File: index.html

package info (click to toggle)
nvidia-cuda-toolkit 12.4.1-2
  • links: PTS, VCS
  • area: non-free
  • in suites: trixie
  • size: 18,505,836 kB
  • sloc: ansic: 203,477; cpp: 64,769; python: 34,699; javascript: 22,006; xml: 13,410; makefile: 3,085; sh: 2,343; perl: 352
file content (2306 lines) | stat: -rw-r--r-- 156,481 bytes parent folder | download | duplicates (6) 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
 
<!DOCTYPE html>
<html class="writer-html5" lang="en" >
<head>
  <meta charset="utf-8" /><meta name="generator" content="Docutils 0.17.1: http://docutils.sourceforge.net/" />
<meta content="The User Guide for Nsight Compute CLI." name="description" />
<meta content="User Guide" name="keywords" />

  <meta name="viewport" content="width=device-width, initial-scale=1.0" />
  <title>4. Nsight Compute CLI &mdash; NsightCompute 12.4 documentation</title>
      <link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
      <link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
      <link rel="stylesheet" href="../_static/design-style.b7bb847fb20b106c3d81b95245e65545.min.css" type="text/css" />
      <link rel="stylesheet" href="../_static/omni-style.css" type="text/css" />
      <link rel="stylesheet" href="../_static/api-styles.css" type="text/css" />
    <link rel="shortcut icon" href="../_static/nsight-compute.ico"/>
  <!--[if lt IE 9]>
    <script src="../_static/js/html5shiv.min.js"></script>
  <![endif]-->
  
        <script data-url_root="../" id="documentation_options" src="../_static/documentation_options.js"></script>
        <script src="../_static/jquery.js"></script>
        <script src="../_static/underscore.js"></script>
        <script src="../_static/doctools.js"></script>
        <script src="../_static/mermaid-init.js"></script>
        <script src="../_static/design-tabs.js"></script>
        <script src="../_static/version.js"></script>
        <script src="../_static/social-media.js"></script>
    <script src="../_static/js/theme.js"></script>
    <link rel="index" title="Index" href="../genindex.html" />
    <link rel="search" title="Search" href="../search.html" />
    <link rel="next" title="1. Customization Guide" href="../CustomizationGuide/index.html" />
    <link rel="prev" title="3. Nsight Compute" href="../NsightCompute/index.html" />
 


</head>

<body class="wy-body-for-nav"> 
  <div class="wy-grid-for-nav">
    <nav data-toggle="wy-nav-shift" class="wy-nav-side">
      <div class="wy-side-scroll">
        <div class="wy-side-nav-search" >


  <a href="../index.html">
  <img src="../_static/nsight-compute.png" class="logo" alt="Logo"/>
</a>

<div role="search">
  <form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
    <input type="text" name="q" placeholder="Search docs" />
    <input type="hidden" name="check_keywords" value="yes" />
    <input type="hidden" name="area" value="default" />
  </form>
</div>
        </div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="Navigation menu">
              <p class="caption" role="heading"><span class="caption-text">Nsight Compute</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../ReleaseNotes/index.html">1. Release Notes</a></li>
<li class="toctree-l1"><a class="reference internal" href="../ProfilingGuide/index.html">2. Kernel Profiling Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../NsightCompute/index.html">3. Nsight Compute</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">4. Nsight Compute CLI</a><ul>
<li class="toctree-l2"><a class="reference internal" href="#introduction">4.1. Introduction</a></li>
<li class="toctree-l2"><a class="reference internal" href="#quickstart">4.2. Quickstart</a></li>
<li class="toctree-l2"><a class="reference internal" href="#usage">4.3. Usage</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#modes">4.3.1. Modes</a></li>
<li class="toctree-l3"><a class="reference internal" href="#multi-process-support">4.3.2. Multi-Process Support</a></li>
<li class="toctree-l3"><a class="reference internal" href="#output-pages">4.3.3. Output Pages</a></li>
<li class="toctree-l3"><a class="reference internal" href="#profile-import">4.3.4. Profile Import</a></li>
<li class="toctree-l3"><a class="reference internal" href="#metrics-and-units">4.3.5. Metrics and Units</a></li>
<li class="toctree-l3"><a class="reference internal" href="#nvtx-filtering">4.3.6. NVTX Filtering</a></li>
<li class="toctree-l3"><a class="reference internal" href="#config-file">4.3.7. Config File</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#command-line-options">4.4. Command Line Options</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#general">4.4.1. General</a></li>
<li class="toctree-l3"><a class="reference internal" href="#launch">4.4.2. Launch</a></li>
<li class="toctree-l3"><a class="reference internal" href="#attach">4.4.3. Attach</a></li>
<li class="toctree-l3"><a class="reference internal" href="#profile">4.4.4. Profile</a></li>
<li class="toctree-l3"><a class="reference internal" href="#pm-sampling">4.4.5. PM Sampling</a></li>
<li class="toctree-l3"><a class="reference internal" href="#warp-sampling">4.4.6. Warp Sampling</a></li>
<li class="toctree-l3"><a class="reference internal" href="#file">4.4.7. File</a></li>
<li class="toctree-l3"><a class="reference internal" href="#console-output">4.4.8. Console Output</a></li>
<li class="toctree-l3"><a class="reference internal" href="#response-file">4.4.9. Response File</a></li>
<li class="toctree-l3"><a class="reference internal" href="#file-macros">4.4.10. File Macros</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#environment-variables">4.5. Environment Variables</a></li>
<li class="toctree-l2"><a class="reference internal" href="#nvprof-transition-guide">4.6. Nvprof Transition Guide</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#trace">4.6.1. Trace</a></li>
<li class="toctree-l3"><a class="reference internal" href="#metric-collection">4.6.2. Metric Collection</a></li>
<li class="toctree-l3"><a class="reference internal" href="#metric-comparison">4.6.3. Metric Comparison</a></li>
<li class="toctree-l3"><a class="reference internal" href="#event-comparison">4.6.4. Event Comparison</a></li>
<li class="toctree-l3"><a class="reference internal" href="#filtering">4.6.5. Filtering</a></li>
<li class="toctree-l3"><a class="reference internal" href="#output">4.6.6. Output</a></li>
<li class="toctree-l3"><a class="reference internal" href="#launch-and-attach">4.6.7. Launch and Attach</a></li>
</ul>
</li>
</ul>
</li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Developer Interfaces</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../CustomizationGuide/index.html">1. Customization Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../NvRulesAPI/index.html">2. NvRules API</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Training</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../Training/index.html">Training</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Release Information</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../Archives/index.html">Archives</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Copyright and Licenses</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../CopyrightAndLicenses/index.html">Copyright and Licenses</a></li>
</ul>

        </div>
      </div>
    </nav>

    <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="Mobile navigation menu" >
          <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
          <a href="../index.html">NsightCompute</a>
      </nav>

      <div class="wy-nav-content">
        <div class="rst-content">
          <div role="navigation" aria-label="Page navigation">
  <ul class="wy-breadcrumbs">


<li><a href="../index.html" class="icon icon-home"></a> &raquo;</li>
<li><span class="section-number">4. </span>Nsight Compute CLI</li>

      <li class="wy-breadcrumbs-aside">
      </li>
<li class="wy-breadcrumbs-aside">


  <span>v2024.1.1 |</span>



  <a href="https://developer.nvidia.com/nsight-compute-history" class="reference external">Archive</a>


  <span>&nbsp;</span>
</li>

  </ul>
  <hr/>
</div>
          <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
           <div itemprop="articleBody">
             
  <section id="nsight-compute-cli">
<h1><span class="section-number">4. </span>Nsight Compute CLI<a class="headerlink" href="#nsight-compute-cli" title="Permalink to this headline"></a></h1>
<p>The User Guide for Nsight Compute CLI.</p>
<section id="introduction">
<h2><span class="section-number">4.1. </span>Introduction<a class="headerlink" href="#introduction" title="Permalink to this headline"></a></h2>
<p>NVIDIA Nsight Compute CLI (ncu) provides a non-interactive way to profile applications from the command line. It can print the results directly on the command line or store them in a report file. It can also be used to simply launch the target application (see <a class="reference external" href="index.html#command-line-options-general">General</a> for details) and later attach with NVIDIA Nsight Compute or another ncu instance.</p>
<p>For users migrating from nvprof to NVIDIA Nsight Compute, please additionally see the <a class="reference external" href="index.html#nvprof-guide">Nvprof Transition Guide</a> for comparison of features and workflows.</p>
</section>
<section id="quickstart">
<h2><span class="section-number">4.2. </span>Quickstart<a class="headerlink" href="#quickstart" title="Permalink to this headline"></a></h2>
<ol class="arabic">
<li><p><strong>Launch the target application with the command line profiler</strong></p>
<p>The command line profiler launches the target application, instruments the target API, and collects profile results for the specified kernels. The CLI executable is called ncu. A shortcut with this name is located in the base directory of the NVIDIA Nsight Compute installation. The actual executable is located in the folder <code class="docutils literal notranslate"><span class="pre">target\windows-desktop-win7-x64</span></code> on Windows or <code class="docutils literal notranslate"><span class="pre">target/linux-desktop-glibc_2_11_3-x64</span></code> on Linux. By default, NVIDIA Nsight Compute is installed in <code class="docutils literal notranslate"><span class="pre">/usr/local/cuda-&lt;cuda-version&gt;/NsightCompute-&lt;version&gt;</span></code> on Linux and in <code class="docutils literal notranslate"><span class="pre">C:\Program</span> <span class="pre">Files\NVIDIA</span> <span class="pre">Corporation\Nsight</span> <span class="pre">Compute</span> <span class="pre">&lt;version&gt;</span></code> on Windows.</p>
<p>To collect the <code class="docutils literal notranslate"><span class="pre">basic</span></code> set for all kernel launches in the target application, launch:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>$ ncu -o profile CuVectorAddMulti.exe
</pre></div>
</div>
<p>The application runs in instrumented mode and for each kernel launch, a profile result is created. The results are written by default to profile.nsight-cuprof. Each output from the compute profiler starts with <code class="docutils literal notranslate"><span class="pre">==PROF==</span></code> The other lines are output from the application itself. For each profiled kernel, the name of the kernel function and the progress of data collection is shown. To collect all requested profile information, it may be required to replay the kernels multiple times. The total number of replay passes per kernel is shown after profiling has completed.</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>[Vector addition of 1144477 elements]
==PROF== Connected to process 5268
Copy input data from the host memory to the CUDA device
CUDA kernel launch A with 4471 blocks of 256 threads
==PROF== Profiling &quot;vectorAdd_A&quot; - 0: 0%....50%....100% - 46 passes
CUDA kernel launch B with 4471 blocks of 256 threads
==PROF== Profiling &quot;vectorAdd_B&quot; - 1: 0%....50%....100% - 46 passes
Copy output data from the CUDA device to the host memory
Done
==PROF== Disconnected from process 5268
==PROF== Report: profile.ncu-rep
</pre></div>
</div>
</li>
<li><p><strong>Customizing data collection</strong></p>
<p>Options are available to specify for which kernels data should be collected. <code class="docutils literal notranslate"><span class="pre">-c</span></code> limits the number of kernel launches collected. <code class="docutils literal notranslate"><span class="pre">-s</span></code> skips the given number of kernels before data collection starts. <code class="docutils literal notranslate"><span class="pre">-k</span></code> allows you to filter the kernels by a regex match of their names. <code class="docutils literal notranslate"><span class="pre">--kernel-id</span></code> allows you to filter kernels by context, stream, name and invocation, similar to nvprof.</p>
<p>To limit what should be collected for each kernel launch, specify the exact *.section (files) by their identifier using <code class="docutils literal notranslate"><span class="pre">--section</span></code>. Each section file defines a set of metrics to be collected, grouped logically to solve a specific performance question. By default, the sections associated with the <code class="docutils literal notranslate"><span class="pre">basic</span></code> set are collected. Use <code class="docutils literal notranslate"><span class="pre">--list-sets</span></code> to see the list of currently available sets. Use <code class="docutils literal notranslate"><span class="pre">--list-sections</span></code> to see the list of currently available sections. The default search directory and location of pre-defined section files is also called <code class="docutils literal notranslate"><span class="pre">sections/</span></code>. See the <a class="reference external" href="../ProfilingGuide/index.html#sets-and-sections">Profiling Guide</a> for more details.</p>
<p>Alternatively, you can collect a set of individual metrics using <code class="docutils literal notranslate"><span class="pre">--metrics</span></code>. The available metrics can be queried using <code class="docutils literal notranslate"><span class="pre">--query-metrics</span></code>. For an explanation of the naming conventions and structuring of metrics, see <a class="reference external" href="../ProfilingGuide/index.html#metrics-structure">Metrics Structure</a>.</p>
<p>Most metrics in NVIDIA Nsight Compute are named using a base name and various suffixes, e.g. <em>sm__throughput.avg.pct_of_peak_sustained_elapsed</em>. The base name is <em>sm__throughput and the suffix is avg.pct_of_peak_sustained_elapsed</em>. This is because most metrics follow the same structure and have the same set of suffixes. You need to pass the base or full name to NVIDIA Nsight Compute when selecting a metric for profiling. Use <code class="docutils literal notranslate"><span class="pre">--query-metrics-mode</span> <span class="pre">suffix</span> <span class="pre">--metrics</span> <span class="pre">&lt;metrics</span> <span class="pre">list&gt;</span></code> to see the full names for the chosen metrics.</p>
<p>Some additional metrics do not follow this structured naming. They are documented in the <a class="reference external" href="../ProfilingGuide/index.html#metrics-reference">Metrics Reference</a>.</p>
</li>
<li><p><strong>Changing command line output</strong></p>
<p>By default, a temporary file is used to store profiling results, and data is printed to the command line. To permanently store the profiler report, use <code class="docutils literal notranslate"><span class="pre">-o</span></code> to specify the output filename.</p>
<p>Besides storing results in a report file, the command line profiler can print results using different pages. Those pages correspond to the respective pages in the UI’s report. By default, the <a class="reference external" href="index.html#output-pages">Details page</a> is printed, if no explicit output file is specified. To select a different page or print in addition to storing in an explicit file, use the <code class="docutils literal notranslate"><span class="pre">--page=&lt;Page&gt;</span></code> command. Currently, the following pages are supported: <code class="docutils literal notranslate"><span class="pre">details,</span> <span class="pre">raw,</span> <span class="pre">source</span></code>.</p>
<p>Use <code class="docutils literal notranslate"><span class="pre">--csv</span></code> to make any output comma separated and easier to process further. See <a class="reference external" href="index.html#command-line-options-console-output">Console Output</a> for further options, e.g. summary views.</p>
</li>
<li><p><strong>Open the report in the UI</strong></p>
<p>The UI executable is called ncu-ui. A shortcut with this name is located in the base directory of the NVIDIA Nsight Compute installation. The actual executable is located in the folder <code class="docutils literal notranslate"><span class="pre">host\windows-desktop-win7-x64</span></code> on Windows or <code class="docutils literal notranslate"><span class="pre">host/linux-desktop-glibc_2_11_3-x64</span></code> on Linux. In the UI window, close the <em>Connection</em> dialog and open the report file through <em>File</em> &gt; <em>Open</em>, by dragging the report file into NVIDIA Nsight Compute.</p>
<p>You can also specify the report file as a command line parameter to the executable, i.e. as <code class="docutils literal notranslate"><span class="pre">ncu-ui</span> <span class="pre">&lt;MyReport.ncu-rep&gt;</span></code>. Alternatively, when using NVIDIA Nsight Compute CLI on a platform with host support, <code class="docutils literal notranslate"><span class="pre">--open-in-ui</span></code> can be used directly with ncu to open a collected report in the user interface.</p>
<p>The report opens in a new document window. For more information about the report, see the <a class="reference external" href="../NsightCompute/index.html#profiler-report">Profiler Report</a> for collecting profile information through NVIDIA Nsight Compute.</p>
</li>
</ol>
</section>
<section id="usage">
<span id="id1"></span><h2><span class="section-number">4.3. </span>Usage<a class="headerlink" href="#usage" title="Permalink to this headline"></a></h2>
<section id="modes">
<h3><span class="section-number">4.3.1. </span>Modes<a class="headerlink" href="#modes" title="Permalink to this headline"></a></h3>
<p>Modes change the fundamental behavior of the command line profiler. Depending on which mode is chosen, different <a class="reference external" href="index.html#command-line-options">Command Line Options</a> become available. For example, <a class="reference external" href="index.html#command-line-options-launch">Launch</a> is invalid if the <em>Attach</em> mode is selected.</p>
<ul>
<li><p><strong>Launch-and-attach:</strong> The target application is launched on the local system with the tool’s injection libraries. Depending on which profiling options are chosen, selected kernels in the application are profiled and the results printed to the console or stored in a report file. The tool exits once the target application finishes or crashes, and once all results are processed.</p>
<p>This is the default, and the only mode that supports profiling of child processes on selected platforms.</p>
</li>
<li><p><strong>Launch:</strong> The target application is launched on the local system with the tool’s injection libraries. As soon as the first intercepted API call is reached (commonly <code class="docutils literal notranslate"><span class="pre">cuInit()</span></code>), all application threads are suspended. The application now expects a tool to attach for profiling. You can attach using NVIDIA Nsight Compute or using the command line profiler’s <em>Attach</em> mode.</p></li>
<li><p><strong>Attach:</strong> The tool tries to connect to a target application previously launched using NVIDIA Nsight Compute or using the command line profiler’s <em>Launch</em> mode. The tool can attach to a target on the local system or using a remote connection.</p></li>
</ul>
</section>
<section id="multi-process-support">
<span id="multi-process"></span><h3><span class="section-number">4.3.2. </span>Multi-Process Support<a class="headerlink" href="#multi-process-support" title="Permalink to this headline"></a></h3>
<p>NVIDIA Nsight Compute CLI supports profiling multi-process applications on the following platforms: x86_64 Windows, x86_64 Linux, DRIVE OS Linux, DRIVE OS QNX, PowerPC. See the <a class="reference external" href="index.html#command-line-options-launch">Launch</a> options on how to enable this feature.</p>
<p>On x86_64 Windows, NVIDIA Nsight Compute CLI supports profiling 64-bit processes launched from 32-bit applications by default . On x86_64 Linux, launching from 32-bit applications requires you to enable the <code class="docutils literal notranslate"><span class="pre">support-32bit</span></code> option, and the required 32-bit libraries must be installed on your system. On DRIVE OS Linux, DRIVE OS QNX and PowerPC, tracking of 32-bit applications is not supported. Profiling of 32-bit processes is not supported on any platform.</p>
<div class="line-block">
<div class="line"><strong>Profiling MPI applications is a special case of multi-process profiling.</strong></div>
</div>
<p>NVIDIA Nsight Compute CLI can be used to profile applications launched with the <code class="docutils literal notranslate"><span class="pre">mpirun</span></code> command.</p>
<ul>
<li><p>To profile all ranks on a node and store all the profiling data in a single report file:</p>
<div class="highlight-c++ notranslate"><div class="highlight"><pre><span></span><span class="n">ncu</span><span class="w"> </span><span class="o">--</span><span class="n">target</span><span class="o">-</span><span class="n">processes</span><span class="w"> </span><span class="n">all</span><span class="w"> </span><span class="o">-</span><span class="n">o</span><span class="w"> </span><span class="o">&lt;</span><span class="n">report</span><span class="o">-</span><span class="n">name</span><span class="o">&gt;</span><span class="w"> </span><span class="n">mpirun</span><span class="w"> </span><span class="p">[</span><span class="n">mpi</span><span class="w"> </span><span class="n">arguments</span><span class="p">]</span><span class="w"> </span><span class="o">&lt;</span><span class="n">app</span><span class="o">&gt;</span><span class="w"> </span><span class="p">[</span><span class="n">app</span><span class="w"> </span><span class="n">arguments</span><span class="p">]</span><span class="w"></span>
</pre></div>
</div>
</li>
<li><p>To profile multi-node submissions, one instance of NVIDIA Nsight Compute CLI can be used per node. Ensure that you specify unique report files per rank.</p>
<div class="highlight-c++ notranslate"><div class="highlight"><pre><span></span><span class="n">mpirun</span><span class="w"> </span><span class="p">[</span><span class="n">mpi</span><span class="w"> </span><span class="n">arguments</span><span class="p">]</span><span class="w"> </span><span class="n">ncu</span><span class="w"> </span><span class="o">-</span><span class="n">o</span><span class="w"> </span><span class="n">report_</span><span class="o">%</span><span class="n">q</span><span class="p">{</span><span class="n">OMPI_COMM_WORLD_RANK</span><span class="p">}</span><span class="w"> </span><span class="o">&lt;</span><span class="n">app</span><span class="o">&gt;</span><span class="w"> </span><span class="p">[</span><span class="n">app</span><span class="w"> </span><span class="n">arguments</span><span class="p">]</span><span class="w"></span>
</pre></div>
</div>
</li>
<li><p>To profile a single rank one can use a wrapper script. The following script (called “wrap.sh”) profiles rank 0 only:</p>
<div class="highlight-c++ notranslate"><div class="highlight"><pre><span></span><span class="cp">#!/bin/bash</span>
<span class="k">if</span><span class="w"> </span><span class="p">[[</span><span class="w"> </span><span class="n">$OMPI_COMM_WORLD_RANK</span><span class="w"> </span><span class="o">==</span><span class="w"> </span><span class="mi">0</span><span class="w"> </span><span class="p">]];</span><span class="w"> </span><span class="n">then</span><span class="w"></span>
<span class="w">   </span><span class="n">ncu</span><span class="w"> </span><span class="o">-</span><span class="n">o</span><span class="w"> </span><span class="n">report_$</span><span class="p">{</span><span class="n">OMPI_COMM_WORLD_RANK</span><span class="p">}</span><span class="w">  </span><span class="o">--</span><span class="n">target</span><span class="o">-</span><span class="n">processes</span><span class="w"> </span><span class="n">all</span><span class="w"> </span><span class="s">&quot;$@&quot;</span><span class="w"></span>
<span class="k">else</span><span class="w"></span>
<span class="w">   </span><span class="s">&quot;$@&quot;</span><span class="w"></span>
<span class="n">fi</span><span class="w"></span>
</pre></div>
</div>
<p>and then execute:</p>
<div class="highlight-c++ notranslate"><div class="highlight"><pre><span></span><span class="n">mpirun</span><span class="w"> </span><span class="p">[</span><span class="n">mpi</span><span class="w"> </span><span class="n">arguments</span><span class="p">]</span><span class="w"> </span><span class="p">.</span><span class="o">/</span><span class="n">wrap</span><span class="p">.</span><span class="n">sh</span><span class="w"> </span><span class="o">&lt;</span><span class="n">app</span><span class="o">&gt;</span><span class="w"> </span><span class="p">[</span><span class="n">app</span><span class="w"> </span><span class="n">arguments</span><span class="p">]</span><span class="w"></span>
</pre></div>
</div>
</li>
</ul>
</section>
<section id="output-pages">
<span id="id2"></span><h3><span class="section-number">4.3.3. </span>Output Pages<a class="headerlink" href="#output-pages" title="Permalink to this headline"></a></h3>
<p>The command line profiler supports printing results to the console using various pages. Each page has an equivalent in NVIDIA Nsight Compute’s <em>Profiler Report</em>. In the command line profiler, they are slightly adapted to fit console output. To select a page, use the <code class="docutils literal notranslate"><span class="pre">--page</span></code> option. By default, the details page is used. Note that if <code class="docutils literal notranslate"><span class="pre">--page</span></code> is not used but <code class="docutils literal notranslate"><span class="pre">--export</span></code> is, no results will be printed to the console.</p>
<ul>
<li><p><strong>Details:</strong> This page represents NVIDIA Nsight Compute’s <em>Details</em> page. For every profiled kernel launch, each collected is printed as section as a three-column table, followed by any rule results applied to this section. Rule results not associated with any section are printed after the kernel’s sections.</p>
<p>The first section table column shows the metric name. If the metric was given a label in the section, it is used instead. The second column shows the metric unit, if available. The third column shows the unit value. Both metric unit and value are automatically adjusted to the most fitting order of magnitude. By default, only metrics defined in section headers are shown. This can be changed by passing the <code class="docutils literal notranslate"><span class="pre">--details-all</span></code> option on the command line.</p>
<p>Some metrics will show multiple values, separated by “;”, e.g. memory_l2_transactions_global Kbytes 240; 240; 240; 240; 240. Those are instanced metrics, which have one value per represented instance. An instance can be a streaming multiprocessor, an assembly source line, etc.</p>
</li>
<li><p><strong>Raw:</strong> This page represents NVIDIA Nsight Compute’s <em>Raw</em> page. For every profiled kernel launch, each collected metric is printed as a three-column table. Besides metrics from sections, this includes automatically collected metrics such as device attributes and kernel launch information.</p>
<p>The first column shows the metric name. The second and third columns show the metric unit and value, respectively. Both metric unit and value are automatically adjusted to the most fitting order of magnitude. No unresolved regex:, group:, or breakdown: metrics are included.</p>
</li>
</ul>
</section>
<section id="profile-import">
<span id="id3"></span><h3><span class="section-number">4.3.4. </span>Profile Import<a class="headerlink" href="#profile-import" title="Permalink to this headline"></a></h3>
<p>Using the <code class="docutils literal notranslate"><span class="pre">--import</span></code> option, saved reports can be imported into the command line profiler. When using this flag, most other options are not available, except for certain result filterting options. They are marked as such in the <a class="reference external" href="index.html#command-line-options-profile">Profile options</a> table.</p>
</section>
<section id="metrics-and-units">
<h3><span class="section-number">4.3.5. </span>Metrics and Units<a class="headerlink" href="#metrics-and-units" title="Permalink to this headline"></a></h3>
<p>When available and applicable, metrics are shown along with their unit. This is to make it apparent if a metric represents cycles, threads, bytes/s, and so on.</p>
<p>By default, units are scaled automatically so that metric values are shown with a reasonable order of magnitude. Units are scaled using their SI-factors, i.e. byte-based units are scaled using a factor of 1000 and the prefixes K, M, G, etc. Time-based units are also scaled using a factor of 1000, with the prefixes n, u and m. This scaling can be changed using a command line option, see <a class="reference external" href="index.html#command-line-options-console-output">Console Output</a> options for details.</p>
</section>
<section id="nvtx-filtering">
<span id="id4"></span><h3><span class="section-number">4.3.6. </span>NVTX Filtering<a class="headerlink" href="#nvtx-filtering" title="Permalink to this headline"></a></h3>
<div class="line-block">
<div class="line"><code class="docutils literal notranslate"><span class="pre">--nvtx-include</span> <span class="pre">&lt;configuration&gt;</span> <span class="pre">--nvtx-exclude</span> <span class="pre">&lt;configuration&gt;</span></code></div>
<div class="line">These options are used to profile only those kernels which satisfy the conditions mentioned in the configuration. Through these options, you can choose which kernel falls into a specific range or collection of ranges.</div>
</div>
<p>You can use both options multiple times, mentioning all the <code class="docutils literal notranslate"><span class="pre">--nvtx-include</span></code> configurations followed by all <code class="docutils literal notranslate"><span class="pre">--nvtx-exclude</span></code> configurations. NVTX filtering requires <code class="docutils literal notranslate"><span class="pre">--nvtx</span></code> option.</p>
<p>NVTX ranges are of two types: NvtxRangeStart/End and NvtxRangePush/Pop. The configuration syntax for both the types are briefly described below.
Both range and domain names can contain whitespace. Note that “Domain” and “range” in below example are for illustration purposes only and are not required to mark domain or range names.</p>
<ul>
<li><p><strong>Push-Pop Ranges</strong></p>
<table class="table-no-stripes docutils align-default">
<colgroup>
<col style="width: 19%" />
<col style="width: 54%" />
<col style="width: 28%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Quantifier</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Example</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>/</p></td>
<td><p>Delimiter between range names. When only a single range name is
given, the delimiter must be appended to indicate that this
refers to a push/pop range.</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">A_range/</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">A_range/B</span> <span class="pre">range</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">A_range/\*/B</span> <span class="pre">range</span></code></p>
</td>
</tr>
<tr class="row-odd"><td><p>[</p></td>
<td><p>Range is at the bottom of the stack</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">[A_range</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">[A_range/+/Range</span> <span class="pre">Z</span></code></p>
</td>
</tr>
<tr class="row-even"><td><p>]</p></td>
<td><p>Range is at the top of the stack</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">Range</span> <span class="pre">Z]</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">Range</span> <span class="pre">C/\*/Range</span> <span class="pre">Z]</span></code></p>
</td>
</tr>
<tr class="row-odd"><td><ul class="simple">
<li></li>
</ul>
</td>
<td><p>Only one B rangeetween the two other ranges</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">B</span> <span class="pre">range/+/Range</span> <span class="pre">D</span></code></p></td>
</tr>
<tr class="row-even"><td><p>*</p></td>
<td><p>Zero or more range(s) between the two other ranges</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">B</span> <span class="pre">range/\*/Range</span> <span class="pre">Z</span></code></p></td>
</tr>
<tr class="row-odd"><td><p>&#64;</p></td>
<td><p>Specify domain name. If not mentioned, assuming &lt;default domain&gt;</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">Domain-A&#64;A_range</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">Domain</span> <span class="pre">B&#64;A_range/\*/Range</span> <span class="pre">Z]</span></code></p>
</td>
</tr>
</tbody>
</table>
<p>Include kernels wrapped inside push/pop range ‘A_range’ of ‘&lt;default-domain&gt;’:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;A_range/&quot; CuNvtx.exe
</pre></div>
</div>
<p>Include kernels wrapped inside push/pop range ‘A_range’ of ‘Domain-A’:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;Domain-A@A_range/&quot; CuNvtx.exe
</pre></div>
</div>
<p>Include kernels wrapped inside push/pop range ‘A_range’ of ‘&lt;default domain&gt;’, where ‘A_range’ is at the bottom of the stack:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;[A_range&quot; CuNvtx.exe
</pre></div>
</div>
<p>Include kernels wrapped inside push/pop ranges ‘A_range’ and ‘B range’ of ‘&lt;default domain&gt;’, with zero or many ranges between them:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;A_range/*/B range&quot; CuNvtx.exe
</pre></div>
</div>
<p>Exclude kernels wrapped inside push/pop ranges ‘A_range’ and ‘B range’ of ‘&lt;default domain&gt;’, with zero or many ranges between them:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-exclude &quot;A_range/*/B range&quot; CuNvtx.exe
</pre></div>
</div>
<p>Include kernels wrapped inside only push/pop range ‘A_range’ of ‘&lt;default domain&gt;’ but not inside ‘B range’ at the top of the stack:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;A_range/&quot; --nvtx-exclude &quot;B range]&quot; CuNvtx.exe
</pre></div>
</div>
</li>
<li><p><strong>Start-End Ranges</strong></p>
<table class="table-no-stripes docutils align-default">
<colgroup>
<col style="width: 19%" />
<col style="width: 55%" />
<col style="width: 25%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Quantifier</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Example</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>,</p></td>
<td><p>Delimiter between range names</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">A_range,B</span> <span class="pre">range</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">B</span> <span class="pre">range,A_range,Range</span> <span class="pre">C</span></code></p>
</td>
</tr>
<tr class="row-odd"><td><p>&#64;</p></td>
<td><p>Specify domain name. If not mentioned, assuming &lt;default domain&gt;</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">Domain-A&#64;A_range</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">Domain</span> <span class="pre">B&#64;B</span> <span class="pre">range,Range</span> <span class="pre">Z</span></code></p>
</td>
</tr>
</tbody>
</table>
<p>Include kernels wrapped inside start/end range ‘A_range’ of ‘Domain-A’:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;Domain-A@A_range&quot; CuNvtx.exe
</pre></div>
</div>
<p>Include kernels wrapped inside both start/end ranges, ‘A_range’ and ‘B range’ of ‘&lt;default domain&gt;’:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;A_range,B range&quot; CuNvtx.exe
</pre></div>
</div>
<p>Include kernels wrapped inside start/end ranges, ‘A_range’ or ‘B range’ of ‘&lt;default domain&gt;’:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;A_range&quot; --nvtx-include &quot;B range&quot; CuNvtx.exe
</pre></div>
</div>
<p>Include all kernels, except those which are wrapped inside start/end range ‘A_range’ of ‘&lt;default domain&gt;’:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-exclude &quot;A_range&quot; CuNvtx.exe
</pre></div>
</div>
<p>Include kernels wrapped inside only start/end ‘B range’ and not ‘A_range’ of ‘&lt;default domain&gt;’:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;B range&quot;--nvtx-exclude &quot;A_range&quot; CuNvtx.exe
</pre></div>
</div>
</li>
<li><p><strong>Regular Expression Support</strong></p>
<p>The configuration syntax for both the types NvtxRangeStart/End and NvtxRangePush/Pop is the same. Additionally, to use regular expressions, follow the following syntax.</p>
<ul>
<li><p>Provide prefix ‘regex:’ to treat nvtx config as regular expression.</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;regex:Domain[A-Z]@Range[0-9]/&quot; CuNvtx.exe
</pre></div>
</div>
<div class="line-block">
<div class="line">The kernels wrapped inside push/pop range with matching regex ‘Range[0-9]’ of domain with matching regex ‘Domain[A-Z]’ are profiled.</div>
</div>
</li>
<li><p>Provide ‘/’ as a prefix to “[” or “]” only for the range part of the config if “[” or “]” is at the start or at the end of the range part, respectively. This is needed so that NCU can distinguish if “[” or “]” is part of the regex or represents the top/bottom of the stack.</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>ncu --nvtx --nvtx-include &quot;regex:[0-9]domainA@/[0-9]rangeA,RangeC[0-9/]&quot; CuNvtx.exe
</pre></div>
</div>
<div class="line-block">
<div class="line">The kernels wrapped inside start/end ranges with matching regex ‘[0-9]rangeA’ and ‘RangeC[0-9]’ of domain with matching regex ‘[0-9]domainA’ are profiled.</div>
</div>
</li>
<li><p>If any quantifier is part of the domain/range name, you need to use ‘\\’ or ‘\’ as a prefix. For the “$” quantifier, only the ‘\\’ prefix is valid.</p></li>
</ul>
</li>
<li><p><strong>Additional Information</strong></p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>--nvtx-include DomainA@RangeA,DomainB@RangeB //Not a valid config
</pre></div>
</div>
<p>In a single NVTX configuration, multiple ranges with regard to a single domain can be specified. Mentioning ranges from different domains inside a single NVTX config is not supported.</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>--nvtx-include &quot;A_range\[i\]&quot;
</pre></div>
</div>
<p>Quantifiers ‘&#64;’ ‘,’ ‘[’ ‘]’ ‘/’ ‘*’ ‘+’ can be used in range names using prefix ‘\’. The kernels wrapped inside ‘A_range[i]’ of ‘&lt;default domain&gt;’ in the application are profiled.</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>--nvtx-include &quot;A_range&quot;  //Start/End configuration
--nvtx-inlcude &quot;A_range/&quot; //Push/Pop configuration
--nvtx-inlcude &quot;A_range]&quot; //Push/Pop configuration
</pre></div>
</div>
<p>If the domain/range name contains ‘\’, you need to provide ‘\\\\’ in the config.</p>
<p>Do not use ‘\\\\’ before any quantifier.</p>
<p>Including/Excluding only single range for Push/Pop configuration without specifying stack frame position ‘[’ or ‘]’, use ‘/’ quantifier at the end.</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>--nvtx-include &quot;A_range/*/B range&quot;
</pre></div>
</div>
<p>The order in which you mention Push/Pop configurations is important. In the above example, ‘A_range’ should be below ‘B range’ in the stack of ranges so that the kernel is profiled.</p>
<p>NVTX filtering honors cudaProfilerStart() and cudaProfilerStop(). There is no support for ranges with no name.</p>
</li>
</ul>
</section>
<section id="config-file">
<h3><span class="section-number">4.3.7. </span>Config File<a class="headerlink" href="#config-file" title="Permalink to this headline"></a></h3>
<div class="line-block">
<div class="line">Using the <code class="docutils literal notranslate"><span class="pre">--config-file</span> <span class="pre">on/off</span></code> option, parsing parameters from config file can be enabled or disabled.</div>
<div class="line">Using the <code class="docutils literal notranslate"><span class="pre">--config-file-path</span> <span class="pre">&lt;path&gt;</span></code> option, default path and name of config file can be overwritten.</div>
<div class="line">By default, config-file with name <code class="docutils literal notranslate"><span class="pre">config.ncu-cfg</span></code> is searched in the current working directory, <code class="docutils literal notranslate"><span class="pre">$HOME/.config/NVIDIA</span> <span class="pre">Corporation</span></code> on Linux and <code class="docutils literal notranslate"><span class="pre">%APPDATA%\NVIDIA</span> <span class="pre">Corporation\</span></code> on Windows. If a valid config file is found, ncu parses the file and initializes any command line parameters to the values set in the file. If the same command line parameter is also set explicitly during the current invocation, the latter takes precedence.</div>
</div>
<p>Parameters can be set under various general modes and ncu command line parameters are used to determine which general-mode needs to be parsed from the config file. See the table below for more details.</p>
<table class="table-no-stripes docutils align-default">
<colgroup>
<col style="width: 81%" />
<col style="width: 19%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Command line parameters</p></th>
<th class="head"><p>General Mode</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>ncu –mode launch-and-attach CuVectorAddMulti.exe</p></td>
<td><p>Launch-and-attach</p></td>
</tr>
<tr class="row-odd"><td><p>ncu –mode launch CuVectorAddMulti.exe</p></td>
<td><p>Launch</p></td>
</tr>
<tr class="row-even"><td><p>ncu –mode attach</p></td>
<td><p>Attach</p></td>
</tr>
<tr class="row-odd"><td><p>ncu –list-sets, ncu –list-sections, ncu –list-rules and ncu –list-metrics</p></td>
<td><p>List</p></td>
</tr>
<tr class="row-even"><td><p>ncu –query-metrics</p></td>
<td><p>Query</p></td>
</tr>
<tr class="row-odd"><td><p>ncu -i &lt;MyReport.ncu-rep&gt;</p></td>
<td><p>Import</p></td>
</tr>
</tbody>
</table>
<p>These general modes should be defined in the config file using INI-like syntax as:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>[&lt;general-mode&gt;]
&lt;parameter&gt;=&lt;value&gt;
;&lt;comments&gt;
</pre></div>
</div>
<p><strong>Sample usage</strong></p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>[Launch-and-attach]
-c = 1
--section = LaunchStats, Occupancy
[Import]
--open-in-ui
-c = 1
--section = LaunchStats, Occupancy
</pre></div>
</div>
<p>From this configuration, ncu will parse parameters set under <code class="docutils literal notranslate"><span class="pre">[Launch-and-attach]</span></code> block whenever an application is profiled in <code class="docutils literal notranslate"><span class="pre">launch-and-attach</span></code> mode. In the same manner, parameters set under <code class="docutils literal notranslate"><span class="pre">[Import]</span></code> block will be parsed whenever a report is imported. Different modes can be clubbed together if there exists a set of parameters which is common to each mode. Sample shown above can be rewritten after clubbing both modes as:</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>[Launch-and-attach, import]
-c = 1
--section = LaunchStats, Occupancy
[Import]
--open-in-ui
</pre></div>
</div>
<p><strong>Additional points</strong></p>
<ul>
<li><p>Options like <code class="docutils literal notranslate"><span class="pre">--open-in-ui</span></code> do not expect any value to be set. These options should not be passed any value.</p></li>
<li><p>Options like <code class="docutils literal notranslate"><span class="pre">--section</span></code> can be passed multiple times in the command line. These options should be written only once under a general-mode with all required values seperated by comma as shown below. Explicitly setting values for these options will not overwrite the config file values. Instead, all values will be composed together and set to the option.</p>
<div class="highlight-text notranslate"><div class="highlight"><pre><span></span>[&lt;general-mode&gt;]
&lt;parameter&gt;=&lt;value1&gt;,&lt;value2&gt;,...
</pre></div>
</div>
</li>
</ul>
</section>
</section>
<section id="command-line-options">
<h2><span class="section-number">4.4. </span>Command Line Options<a class="headerlink" href="#command-line-options" title="Permalink to this headline"></a></h2>
<p>For long command line options, passing a unique initial substring can be sufficient.</p>
<section id="general">
<span id="command-line-options-general"></span><h3><span class="section-number">4.4.1. </span>General<a class="headerlink" href="#general" title="Permalink to this headline"></a></h3>
<table class="table-no-stripes docutils align-default" id="id18">
<caption><span class="caption-text">Table 1. General Command Line Options</span><a class="headerlink" href="#id18" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 10%" />
<col style="width: 81%" />
<col style="width: 10%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>h,help</p></td>
<td><p>Show help message</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>v,version</p></td>
<td><p>Show version information</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>mode</p></td>
<td><p>Select the mode of interaction with the target application</p>
<ul class="simple">
<li><p><strong>launch-and-attach:</strong> Launch the target application and immediately attach for profiling.</p></li>
<li><p><strong>launch:</strong> Launch the target application and suspend in the first intercepted API call, wait for tool to attach.</p></li>
<li><p><strong>attach:</strong> Attach to a previously launched application to which no other tool is attached.</p></li>
</ul>
</td>
<td><p>launch-and-attach</p></td>
</tr>
<tr class="row-odd"><td><p>p,port</p></td>
<td><p>Base port used for connecting to target applications for <code class="docutils literal notranslate"><span class="pre">--mode</span> <span class="pre">launch/attach</span></code></p></td>
<td><p>49152</p></td>
</tr>
<tr class="row-even"><td><p>max-connections</p></td>
<td><p>Maximum number of ports for connecting to target applications</p></td>
<td><p>64</p></td>
</tr>
<tr class="row-odd"><td><p>config-file</p></td>
<td><p>Use config.ncu-cfg config file to set parameters. Searches in the current working directory, in “$HOME/.config/NVIDIA Corporation” on Linux and in “%APPDATA%\NVIDIA Corporation\” on Windows.</p></td>
<td><p>on</p></td>
</tr>
<tr class="row-even"><td><p>config-file-path</p></td>
<td><p>Override the default path for config file.</p></td>
<td></td>
</tr>
</tbody>
</table>
</section>
<section id="launch">
<span id="command-line-options-launch"></span><h3><span class="section-number">4.4.2. </span>Launch<a class="headerlink" href="#launch" title="Permalink to this headline"></a></h3>
<table class="table-no-stripes docutils align-default" id="id19">
<caption><span class="caption-text">Table 2. Launch Command Line Options</span><a class="headerlink" href="#id19" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 6%" />
<col style="width: 53%" />
<col style="width: 41%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>check-exit-code</p></td>
<td><p>Check the application exit code and print an error if it is different than 0. If set, <code class="docutils literal notranslate"><span class="pre">--replay-mode</span> <span class="pre">application</span></code> will stop after the first pass if the exit code is not 0.</p></td>
<td><p>yes</p></td>
</tr>
<tr class="row-odd"><td><p>injection-path-64</p></td>
<td><p>Override the default path for the injection libraries. The injection libraries are used by the tools to intercept relevant APIs (like CUDA or NVTX).</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>preload-library</p></td>
<td><p>Prepend a shared library to be loaded by the application before the injection libraries. This option can be given multiple times and the libraries will be loaded in the order they were specified.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>call-stack</p></td>
<td><p>Enable CPU Call Stack collection.</p></td>
<td><p>false</p></td>
</tr>
<tr class="row-even"><td><p>nvtx</p></td>
<td><p>Enable NVTX support for tools.</p></td>
<td><p>false</p></td>
</tr>
<tr class="row-odd"><td><p>target-processes</p></td>
<td><p>Select the processes you want to profile. Available modes are:</p>
<ul class="simple">
<li><p><strong>application-only</strong> Profile only the root application process.</p></li>
<li><p><strong>all</strong> Profile the application and all its child processes.</p></li>
</ul>
</td>
<td><p>all</p></td>
</tr>
<tr class="row-even"><td><p>target-processes-filter</p></td>
<td><p>Set the comma separated expressions to filter which processes are profiled.</p>
<ul>
<li><p><code class="docutils literal notranslate"><span class="pre">&lt;process</span> <span class="pre">name&gt;</span></code> Set the exact process name to include for profiling.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">regex:&lt;expression&gt;</span></code> Set the regex to filter matching process name profiling. On shells that recognize regular expression symbols as special characters (e.g. Linux bash),
the expression needs to be escaped with quotes, e.g. <code class="docutils literal notranslate"><span class="pre">--target-processes-filter</span> <span class="pre">regex:&quot;.*Process&quot;</span></code>.</p>
<p>When using <code class="docutils literal notranslate"><span class="pre">regex:</span></code>, the expression must not include any commas.</p>
</li>
<li><p><code class="docutils literal notranslate"><span class="pre">exclude:&lt;process</span> <span class="pre">name&gt;</span></code> Set the exact process name to exclude for profiling.</p></li>
<li><p>exclude-tree:&lt;process name&gt; Set the exact process name to exclude for profiling and further process tracking. None of its child processes will be profiled, even if they match a positive filter.
This option is not available on Windows.</p></li>
</ul>
<p>The executable name part of the process will be considered in the match. Processing of filters stops at the first match.
If any positive filter is specified, no process that is not matching a positive filter is profiled.</p>
</td>
<td><p><strong>Examples</strong></p>
<p><code class="docutils literal notranslate"><span class="pre">--target-processes-filter</span> <span class="pre">MatrixMul</span></code> Filter all processes having executable name exactly as “MatrixMul”.</p>
<p><code class="docutils literal notranslate"><span class="pre">--target-processes-filter</span> <span class="pre">regex:Matrix</span></code>Filter all processes that include the string “Matrix” in their executable name, e.g. “MatrixMul” and “MatrixAdd”.</p>
<p><code class="docutils literal notranslate"><span class="pre">--target-processes-filter</span> <span class="pre">MatrixMul,MatrixAdd</span></code>Filter all processes having executable name exactly as “MatrixMul” or “MatrixAdd”.</p>
<p><code class="docutils literal notranslate"><span class="pre">--target-processes-filter</span> <span class="pre">exclude:MatrixMul.exe</span></code> Exclude only “MatrixMul.exe”.</p>
<p><code class="docutils literal notranslate"><span class="pre">--target-processes-filter</span> <span class="pre">exclude-tree:ChildLauncher,ParentProcess</span></code> Exclude “ChildLauncher” and all its sub-processes.
Include (only) “ParentProcess”, but not if it’s a child of “ChildLauncher”.</p>
</td>
</tr>
<tr class="row-odd"><td><p>support-32bit</p></td>
<td><p>Support profiling processes launched from 32-bit applications. This option is only available on x86_64 Linux. On Windows, tracking 32-bit applications is enabled by default.</p></td>
<td><p>no</p></td>
</tr>
<tr class="row-even"><td><p>null-stdin</p></td>
<td><p>Launch the application with ‘/dev/null’ as its standard input. This avoids applications reading from standard input being stopped by <code class="docutils literal notranslate"><span class="pre">SIGTTIN</span></code> signals and hanging when running as backgrounded processes.</p></td>
<td><p>false</p></td>
</tr>
</tbody>
</table>
</section>
<section id="attach">
<span id="command-line-options-attach"></span><h3><span class="section-number">4.4.3. </span>Attach<a class="headerlink" href="#attach" title="Permalink to this headline"></a></h3>
<table class="table-no-stripes docutils align-default" id="id20">
<caption><span class="caption-text">Table 3. Attach Command Line Options</span><a class="headerlink" href="#id20" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 8%" />
<col style="width: 86%" />
<col style="width: 6%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>hostname</p></td>
<td><p>Set the hostname or IP address for connecting to the machine on which the target application is running. When attaching to a local target application, use 127.0.0.1.</p></td>
<td><p>127.0.0.1</p></td>
</tr>
</tbody>
</table>
</section>
<section id="profile">
<span id="command-line-options-profile"></span><h3><span class="section-number">4.4.4. </span>Profile<a class="headerlink" href="#profile" title="Permalink to this headline"></a></h3>
<table class="table-no-stripes docutils align-default" id="id21">
<caption><span class="caption-text">Table 4. Profile Command Line Options</span><a class="headerlink" href="#id21" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 5%" />
<col style="width: 69%" />
<col style="width: 26%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default/Examples</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>devices</p></td>
<td><p>List the GPU devices to enable profiling on, separated by comma. <a class="footnote-reference brackets" href="#fn1" id="id5">1</a></p></td>
<td><p>All devices</p>
<p><strong>Examples</strong></p>
<p><code class="docutils literal notranslate"><span class="pre">--devices</span> <span class="pre">0,2</span></code></p>
</td>
</tr>
<tr class="row-odd"><td><p>filter-mode</p></td>
<td><p>Set the filtering mode for kernel launches. Available modes:</p>
<ul class="simple">
<li><p><strong>global:</strong> Apply provided launch filters on kernel launches collectively.</p></li>
<li><p><strong>per-gpu:</strong> Apply provided launch filters on kernel launches separately on each device. Effective launch filters for this mode are <code class="docutils literal notranslate"><span class="pre">--launch-count</span></code> and <code class="docutils literal notranslate"><span class="pre">--launch-skip</span></code></p></li>
<li><p><strong>per-launch-config:</strong> Apply kernel filters and launch filters on kernel launches separately for each GPU launch parameter i.e. Grid Size, Block Size and Shared Memory.</p></li>
</ul>
</td>
<td><p>global</p></td>
</tr>
<tr class="row-even"><td><p>kernel-id</p></td>
<td><p>Set the identifier to use for matching kernels. If the kernel does not match the identifier, it will be ignored for profiling.</p>
<p>The identifier must be of the following format: <em>context-id:stream-id:[name-operator:]kernel-name:invocation-nr</em></p>
<ul class="simple">
<li><p><strong>context-id</strong> is the CUDA context ID or regular expression to match the NVTX name.</p></li>
<li><p><strong>stream-id</strong> is the CUDA stream ID or regular expression to match the NVTX name.</p></li>
<li><p><strong>name-operator</strong> is an optional operator to <em>kernel-name</em>. Currently, <em>regex</em> is the only supported operator.</p></li>
<li><p><strong>kernel-name</strong> is the expression to match the kernel name. By default, this is a full, literal match to what is specified by <code class="docutils literal notranslate"><span class="pre">--kernel-name-base</span></code>. When specifying the optional <em>regex</em> name operator, this is a partial regular expression match to what is specified by <code class="docutils literal notranslate"><span class="pre">--kernel-name-base</span></code>.</p></li>
<li><p><strong>invocation-nr</strong> is the N’th invocation of this kernel function. Multiple invocations can also be specified using regular expressions.</p></li>
</ul>
<p>If the context/stream ID is a positive number, it will be strictly matched against the CUDA context/stream ID. Otherwise it will be treated as a regular expression and matched against the context/stream name specified using the NVTX library. <a class="footnote-reference brackets" href="#fn1" id="id6">1</a></p>
</td>
<td><p><strong>Examples</strong></p>
<p><code class="docutils literal notranslate"><span class="pre">--kernel-id</span> <span class="pre">::foo:2</span></code> For kernel “foo”, match the second invocation.</p>
<p><code class="docutils literal notranslate"><span class="pre">--kernel-id</span> <span class="pre">:::&quot;.*5|3&quot;</span></code> For all kernels, match the third invocation, and all for which the invocation number ends in “5”.</p>
<p><code class="docutils literal notranslate"><span class="pre">--kernel-id</span> <span class="pre">::regex:^.*foo$:</span></code> Match all kernels ending in “foo”.</p>
<p><code class="docutils literal notranslate"><span class="pre">--kernel-id</span> <span class="pre">::regex:^(?!foo):</span></code> Match all kernels except those starting with “foo”. Note that depending on your OS and shell, `
you might need to quote the expression, e.g. using single quotes in Linux <em>bash</em>: <code class="docutils literal notranslate"><span class="pre">--kernel-id</span> <span class="pre">::regex:'^(?!foo)':</span></code></p>
<p><code class="docutils literal notranslate"><span class="pre">--kernel-id</span> <span class="pre">1:2::7</span></code> Match all seventh kernel invocations on context 1, stream 2.</p>
</td>
</tr>
<tr class="row-odd"><td><p>k,kernel-name</p></td>
<td><p>Set the expression to use when matching kernel names.</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">&lt;kernel</span> <span class="pre">name&gt;</span></code> Set the kernel name for an exact match.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">regex:&lt;expression&gt;</span></code> Set the regex to use for matching the kernel name. On shells that recognize regular expression symbols as special characters (e.g. Linux bash), the expression needs to be escaped with quotes, e.g. <code class="docutils literal notranslate"><span class="pre">--kernel-name</span> <span class="pre">regex:&quot;.*Foo&quot;</span></code>.</p></li>
</ul>
<p>If the kernel name or the provided expression do not match, it will be ignored for profiling. <a class="footnote-reference brackets" href="#fn1" id="id7">1</a></p>
</td>
<td><p><strong>Examples</strong></p>
<p><code class="docutils literal notranslate"><span class="pre">-k</span> <span class="pre">foo</span></code> Match all kernels named exactly “foo”.</p>
<p><code class="docutils literal notranslate"><span class="pre">-k</span> <span class="pre">regex:foo</span></code> Match all kernels that include the string “foo”, e.g. “foo” and “fooBar”.</p>
<p><code class="docutils literal notranslate"><span class="pre">-k</span> <span class="pre">regex:&quot;foo|bar&quot;</span></code> Match all kernels including the strings “foo” or “bar”, e.g. “foo”, “foobar”, “_bar2”.</p>
</td>
</tr>
<tr class="row-even"><td><p>kernel-name-base</p></td>
<td><p>Set the basis for <code class="docutils literal notranslate"><span class="pre">--kernel-name</span></code>, and <code class="docutils literal notranslate"><span class="pre">--kernel-id</span></code> kernel-name.  <a class="footnote-reference brackets" href="#fn1" id="id8">1</a> Options are:</p>
<ul class="simple">
<li><p><strong>function:</strong> Function name without parameters, templates etc. e.g. <code class="docutils literal notranslate"><span class="pre">dmatrixmul</span></code></p></li>
<li><p><strong>demangled:</strong> Demangled function name, including parameters, templates, etc. e.g. <code class="docutils literal notranslate"><span class="pre">dmatrixmul(float*,int,int)</span></code></p></li>
<li><p><strong>mangled:</strong> Mangled function name. e.g. <code class="docutils literal notranslate"><span class="pre">_Z10dmatrixmulPfiiS_iiS_</span></code></p></li>
</ul>
</td>
<td><p>function</p></td>
</tr>
<tr class="row-odd"><td><p>c,launch-count</p></td>
<td><p>Limit the number of profiled kernel launches. The count is only incremented for launches that match the kernel filters.<a class="footnote-reference brackets" href="#fn1" id="id9">1</a></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>s,launch-skip</p></td>
<td><p>Set the number of kernel launches to skip before starting to profile kernels. The number takes into account only launches that match the kernel filters. <a class="footnote-reference brackets" href="#fn1" id="id10">1</a></p></td>
<td><p>0</p></td>
</tr>
<tr class="row-odd"><td><p>launch-skip-before-match</p></td>
<td><p>Set the number of kernel launches to skip before starting to profile. The count is incremented for all launches, regardless of the kernel filters. <a class="footnote-reference brackets" href="#fn1" id="id11">1</a></p></td>
<td><p>0</p></td>
</tr>
<tr class="row-even"><td><p>range-filter</p></td>
<td><p>Filter to profile specified instance(s) of matching NVTX ranges or start/stop ranges created through cu(da)ProfilerStart/Stop APIs.</p>
<p>Specify in format <em>[yes/no/on/off]:[start/stop range instance(s)]:[NVTX range instance(s)]</em></p>
<ul class="simple">
<li><p>[yes/no/on/off] : default is ‘no/off’. If set to ‘yes/on’ then NVTX range numbering starts from 1 inside every start/stop range.</p></li>
<li><p>provide numbers in regex form e.g, [2-4] or 2|3|4 to profile 2nd, 3rd and 4th instance of the matching range.</p></li>
<li><p>NVTX range numbers will be counted for matching range provided using –nvtx-include.</p></li>
</ul>
</td>
<td><p><strong>Examples</strong></p>
<p><code class="docutils literal notranslate"><span class="pre">--range-filter</span> <span class="pre">:2:3</span> <span class="pre">--nvtx-include</span> <span class="pre">A/</span></code> Match 2nd start/stop range and also 3rd NVTX push/pop range A in the app.</p>
<p><code class="docutils literal notranslate"><span class="pre">--range-filter</span> <span class="pre">yes:2:3</span> <span class="pre">--nvtx-include</span> <span class="pre">A/</span></code> Match 3rd NVTX push/pop range A from 2nd start/stop range.</p>
</td>
</tr>
<tr class="row-odd"><td><p>kill</p></td>
<td><p>Terminate the target application when the requested –launch-count was profiled. Allowed values:</p>
<ul class="simple">
<li><p>on/off</p></li>
<li><p>yes/no</p></li>
</ul>
</td>
<td><p>no</p></td>
</tr>
<tr class="row-even"><td><p>replay-mode</p></td>
<td><p>Mechanism used for replaying a kernel launch multiple times to collect all requested profiling data:</p>
<ul class="simple">
<li><p><strong>kernel:</strong> Replay individual kernel launches “transparently” during the execution of the application. See <a class="reference external" href="../ProfilingGuide/index.html#kernel-replay">Kernel Replay</a> for more details.</p></li>
<li><p><strong>application:</strong> Relaunch the entire application multiple times. Requires deterministic program execution. See <a class="reference external" href="../ProfilingGuide/index.html#application-replay">Application Replay</a> for more details.</p></li>
<li><p><strong>range:</strong> Replay ranges of CUDA API calls and kernel launches “transparently” during the execution of the application. Ranges must be defined using <code class="docutils literal notranslate"><span class="pre">cu(da)ProfilerStart/Stop</span></code> API pairs or <a class="reference external" href="index.html#nvtx-filtering">NVTX expressions</a>. See <a class="reference external" href="../ProfilingGuide/index.html#range-replay">Range Replay</a> for more details.</p></li>
<li><p><strong>app-range:</strong> Profile ranges without API capture by relaunching the entire application multiple times. Requires deterministic program execution. Ranges must be defined using ``cu(da)ProfilerStart/Stop`` API pairs or `NVTX expressions &lt;index.html#nvtx-filtering&gt;`__. See `Application Range Replay &lt;../ProfilingGuide/index.html#application-range-replay&gt;`__ for more details.</p></li>
</ul>
</td>
<td><p>kernel</p></td>
</tr>
<tr class="row-odd"><td><p>app-replay-buffer</p></td>
<td><p>Application replay buffer location.</p>
<ul class="simple">
<li><p><strong>file:</strong> Replay pass data is buffered in a temporary file. The report is created after profiling completed. This mode is more scalable, as the amount of required memory does not scale with the number of profiled kernels.</p></li>
<li><p><strong>memory:</strong> Replay pass data is buffered in memory, and the report is created while profiling. This mode can result in better performance if the filesystem is slow, but the amount of required memory scales with the number of profiled kernels.</p></li>
</ul>
</td>
<td><p>file</p></td>
</tr>
<tr class="row-even"><td><p>app-replay-match</p></td>
<td><p>Application replay kernel matching strategy. For all options, kernels are matched on a per-process and per-device (GPU) basis. Below options are used to configure the applied strategy in more detail.</p>
<ul class="simple">
<li><p><strong>name:</strong> Kernels are matched in the following order: 1. (mangled) name, 2. order of execution</p></li>
<li><p><strong>grid:</strong> Kernels are matched in the following order: 1. (mangled) name, 2. CUDA grid/block size, 3. order of execution</p></li>
<li><p><strong>all:</strong> Kernels are matched in the following order: 1. (mangled) name, 2. CUDA grid/block size, 3. CUDA context ID, 4. CUDA stream ID, 5. order of execution</p></li>
</ul>
</td>
<td><p>grid</p></td>
</tr>
<tr class="row-odd"><td><p>app-replay-mode</p></td>
<td><p>Application replay kernel matching mode:</p>
<ul class="simple">
<li><p><strong>strict:</strong> Requires all kernels to match across all replay passes.</p></li>
<li><p><strong>relaxed:</strong> Produces results only for kernels that could be matched across replay passes.</p></li>
</ul>
</td>
<td><p>strict</p></td>
</tr>
<tr class="row-even"><td><p>range-replay-options</p></td>
<td><p>Range replay options, separated by comma. Below options are supported:</p>
<ul>
<li><p><strong>enable-greedy-sync</strong></p>
<p>Insert ctx sync for applicable deferred APIs during capture.</p>
</li>
<li><p><strong>disable-host-restore</strong></p>
<p>Disable restoring device-written host allocations.</p>
</li>
</ul>
</td>
<td><p>none</p></td>
</tr>
<tr class="row-odd"><td><p>graph-profiling</p></td>
<td><p>CUDA graph profiling mode:</p>
<ul>
<li><p><strong>node</strong></p>
<p>Profile individual kernel nodes as regular CUDA kernels.</p>
</li>
<li><p><strong>graph</strong></p>
<p>Profile entire graphs as one workload (but disable profiling of individual graph kernel nodes). See the <a class="reference external" href="../ProfilingGuide/index.html#graph-profiling">Kernel Profiling Guide</a> for more information on this mode.</p>
</li>
</ul>
</td>
<td><p>node</p></td>
</tr>
<tr class="row-even"><td><p>list-sections</p></td>
<td><p>List all sections found in the searched section folders and exit.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>section</p></td>
<td><p>Add a section identifier to collect in one of the following ways:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">&lt;section</span> <span class="pre">identifier&gt;</span></code> Set the section identifier for an exact match.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">regex:&lt;expression&gt;</span></code> Regular expression allows matching full section identifier. For example, <code class="docutils literal notranslate"><span class="pre">.*Stats</span></code>, matches all sections ending with ‘Stats’. On shells that recognize regular expression symbols as special characters (e.g. Linux bash), the expression needs to be escaped with quotes, e.g. <code class="docutils literal notranslate"><span class="pre">--section</span> <span class="pre">&quot;regex:.*Stats&quot;</span></code>.</p></li>
</ul>
<p>This option is ignored when used with <code class="docutils literal notranslate"><span class="pre">--import</span></code> and <code class="docutils literal notranslate"><span class="pre">--page</span> <span class="pre">raw</span></code> or <code class="docutils literal notranslate"><span class="pre">--page</span> <span class="pre">source</span></code>. <a class="footnote-reference brackets" href="#fn1" id="id12">1</a></p>
</td>
<td><p>If no <code class="docutils literal notranslate"><span class="pre">--section</span></code> options are given, the sections associated with the <code class="docutils literal notranslate"><span class="pre">basic</span></code> set are collected. If no sets are found, all sections are collected.</p></td>
</tr>
<tr class="row-even"><td><p>section-folder</p></td>
<td><p>Add a non-recursive search path for .section files. Section files in this folder will be made available to the <code class="docutils literal notranslate"><span class="pre">--section</span></code> option.</p></td>
<td><p>If no <code class="docutils literal notranslate"><span class="pre">--section-folder</span></code> options are given, the <code class="docutils literal notranslate"><span class="pre">sections</span></code> folder is added by default.</p></td>
</tr>
<tr class="row-odd"><td><p>section-folder-recursive</p></td>
<td><p>Add a recursive search path for .section files. Section files in this folder and all folders below will be made available to the <code class="docutils literal notranslate"><span class="pre">--section</span></code> option.</p></td>
<td><p>If no <code class="docutils literal notranslate"><span class="pre">--section-folder</span></code> options are given, the <code class="docutils literal notranslate"><span class="pre">sections</span></code> folder is added by default.</p></td>
</tr>
<tr class="row-even"><td><p>list-rules</p></td>
<td><p>List all rules found in the searched section folders and exit.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>apply-rules</p></td>
<td><p>Apply active and applicable rules to each profiling result. Use <code class="docutils literal notranslate"><span class="pre">--rule</span></code> to limit which rules to apply. Allowed values:</p>
<ul class="simple">
<li><p>on/off</p></li>
<li><p>yes/no</p></li>
</ul>
</td>
<td><p>yes</p></td>
</tr>
<tr class="row-even"><td><p>rule</p></td>
<td><p>Add a rule identifier to apply. Implies <code class="docutils literal notranslate"><span class="pre">--apply-rules</span> <span class="pre">yes</span></code>.</p></td>
<td><p>If no <code class="docutils literal notranslate"><span class="pre">--rule</span></code> options are given, all applicable rules in the <code class="docutils literal notranslate"><span class="pre">sections</span></code> folder are applied.</p></td>
</tr>
<tr class="row-odd"><td><p>import-source</p></td>
<td><p>If available from -lineinfo, correlated CUDA source files are permanently imported into the report. Allowed values:</p>
<ul class="simple">
<li><p>on/off</p></li>
<li><p>yes/no</p></li>
</ul>
<p>Use <code class="docutils literal notranslate"><span class="pre">--source-folders</span></code> option to provide missing source files.</p>
</td>
<td><p>no</p></td>
</tr>
<tr class="row-even"><td><p>source-folders</p></td>
<td><p>Add comma separated recursive search paths for missing CUDA source files to import into the report.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>list-metrics</p></td>
<td><p>List all metrics collected from active sections. If the list of active sections is restricted using the <code class="docutils literal notranslate"><span class="pre">--section</span></code> option, only metrics from those sections will be listed.</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>query-metrics</p></td>
<td><p>Query available metrics for the devices on system. Use <code class="docutils literal notranslate"><span class="pre">--devices</span></code> and <code class="docutils literal notranslate"><span class="pre">--chips</span></code> to filter which devices to query. Note that by default, listed metric names need to be appended a valid suffix in order for them to become valid metrics. See <code class="docutils literal notranslate"><span class="pre">--query-metrics-mode</span></code> for how to get the list of valid suffixes, or check the <a class="reference external" href="../ProfilingGuide/index.html#metrics-structure">Kernel Profiling Guide</a>.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>query-metrics-mode</p></td>
<td><p>Set the mode for querying metrics. Implies <code class="docutils literal notranslate"><span class="pre">--query-metrics</span></code>. Available modes:</p>
<ul class="simple">
<li><p><strong>base:</strong> Only the base names of the metrics.</p></li>
<li><p><strong>suffix:</strong> Suffix names for the base metrics. This gives the list of all metrics derived from the base metrics. Use <code class="docutils literal notranslate"><span class="pre">--metrics</span></code> to specify the base metrics to query.</p></li>
<li><p><strong>all:</strong> Full names for all metrics. This gives the list of all base metrics and their suffix metrics.</p></li>
</ul>
</td>
<td><p>base</p></td>
</tr>
<tr class="row-even"><td><p>query-metrics-collection</p></td>
<td><p>Set which metric collection kind to query. Implies <code class="docutils literal notranslate"><span class="pre">--query-metrics</span></code>. Available collections:</p>
<ul class="simple">
<li><p><strong>profiling:</strong> Query metrics available for profiling.</p></li>
<li><p><strong>pmsampling:</strong> Query metrics available for <a class="reference external" href="../ProfilingGuide/index.html#pm-sampling">PM sampling</a>.</p></li>
</ul>
</td>
<td><p>profiling</p></td>
</tr>
<tr class="row-odd"><td><p>metrics</p></td>
<td><p>Specify all metrics to be profiled, separated by comma. If no <code class="docutils literal notranslate"><span class="pre">--section</span></code> options are given, only the temporary section containing all metrics listed using this option is collected. If <code class="docutils literal notranslate"><span class="pre">--section</span></code> options are given in addition to <code class="docutils literal notranslate"><span class="pre">--metrics</span></code>, all metrics from those sections and from <code class="docutils literal notranslate"><span class="pre">--metrics</span></code> are collected.</p>
<p>Names passed to this option support the following prefixes:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">regex:&lt;expression&gt;</span></code> expands to all metrics that partially match the expression. Enclose the regular expression in ^…$ to force a full match.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">group:&lt;name&gt;</span></code> lists all metrics of the metric group with that name. See section files for valid group names.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">breakdown:&lt;metric&gt;</span></code> expands to the input metrics of the high-level throughput metric.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">pmsampling:&lt;metric&gt;</span></code> collects the metric using PM sampling. Only single-pass metrics that don’t require SASS-patching (_sass_) are supported. Using this prefix adds a timeline element to the report’s details page.</p></li>
</ul>
<p>Combining multiple prefixes is not supported.</p>
<p>If a metric requires a suffix to be valid, and neither <code class="docutils literal notranslate"><span class="pre">regex:</span></code> nor <code class="docutils literal notranslate"><span class="pre">group:</span></code> are used, this option automatically expands the name to all available first-level sub-metrics.</p>
<p>When importing a report, <code class="docutils literal notranslate"><span class="pre">:group</span></code> and <code class="docutils literal notranslate"><span class="pre">:breakdown</span></code> are not supported.</p>
<p>When using <code class="docutils literal notranslate"><span class="pre">regex:</span></code>, the expression must not include any commas.   <a class="footnote-reference brackets" href="#fn1" id="id13">1</a></p>
</td>
<td></td>
</tr>
<tr class="row-even"><td><p>disable-extra-suffixes</p></td>
<td><p>Disable the collection of extra suffixes (avg, min, max, sum) for all metrics. Only collect what is explicity specified.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>list-chips</p></td>
<td><p>List all supported chips that can be used with <code class="docutils literal notranslate"><span class="pre">--chips</span></code>.</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>chips</p></td>
<td><p>Specify the chips for querying metrics, separated by comma.</p></td>
<td><p><strong>Examples</strong></p>
<p><code class="docutils literal notranslate"><span class="pre">--chips</span> <span class="pre">gv100,tu102</span></code></p>
</td>
</tr>
<tr class="row-odd"><td><p>profile-from-start</p></td>
<td><p>Set if application should be profiled from its start. Allowed values:</p>
<ul class="simple">
<li><p>on/off</p></li>
<li><p>yes/no</p></li>
</ul>
</td>
<td><p>yes</p></td>
</tr>
<tr class="row-even"><td><p>disable-profiler-start-stop</p></td>
<td><p>Disable profiler start/stop. When enabled, <code class="docutils literal notranslate"><span class="pre">cu(da)ProfilerStart/Stop</span></code> API calls are ignored.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>quiet</p></td>
<td><p>Suppress all profiling output.</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>verbose</p></td>
<td><p>Make profiler output more verbose.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>cache-control</p></td>
<td><p>Control the behavior of the GPU caches during profiling. Allowed values:</p>
<ul class="simple">
<li><p><strong>all:</strong> All GPU caches are flushed before each kernel replay iteration during profiling. While metric values in the execution environment of the application might be slightly different without invalidating the caches, this mode offers the most reproducible metric results across the replay passes and also across multiple runs of the target application.</p></li>
<li><p><strong>none:</strong> No GPU caches are flushed during profiling. This can improve performance and better replicates the application behavior if only a single kernel replay pass is necessary for metric collection. However, some metric results will vary depending on prior GPU work, and between replay iterations. This can lead to inconsistent and out-of-bounds metric values.</p></li>
</ul>
</td>
<td><p>all</p></td>
</tr>
<tr class="row-even"><td><p>clock-control</p></td>
<td><p>Control the behavior of the GPU clocks during profiling. Allowed values:</p>
<ul class="simple">
<li><p><strong>base:</strong> GPC and memory clocks are locked to their respective base frequency during profiling. This has no impact on thermal throttling. Note that actual clocks might still vary, depending on the level of driver support for this feature. As an alternative, use <code class="docutils literal notranslate"><span class="pre">nvidia-smi</span></code> to lock the clocks externally and set this option to <code class="docutils literal notranslate"><span class="pre">none</span></code>.</p></li>
<li><p><strong>none:</strong> No GPC or memory frequencies are changed during profiling.</p></li>
<li><p><strong>reset:</strong> Reset GPC and memory clocks for all or the selected devices and exit. Use if a previous, killed execution of ncu left the GPU clocks in a locked state.</p></li>
</ul>
</td>
<td><p>base</p></td>
</tr>
<tr class="row-odd"><td><p>nvtx-include</p></td>
<td><p>Adds an include statement to the <a class="reference external" href="index.html#nvtx-filtering">NVTX filter</a>, which allows selecting kernels to profile based on NVTX ranges. <a class="footnote-reference brackets" href="#fn1" id="id14">1</a></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>nvtx-exclude</p></td>
<td><p>Adds an exclude statement to the <a class="reference external" href="index.html#nvtx-filtering">NVTX filter</a>, which allows selecting kernels to profile based on NVTX ranges. <a class="footnote-reference brackets" href="#fn1" id="id15">1</a></p></td>
<td></td>
</tr>
</tbody>
</table>
<dl class="footnote brackets">
<dt class="label" id="fn1"><span class="brackets">1</span><span class="fn-backref">(<a href="#id5">1</a>,<a href="#id6">2</a>,<a href="#id7">3</a>,<a href="#id8">4</a>,<a href="#id9">5</a>,<a href="#id10">6</a>,<a href="#id11">7</a>,<a href="#id12">8</a>,<a href="#id13">9</a>,<a href="#id14">10</a>,<a href="#id15">11</a>)</span></dt>
<dd><p>This filtering option is available when using <code class="docutils literal notranslate"><span class="pre">--import</span></code>.</p>
</dd>
</dl>
</section>
<section id="pm-sampling">
<h3><span class="section-number">4.4.5. </span>PM Sampling<a class="headerlink" href="#pm-sampling" title="Permalink to this headline"></a></h3>
<p>These options apply to <a class="reference external" href="../ProfilingGuide/index.html#overhead">PM sampling</a>. See <a class="reference external" href="index.html#command-line-options-sampling">here</a> for options used in warp state sampling.</p>
<table class="table-no-stripes docutils align-default" id="id22">
<caption><span class="caption-text">Table 5. PM Sampling Command Line Options</span><a class="headerlink" href="#id22" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 17%" />
<col style="width: 75%" />
<col style="width: 7%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>pm-sampling-interval</p></td>
<td><p>Set the PM sampling interval in cycles or ns (depending on the architecture), or determine dynamically when 0.</p></td>
<td><p>0 (auto)</p></td>
</tr>
<tr class="row-odd"><td><p>pm-sampling-buffer-size</p></td>
<td><p>Set the size of the device-sided allocation for PM sampling in bytes, or determine dynamically when 0.</p></td>
<td><p>0 (auto)</p></td>
</tr>
<tr class="row-even"><td><p>pm-sampling-max-passes</p></td>
<td><p>Set the maximum number of passes used for PM sampling, or determine dynamically when 0.</p></td>
<td><p>0 (auto)</p></td>
</tr>
</tbody>
</table>
</section>
<section id="warp-sampling">
<h3><span class="section-number">4.4.6. </span>Warp Sampling<a class="headerlink" href="#warp-sampling" title="Permalink to this headline"></a></h3>
<table class="table-no-stripes docutils align-default" id="id23">
<caption><span class="caption-text">Table 6. Warp Sampling Command Line Options</span><a class="headerlink" href="#id23" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 9%" />
<col style="width: 86%" />
<col style="width: 5%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>warp-sampling-interval</p></td>
<td><p>Set the sampling period in the range of [0..31]. The actual frequency is 2 ^ (5 + value) cycles. If set to ‘auto’, the profiler tries to automatically determine a high sampling frequency without skipping samples or overflowing the output buffer.</p></td>
<td><p>auto</p></td>
</tr>
<tr class="row-odd"><td><p>warp-sampling-max-passes</p></td>
<td><p>Set maximum number of passes used for sampling (see the <a class="reference external" href="../ProfilingGuide/index.html#overhead">Kernel Profiling Guide</a> for more details on profiling overhead).</p></td>
<td><p>5</p></td>
</tr>
<tr class="row-even"><td><p>warp-sampling-buffer-size</p></td>
<td><p>Set the size of the device-sided allocation for samples in bytes.</p></td>
<td><p>32*1024*1024</p></td>
</tr>
</tbody>
</table>
</section>
<section id="file">
<span id="command-line-options-file"></span><h3><span class="section-number">4.4.7. </span>File<a class="headerlink" href="#file" title="Permalink to this headline"></a></h3>
<table class="table-no-stripes docutils align-default" id="id24">
<caption><span class="caption-text">Table 7. File Command Line Options</span><a class="headerlink" href="#id24" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 5%" />
<col style="width: 70%" />
<col style="width: 24%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>log-file</p></td>
<td><p>Send all tool output to the specified file, or one of the standard channels. The file will be overwritten. If the file doesn’t exist, a new one will be created.”stdout” as the whole file name indicates standard output channel (stdout). “stderr” as the whole file name indicates standard error channel (stderr).”</p></td>
<td><p>If <code class="docutils literal notranslate"><span class="pre">--log-file</span></code> is not set , profile results will be printed on the console.</p></td>
</tr>
<tr class="row-odd"><td><p>o,export</p></td>
<td><p>Set the output file for writing the profile report. If not set, a temporary file will be used which is removed afterwards. The specified name supports macro expansion. See <a class="reference external" href="index.html#command-line-options-file-macros">File Macros</a> for more details.</p></td>
<td><p>If <code class="docutils literal notranslate"><span class="pre">--export</span></code> is set and no <code class="docutils literal notranslate"><span class="pre">--page</span></code> option is given, no profile results will be printed on the console.</p></td>
</tr>
<tr class="row-even"><td><p>f,force-overwrite</p></td>
<td><p>Force overwriting all output files.</p></td>
<td><p>By default, the profiler won’t overwrite existing output files and show an error instead.</p></td>
</tr>
<tr class="row-odd"><td><p>i,import</p></td>
<td><p>Set the input file for reading the profile results.</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>open-in-ui</p></td>
<td><p>Open report in UI instead of showing result on terminal. (Only available on host platforms)</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>section-folder-restore</p></td>
<td><p>Restores stock files to the default section folder or the folder specified by an accompanying <em>–section-folder</em> option. If the operation will overwrite modified files then the <em>–force-overwrite</em> option is required.</p></td>
<td></td>
</tr>
</tbody>
</table>
</section>
<section id="console-output">
<span id="command-line-options-console-output"></span><h3><span class="section-number">4.4.8. </span>Console Output<a class="headerlink" href="#console-output" title="Permalink to this headline"></a></h3>
<table class="table-no-stripes docutils align-default" id="id25">
<caption><span class="caption-text">Table 8. Console Output Command Line Options</span><a class="headerlink" href="#id25" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 6%" />
<col style="width: 67%" />
<col style="width: 28%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>csv</p></td>
<td><p>Use comma-separated values as console output. Implies –print-units base by default.</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>page</p></td>
<td><p>Select the report page to print console output for. Available pages are:</p>
<ul class="simple">
<li><p><strong>details</strong> Show results grouped as sections, include rule results. Some metrics that are collected by default (e.g. device attributes) are omitted if not specified explicitly in any section or using <code class="docutils literal notranslate"><span class="pre">--metrics</span></code>.</p></li>
<li><p><strong>raw</strong> Show all collected metrics by kernel launch.</p></li>
<li><p><strong>source</strong> Show source. See <code class="docutils literal notranslate"><span class="pre">--print-source</span></code> to select the source view.</p></li>
<li><p><strong>session</strong> Show launch settings, session info, process info and device attributes.</p></li>
</ul>
</td>
<td><p><strong>details</strong>. If no <code class="docutils literal notranslate"><span class="pre">--page</span></code> option is given and <code class="docutils literal notranslate"><span class="pre">--export</span></code> is set, no results are printed to the console output.</p></td>
</tr>
<tr class="row-even"><td><p>print-source</p></td>
<td><p>Select the source view:</p>
<ul class="simple">
<li><p><strong>sass</strong> Show SASS (assembly) instructions for each kernel launch.</p></li>
<li><p><strong>ptx</strong> Show PTX source of every cubin whose at least one kernel is profiled.</p></li>
<li><p><strong>cuda</strong> Show entire CUDA-C source file which has kernel code as per kernel launch. CLI shows CUDA source only if file exists on the host machine.</p></li>
<li><p><strong>cuda,sass</strong> Show SASS CUDA-C source correlation for each kernel launch. CLI shows CUDA source only if file exists on the host machine.</p></li>
</ul>
<p>Metric correlation with source is available in <strong>sass</strong>, and <strong>cuda,sass</strong> source view. Metrics specified with <code class="docutils literal notranslate"><span class="pre">--metrics</span></code> and specified section file with <code class="docutils literal notranslate"><span class="pre">--section</span></code> are correlated. Consider restricting the number of selected metrics such that values fit into a single output row.</p>
</td>
<td><p>sass</p></td>
</tr>
<tr class="row-odd"><td><p>resolve-source-file</p></td>
<td><p>Resolve CUDA source file in the <code class="docutils literal notranslate"><span class="pre">--page</span> <span class="pre">source</span></code> output. Provide comma separated files full path.</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>print-details</p></td>
<td><p>Select which part of a section should be shown in the details page output:</p>
<ul class="simple">
<li><p><strong>header</strong> Show all metrics from header of the section.</p></li>
<li><p><strong>body</strong> Show all metrics from body of the section.</p></li>
<li><p><strong>all</strong> Show all metrics from the section.</p></li>
</ul>
<p>Replaces deprecated option <code class="docutils literal notranslate"><span class="pre">--details-all</span></code>.</p>
</td>
<td><p>header</p></td>
</tr>
<tr class="row-odd"><td><p>print-metric-name</p></td>
<td><p>Select one of the option to show it in the Metric Name column:</p>
<ul class="simple">
<li><p><strong>label</strong> Show metric label.</p></li>
<li><p><strong>name</strong> Show metric name.</p></li>
<li><p><strong>label-name</strong> Show both metric label and metric name.</p></li>
</ul>
</td>
<td><p>label</p></td>
</tr>
<tr class="row-even"><td><p>print-units</p></td>
<td><p>Select the mode for scaling of metric units. Available modes are:</p>
<ul class="simple">
<li><p><strong>auto</strong> Show all metrics automatically scaled to the most fitting order of magnitude.</p></li>
<li><p><strong>base</strong> Show all metrics in their base unit.</p></li>
</ul>
</td>
<td><p>auto</p></td>
</tr>
<tr class="row-odd"><td><p>print-fp</p></td>
<td><p>Show all numeric metrics in the console output as floating point numbers.</p></td>
<td><p>false</p></td>
</tr>
<tr class="row-even"><td><p>print-kernel-base</p></td>
<td><p>Set the basis for kernel name output. See <code class="docutils literal notranslate"><span class="pre">--kernel-regex-base</span></code> for options.</p></td>
<td><p>demangled</p></td>
</tr>
<tr class="row-odd"><td><p>print-metric-instances</p></td>
<td><p>Set output mode for metrics with instance values:</p>
<ul class="simple">
<li><p><strong>none</strong> Only show GPU aggregate value.</p></li>
<li><p><strong>values</strong> Show GPU aggregate followed by all instance values.</p></li>
</ul>
</td>
<td><p>none</p></td>
</tr>
<tr class="row-even"><td><p>print-nvtx-rename</p></td>
<td><p>Select how NVTX should be used for renaming:</p>
<ul class="simple">
<li><p><strong>none</strong> Don’t use NVTX for renaming.</p></li>
<li><p><strong>kernel</strong> Rename kernels with the most recent enclosing NVTX push/pop range.</p></li>
</ul>
</td>
<td><p>none</p></td>
</tr>
<tr class="row-odd"><td><p>print-rule-details</p></td>
<td><p>Print additional details for rule results, such as the triggering metrics. Currently has no effect in CSV mode.</p></td>
<td><p>false</p></td>
</tr>
<tr class="row-even"><td><p>print-summary</p></td>
<td><p>Select the summary output mode. Available modes are:</p>
<ul class="simple">
<li><p><strong>none</strong> No summary.</p></li>
<li><p><strong>per-gpu</strong> Summary for each GPU.</p></li>
<li><p><strong>per-kernel</strong> Summary for each kernel type.</p></li>
<li><p><strong>per-nvtx</strong> Summary for each NVTX context.</p></li>
</ul>
</td>
<td><p>none</p></td>
</tr>
</tbody>
</table>
</section>
<section id="response-file">
<h3><span class="section-number">4.4.9. </span>Response File<a class="headerlink" href="#response-file" title="Permalink to this headline"></a></h3>
<p>Response files can be specified by adding <code class="docutils literal notranslate"><span class="pre">&#64;FileName</span></code> to the command line. The file name must immediately follow the <code class="docutils literal notranslate"><span class="pre">&#64;</span></code> character. The content of each response file is inserted in place of the corresponding response file option.</p>
</section>
<section id="file-macros">
<span id="command-line-options-file-macros"></span><h3><span class="section-number">4.4.10. </span>File Macros<a class="headerlink" href="#file-macros" title="Permalink to this headline"></a></h3>
<p>The file name specified with option <code class="docutils literal notranslate"><span class="pre">-o</span></code> or <code class="docutils literal notranslate"><span class="pre">--export</span></code> supports the following macro expansions. Occurrences of these macros in the report file name are replaced by the corresponding character sequence. If not specified otherwise, the macros cannot be used as part of the file path.</p>
<table class="table-no-stripes docutils align-default" id="id26">
<caption><span class="caption-text">Table 9. Macro Expansions</span><a class="headerlink" href="#id26" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 8%" />
<col style="width: 92%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Macro</p></th>
<th class="head"><p>Description</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>%h</p></td>
<td><p>Expands to the host name of the machine on which the command line profiler is running.</p></td>
</tr>
<tr class="row-odd"><td><p>%q{ENV_NAME}</p></td>
<td><p>Expands to the content of the variable with the given name <code class="docutils literal notranslate"><span class="pre">ENV_NAME</span></code> from the environment of the command line profiler.</p></td>
</tr>
<tr class="row-even"><td><p>%p</p></td>
<td><p>Expands to the process ID of the command line profiler.</p></td>
</tr>
<tr class="row-odd"><td><p>%i</p></td>
<td><p>Expands to the lowest unused positive integer number that guarantees the resulting file name is not yet used. This macro can only be used once in the output file name.</p></td>
</tr>
<tr class="row-even"><td><p>%%</p></td>
<td><p>Expands to a single <code class="docutils literal notranslate"><span class="pre">%</span></code> character in the output file name. This macro can be used in the file path and the file name.</p></td>
</tr>
</tbody>
</table>
</section>
</section>
<section id="environment-variables">
<h2><span class="section-number">4.5. </span>Environment Variables<a class="headerlink" href="#environment-variables" title="Permalink to this headline"></a></h2>
<p>The following environment variables can be set before launching NVIDIA Nsight Compute CLI, or the UI, respectively.</p>
<table class="table-no-stripes docutils align-default" id="id27">
<caption><span class="caption-text">Table 10. Environment Variables</span><a class="headerlink" href="#id27" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 7%" />
<col style="width: 65%" />
<col style="width: 29%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Name</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Default/Values</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>NV_COMPUTE_PROFILER_DISABLE_STOCK_FILE_DEPLOYMENT</p></td>
<td><p>Disable file deployment to the versioned <code class="docutils literal notranslate"><span class="pre">Sections</span></code> directory, using section and rule files from the stock directory within the installation instead.</p>
<p>By default, the versioned directory from the user’s documents folder is used to ensure that any potential user updates are taken into account.</p>
<p>Only supported in the NVIDIA Nsight Compute CLI.</p>
</td>
<td><p>Default: unset</p>
<p>Set to “1” to disable deployment.</p>
</td>
</tr>
<tr class="row-odd"><td><p>NV_COMPUTE_PROFILER_LOCAL_CONNECTION_OVERRIDE</p></td>
<td><p>Override the default local connection mechanism between frontend and profiled target processes. The default mechanism is platform-dependent. This should only be used if there are connection problems between frontend and target processes in a local launch.</p></td>
<td><p>Default: unset (use default mechanism)</p>
<p>Set to “uds” to use Unix Domain Socket connections (available on Posix platforms, only). Set to “tcp” to use TCP (available on all platforms). Set to “named-pipes” to use Windows Named Pipes (available on Windows, only).</p>
</td>
</tr>
<tr class="row-even"><td><p>NV_COMPUTE_PROFILER_DISABLE_SW_PRE_PASS</p></td>
<td><p>Disable the instruction-level software (SW) metric pre-pass. When collecting SW-patched metrics, such as <code class="docutils literal notranslate"><span class="pre">inst_executed</span></code>, the pre-pass is used to determine which functions are executed as part of the kernel and should be patched. This requires a separate replay pass, and if only instruction-level SW metrics are to be collected, prevents single-pass data collection. Disabling the pre-pass can improve performance if memory save-and-restore is undesirable and application replay is not possible.</p></td>
<td><p>Default: unset (use pre-pass when applicable)</p>
<p>Set to “1” to disable pre-pass.</p>
</td>
</tr>
</tbody>
</table>
</section>
<section id="nvprof-transition-guide">
<span id="nvprof-guide"></span><h2><span class="section-number">4.6. </span>Nvprof Transition Guide<a class="headerlink" href="#nvprof-transition-guide" title="Permalink to this headline"></a></h2>
<p>This guide provides tips for moving from nvprof to NVIDIA Nsight Compute CLI. NVIDIA Nsight Compute CLI tries to provide as much feature and usage parity as possible with nvprof, but some features are now covered by different tools and some command line options have changed their name or meaning.</p>
<section id="trace">
<span id="nvprof-trace"></span><h3><span class="section-number">4.6.1. </span>Trace<a class="headerlink" href="#trace" title="Permalink to this headline"></a></h3>
<ul>
<li><p><strong>GPU and API trace</strong></p>
<p>NVIDIA Nsight Compute CLI does not support any form of tracing GPU or API activities. This functionality is covered by <a class="reference external" href="https://developer.nvidia.com/nsight-systems">NVIDIA Nsight Systems</a>.</p>
</li>
</ul>
</section>
<section id="metric-collection">
<span id="nvprof-metric-collection"></span><h3><span class="section-number">4.6.2. </span>Metric Collection<a class="headerlink" href="#metric-collection" title="Permalink to this headline"></a></h3>
<ul>
<li><p><strong>Finding available metrics</strong></p>
<p>For nvprof, you can use <code class="docutils literal notranslate"><span class="pre">--query-metrics</span></code> to see the list of metrics available for the current devices on your machine. You can also use <code class="docutils literal notranslate"><span class="pre">--devices</span></code> to filter which local devices to query. For NVIDIA Nsight Compute CLI, this functionality is the same. However, in addition, you can combine <code class="docutils literal notranslate"><span class="pre">--query-metrics</span></code> with <code class="docutils literal notranslate"><span class="pre">--chip</span> <span class="pre">[chipname]</span></code> to query the available metrics for any chip, not only the ones in your present CUDA devices.</p>
<p>Note that metric names have changed between nvprof and NVIDIA Nsight Compute CLI and metric names also differ between chips after (and including) GV100 and those before. See <a class="reference external" href="index.html#nvprof-metric-comparison">Metric Comparison</a> for a comparison of nvprof and NVIDIA Nsight Compute metric names.</p>
<p>On Volta and newer GPUs, most metrics are named using a base name and various suffixes, e.g. <em>sm__throughput.avg.pct_of_peak_sustained_elapsed</em>. The base name is <em>sm__throughput</em> and the suffix is <em>avg.pct_of_peak_sustained_elapsed</em>. This is because most metrics follow the same structure and have the same set of suffixes. You need to pass the full name to NVIDIA Nsight Compute when selecting a metric for profiling.</p>
<p>To reduce the number of metrics shown for Volta and newer GPUs when using <code class="docutils literal notranslate"><span class="pre">--query-metrics</span></code>, by default only the base names are shown. Use <code class="docutils literal notranslate"><span class="pre">--query-metrics-mode</span> <span class="pre">suffix</span> <span class="pre">--metrics</span> <span class="pre">&lt;metrics</span> <span class="pre">list&gt;</span></code> to see the full names for the chosen metrics. Use <code class="docutils literal notranslate"><span class="pre">--query-metrics-mode</span> <span class="pre">all</span></code> to see all metrics with their full name directly.</p>
</li>
<li><p><strong>Selecting which metrics to collect</strong></p>
<p>In both nvprof and NVIDIA Nsight Compute CLI, you can specify a comma-separated list of metric names to the <code class="docutils literal notranslate"><span class="pre">--metrics</span></code> option. While nvprof would allow you to collect either a list or all metrics, in NVIDIA Nsight Compute CLI you can use regular expressions to select a more fine-granular subset of all available metrics. For example, you can use <code class="docutils literal notranslate"><span class="pre">--metrics</span> <span class="pre">&quot;regex:.*&quot;</span></code> to collect all metrics, or <code class="docutils literal notranslate"><span class="pre">--metrics</span> <span class="pre">&quot;regex:smsp__cycles_elapsed&quot;</span></code> to collect all “smsp__cycles_elapsed” metrics.</p>
</li>
<li><p><strong>Selecting which events to collect</strong></p>
<p>You cannot collect any events in NVIDIA Nsight Compute CLI.</p>
</li>
<li><p><strong>Selecting which section to collect</strong></p>
<p>In nvprof, you can either collect individual metrics or events, or a pre-configured set (all, analysis-metrics). NVIDIA Nsight Compute CLI adds the concept of a section. A section is a file that describes which metrics to collect for which GPU architecture, or architecture range. Furthermore, it defines how those metrics will be shown in both the command line output or the user interface. This includes structuring in tables, charts, histograms etc.</p>
<p>NVIDIA Nsight Compute CLI comes with a set of pre-defined sections, located in the <code class="docutils literal notranslate"><span class="pre">sections</span></code> directory. You can inspect, modify or extend those, as well as add new ones, e.g. to easily collect recurring metric sets. Each section specifies a unique <em>section identifier</em>, and there must not be two sections with the same identifier in the search path.</p>
<p>By default, the sections associated with the <code class="docutils literal notranslate"><span class="pre">basic</span></code> section set are collected. You can select one or more individual sections using the <code class="docutils literal notranslate"><span class="pre">--section</span> <span class="pre">[section</span> <span class="pre">identifier]</span></code> option one or more times. If no <code class="docutils literal notranslate"><span class="pre">--section</span></code> option is given, but <code class="docutils literal notranslate"><span class="pre">--metrics</span></code> is used, no sections will be collected.</p>
</li>
<li><p><strong>Selecting which section set to collect</strong></p>
<p>In nvprof, you can either collect individual metrics or events, or a pre-configured set (all, analysis-metrics). NVIDIA Nsight Compute CLI adds the concept of <em>section sets</em>. A section set defines a group of sections to collect together, in order to achieve different profiling overheads, depending on the required analysis level of detail.</p>
<p>If no other options are selected, the <code class="docutils literal notranslate"><span class="pre">basic</span></code> section set is collected. You can select one or more sets using the <code class="docutils literal notranslate"><span class="pre">--set</span> <span class="pre">[set</span> <span class="pre">identifier]</span></code> option one or more times. If no <code class="docutils literal notranslate"><span class="pre">--set</span></code> option is given, but <code class="docutils literal notranslate"><span class="pre">--section</span></code> or <code class="docutils literal notranslate"><span class="pre">--metrics</span></code> is used, no sets will be collected.</p>
</li>
</ul>
</section>
<section id="metric-comparison">
<span id="nvprof-metric-comparison"></span><h3><span class="section-number">4.6.3. </span>Metric Comparison<a class="headerlink" href="#metric-comparison" title="Permalink to this headline"></a></h3>
<p>NVIDIA Nsight Compute uses two groups of metrics, depending on which GPU architecture is profiled. For nvprof metrics, the following table lists the equivalent metrics in NVIDIA Nsight Compute, if available. For a detailed explanation of the structuring of PerfWorks metrics, see <a class="reference external" href="../ProfilingGuide/index.html#metrics-structure">Metrics Structure</a>.</p>
<p>Metrics starting with <em>sm__&amp; are collected per-SM. Metrics starting with *smsp__</em> are collected per-SM subpartition. However, all corresponding nvprof events are collected per-SM, only. Check the <a class="reference external" href="../ProfilingGuide/index.html#metrics-guide">Metrics Guide</a> for more details on these terms.</p>
<table class="table-no-stripes docutils align-default" id="id28">
<caption><span class="caption-text">Table 11. Metrics Mapping Table from CUPTI to PerfWorks</span><a class="headerlink" href="#id28" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 8%" />
<col style="width: 92%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>nvprof Metric</p></th>
<th class="head"><p>PerfWorks Metric or Formula (&gt;= SM 7.0)</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>achieved_occupancy</p></td>
<td><p>sm__warps_active.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-odd"><td><p>atomic_transactions</p></td>
<td><p>l1tex__t_set_accesses_pipe_lsu_mem_global_op_atom.sum + l1tex__t_set_accesses_pipe_lsu_mem_global_op_red.sum</p></td>
</tr>
<tr class="row-even"><td><p>atomic_transactions_per_request</p></td>
<td><p>(l1tex__t_sectors_pipe_lsu_mem_global_op_atom.sum + l1tex__t_sectors_pipe_lsu_mem_global_op_red.sum) / (l1tex__t_requests_pipe_lsu_mem_global_op_atom.sum + l1tex__t_requests_pipe_lsu_mem_global_op_red.sum)</p></td>
</tr>
<tr class="row-odd"><td><p>branch_efficiency</p></td>
<td><p>smsp__sass_average_branch_targets_threads_uniform.pct</p></td>
</tr>
<tr class="row-even"><td><p>cf_executed</p></td>
<td><p>smsp__inst_executed_pipe_cbu.sum + smsp__inst_executed_pipe_adu.sum</p></td>
</tr>
<tr class="row-odd"><td><p>cf_fu_utilization</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>cf_issued</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>double_precision_fu_utilization</p></td>
<td><p>smsp__inst_executed_pipe_fp64.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-even"><td><p>dram_read_bytes</p></td>
<td><p>dram__bytes_read.sum</p></td>
</tr>
<tr class="row-odd"><td><p>dram_read_throughput</p></td>
<td><p>dram__bytes_read.sum.per_second</p></td>
</tr>
<tr class="row-even"><td><p>dram_read_transactions</p></td>
<td><p>dram__sectors_read.sum</p></td>
</tr>
<tr class="row-odd"><td><p>dram_utilization</p></td>
<td><p>dram__throughput.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-even"><td><p>dram_write_bytes</p></td>
<td><p>dram__bytes_write.sum</p></td>
</tr>
<tr class="row-odd"><td><p>dram_write_throughput</p></td>
<td><p>dram__bytes_write.sum.per_second</p></td>
</tr>
<tr class="row-even"><td><p>dram_write_transactions</p></td>
<td><p>dram__sectors_write.sum</p></td>
</tr>
<tr class="row-odd"><td><p>eligible_warps_per_cycle</p></td>
<td><p>smsp__warps_eligible.sum.per_cycle_active</p></td>
</tr>
<tr class="row-even"><td><p>flop_count_dp</p></td>
<td><p>smsp__sass_thread_inst_executed_op_dadd_pred_on.sum + smsp__sass_thread_inst_executed_op_dmul_pred_on.sum + smsp__sass_thread_inst_executed_op_dfma_pred_on.sum * 2</p></td>
</tr>
<tr class="row-odd"><td><p>flop_count_dp_add</p></td>
<td><p>smsp__sass_thread_inst_executed_op_dadd_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>flop_count_dp_fma</p></td>
<td><p>smsp__sass_thread_inst_executed_op_dfma_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>flop_count_dp_mul</p></td>
<td><p>smsp__sass_thread_inst_executed_op_dmul_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>flop_count_hp</p></td>
<td><p>smsp__sass_thread_inst_executed_op_hadd_pred_on.sum + smsp__sass_thread_inst_executed_op_hmul_pred_on.sum + smsp__sass_thread_inst_executed_op_hfma_pred_on.sum * 2</p></td>
</tr>
<tr class="row-odd"><td><p>flop_count_hp_add</p></td>
<td><p>smsp__sass_thread_inst_executed_op_hadd_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>flop_count_hp_fma</p></td>
<td><p>smsp__sass_thread_inst_executed_op_hfma_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>flop_count_hp_mul</p></td>
<td><p>smsp__sass_thread_inst_executed_op_hmul_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>flop_count_sp</p></td>
<td><p>smsp__sass_thread_inst_executed_op_fadd_pred_on.sum + smsp__sass_thread_inst_executed_op_fmul_pred_on.sum + smsp__sass_thread_inst_executed_op_ffma_pred_on.sum * 2</p></td>
</tr>
<tr class="row-odd"><td><p>flop_count_sp_add</p></td>
<td><p>smsp__sass_thread_inst_executed_op_fadd_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>flop_count_sp_fma</p></td>
<td><p>smsp__sass_thread_inst_executed_op_ffma_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>flop_count_sp_mul</p></td>
<td><p>smsp__sass_thread_inst_executed_op_fmul_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>flop_count_sp_special</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>flop_dp_efficiency</p></td>
<td><p>smsp__sass_thread_inst_executed_ops_dadd_dmul_dfma_pred_on.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-even"><td><p>flop_hp_efficiency</p></td>
<td><p>smsp__sass_thread_inst_executed_ops_hadd_hmul_hfma_pred_on.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-odd"><td><p>flop_sp_efficiency</p></td>
<td><p>smsp__sass_thread_inst_executed_ops_fadd_fmul_ffma_pred_on.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-even"><td><p>gld_efficiency</p></td>
<td><p>smsp__sass_average_data_bytes_per_sector_mem_global_op_ld.pct</p></td>
</tr>
<tr class="row-odd"><td><p>gld_requested_throughput</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>gld_throughput</p></td>
<td><p>l1tex__t_bytes_pipe_lsu_mem_global_op_ld.sum.per_second</p></td>
</tr>
<tr class="row-odd"><td><p>gld_transactions</p></td>
<td><p>l1tex__t_sectors_pipe_lsu_mem_global_op_ld.sum</p></td>
</tr>
<tr class="row-even"><td><p>gld_transactions_per_request</p></td>
<td><p>l1tex__average_t_sectors_per_request_pipe_lsu_mem_global_op_ld.ratio</p></td>
</tr>
<tr class="row-odd"><td><p>global_atomic_requests</p></td>
<td><p>l1tex__t_requests_pipe_lsu_mem_global_op_atom.sum</p></td>
</tr>
<tr class="row-even"><td><p>global_hit_rate</p></td>
<td><p>(l1tex__t_sectors_pipe_lsu_mem_global_op_ld_lookup_hit.sum + l1tex__t_sectors_pipe_lsu_mem_global_op_st_lookup_hit.sum + l1tex__t_sectors_pipe_lsu_mem_global_op_red_lookup_hit.sum + l1tex__t_sectors_pipe_lsu_mem_global_op_atom_lookup_hit.sum) / (l1tex__t_sectors_pipe_lsu_mem_global_op_ld.sum + l1tex__t_sectors_pipe_lsu_mem_global_op_st.sum + l1tex__t_sectors_pipe_lsu_mem_global_op_red.sum + l1tex__t_sectors_pipe_lsu_mem_global_op_atom.sum)</p></td>
</tr>
<tr class="row-odd"><td><p>global_load_requests</p></td>
<td><p>l1tex__t_requests_pipe_lsu_mem_global_op_ld.sum</p></td>
</tr>
<tr class="row-even"><td><p>global_reduction_requests</p></td>
<td><p>l1tex__t_requests_pipe_lsu_mem_global_op_red.sum</p></td>
</tr>
<tr class="row-odd"><td><p>global_store_requests</p></td>
<td><p>l1tex__t_requests_pipe_lsu_mem_global_op_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>gst_efficiency</p></td>
<td><p>smsp__sass_average_data_bytes_per_sector_mem_global_op_st.pct</p></td>
</tr>
<tr class="row-odd"><td><p>gst_requested_throughput</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>gst_throughput</p></td>
<td><p>l1tex__t_bytes_pipe_lsu_mem_global_op_st.sum.per_second</p></td>
</tr>
<tr class="row-odd"><td><p>gst_transactions</p></td>
<td><p>l1tex__t_sectors_pipe_lsu_mem_global_op_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>gst_transactions_per_request</p></td>
<td><p>l1tex__average_t_sectors_per_request_pipe_lsu_mem_global_op_st.ratio</p></td>
</tr>
<tr class="row-odd"><td><p>half_precision_fu_utilization</p></td>
<td><p>smsp__inst_executed_pipe_fp16.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-even"><td><p>inst_bit_convert</p></td>
<td><p>smsp__sass_thread_inst_executed_op_conversion_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_compute_ld_st</p></td>
<td><p>smsp__sass_thread_inst_executed_op_memory_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_control</p></td>
<td><p>smsp__sass_thread_inst_executed_op_control_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed</p></td>
<td><p>smsp__inst_executed.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_global_atomics</p></td>
<td><p>smsp__sass_inst_executed_op_global_atom.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_global_loads</p></td>
<td><p>smsp__inst_executed_op_global_ld.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_global_reductions</p></td>
<td><p>smsp__inst_executed_op_global_red.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_global_stores</p></td>
<td><p>smsp__inst_executed_op_global_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_local_loads</p></td>
<td><p>smsp__inst_executed_op_local_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_local_stores</p></td>
<td><p>smsp__inst_executed_op_local_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_shared_atomics</p></td>
<td><p>smsp__inst_executed_op_shared_atom.sum + smsp__inst_executed_op_shared_atom_dot_alu.sum + smsp__inst_executed_op_shared_atom_dot_cas.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_shared_loads</p></td>
<td><p>smsp__inst_executed_op_shared_ld.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_shared_stores</p></td>
<td><p>smsp__inst_executed_op_shared_st.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_surface_atomics</p></td>
<td><p>smsp__inst_executed_op_surface_atom.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_surface_loads</p></td>
<td><p>smsp__inst_executed_op_surface_ld.sum + smsp__inst_executed_op_shared_atom_dot_alu.sum + smsp__inst_executed_op_shared_atom_dot_cas.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_surface_reductions</p></td>
<td><p>smsp__inst_executed_op_surface_red.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_surface_stores</p></td>
<td><p>smsp__inst_executed_op_surface_st.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_tex_ops</p></td>
<td><p>smsp__inst_executed_op_texture.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_fp_16</p></td>
<td><p>smsp__sass_thread_inst_executed_op_fp16_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_fp_32</p></td>
<td><p>smsp__sass_thread_inst_executed_op_fp32_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_fp_64</p></td>
<td><p>smsp__sass_thread_inst_executed_op_fp64_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_integer</p></td>
<td><p>smsp__sass_thread_inst_executed_op_integer_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_inter_thread_communication</p></td>
<td><p>smsp__sass_thread_inst_executed_op_inter_thread_communication_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_issued</p></td>
<td><p>smsp__inst_issued.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_misc</p></td>
<td><p>smsp__sass_thread_inst_executed_op_misc_pred_on.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_per_warp</p></td>
<td><p>smsp__average_inst_executed_per_warp.ratio</p></td>
</tr>
<tr class="row-even"><td><p>inst_replay_overhead</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>ipc</p></td>
<td><p>smsp__inst_executed.avg.per_cycle_active</p></td>
</tr>
<tr class="row-even"><td><p>issue_slot_utilization</p></td>
<td><p>smsp__issue_active.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-odd"><td><p>issue_slots</p></td>
<td><p>smsp__inst_issued.sum</p></td>
</tr>
<tr class="row-even"><td><p>issued_ipc</p></td>
<td><p>smsp__inst_issued.avg.per_cycle_active</p></td>
</tr>
<tr class="row-odd"><td><p>l2_atomic_throughput</p></td>
<td><p>2 * ( lts__t_sectors_op_atom.sum.per_second + lts__t_sectors_op_red.sum.per_second )</p></td>
</tr>
<tr class="row-even"><td><p>l2_atomic_transactions</p></td>
<td><p>2 * ( lts__t_sectors_op_atom.sum + lts__t_sectors_op_red.sum )</p></td>
</tr>
<tr class="row-odd"><td><p>l2_global_atomic_store_bytes</p></td>
<td><p>lts__t_bytes_equiv_l1sectormiss_pipe_lsu_mem_global_op_atom.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_global_load_bytes</p></td>
<td><p>lts__t_bytes_equiv_l1sectormiss_pipe_lsu_mem_global_op_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_local_global_store_bytes</p></td>
<td><p>lts__t_bytes_equiv_l1sectormiss_pipe_lsu_mem_local_op_st.sum + lts__t_bytes_equiv_l1sectormiss_pipe_lsu_mem_global_op_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_local_load_bytes</p></td>
<td><p>lts__t_bytes_equiv_l1sectormiss_pipe_lsu_mem_local_op_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_read_throughput</p></td>
<td><p>lts__t_sectors_op_read.sum.per_second + lts__t_sectors_op_atom.sum.per_second + lts__t_sectors_op_red.sum.per_second <a class="footnote-reference brackets" href="#fn2" id="id16">2</a></p></td>
</tr>
<tr class="row-even"><td><p>l2_read_transactions</p></td>
<td><p>lts__t_sectors_op_read.sum + lts__t_sectors_op_atom.sum + lts__t_sectors_op_red.sum <a class="footnote-reference brackets" href="#fn2" id="id17">2</a></p></td>
</tr>
<tr class="row-odd"><td><p>l2_surface_load_bytes</p></td>
<td><p>lts__t_bytes_equiv_l1sectormiss_pipe_tex_mem_surface_op_ld.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_surface_store_bytes</p></td>
<td><p>lts__t_bytes_equiv_l1sectormiss_pipe_tex_mem_surface_op_st.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_tex_hit_rate</p></td>
<td><p>lts__t_sector_hit_rate.pct</p></td>
</tr>
<tr class="row-even"><td><p>l2_tex_read_hit_rate</p></td>
<td><p>lts__t_sector_op_read_hit_rate.pct</p></td>
</tr>
<tr class="row-odd"><td><p>l2_tex_read_throughput</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_read.sum.per_second</p></td>
</tr>
<tr class="row-even"><td><p>l2_tex_read_transactions</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_read.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_tex_write_hit_rate</p></td>
<td><p>lts__t_sector_op_write_hit_rate.pct</p></td>
</tr>
<tr class="row-even"><td><p>l2_tex_write_throughput</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_write.sum.per_second</p></td>
</tr>
<tr class="row-odd"><td><p>l2_tex_write_transactions</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_write.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_utilization</p></td>
<td><p>lts__t_sectors.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-odd"><td><p>l2_write_throughput</p></td>
<td><p>lts__t_sectors_op_write.sum.per_second + lts__t_sectors_op_atom.sum.per_second + lts__t_sectors_op_red.sum.per_second</p></td>
</tr>
<tr class="row-even"><td><p>l2_write_transactions</p></td>
<td><p>lts__t_sectors_op_write.sum + lts__t_sectors_op_atom.sum + lts__t_sectors_op_red.sum</p></td>
</tr>
<tr class="row-odd"><td><p>ldst_executed</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>ldst_fu_utilization</p></td>
<td><p>smsp__inst_executed_pipe_lsu.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-odd"><td><p>ldst_issued</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>local_hit_rate</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>local_load_requests</p></td>
<td><p>l1tex__t_requests_pipe_lsu_mem_local_op_ld.sum</p></td>
</tr>
<tr class="row-even"><td><p>local_load_throughput</p></td>
<td><p>l1tex__t_bytes_pipe_lsu_mem_local_op_ld.sum.per_second</p></td>
</tr>
<tr class="row-odd"><td><p>local_load_transactions</p></td>
<td><p>l1tex__t_sectors_pipe_lsu_mem_local_op_ld.sum</p></td>
</tr>
<tr class="row-even"><td><p>local_load_transactions_per_request</p></td>
<td><p>l1tex__average_t_sectors_per_request_pipe_lsu_mem_local_op_ld.ratio</p></td>
</tr>
<tr class="row-odd"><td><p>local_memory_overhead</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>local_store_requests</p></td>
<td><p>l1tex__t_requests_pipe_lsu_mem_local_op_st.sum</p></td>
</tr>
<tr class="row-odd"><td><p>local_store_throughput</p></td>
<td><p>l1tex__t_sectors_pipe_lsu_mem_local_op_st.sum.per_second</p></td>
</tr>
<tr class="row-even"><td><p>local_store_transactions</p></td>
<td><p>l1tex__t_sectors_pipe_lsu_mem_local_op_st.sum</p></td>
</tr>
<tr class="row-odd"><td><p>local_store_transactions_per_request</p></td>
<td><p>l1tex__average_t_sectors_per_request_pipe_lsu_mem_local_op_st.ratio</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_data_receive_efficiency</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_data_transmission_efficiency</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_overhead_data_received</p></td>
<td><p>(nvlrx__bytes_data_protocol.sum / nvlrx__bytes.sum) * 100</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_overhead_data_transmitted</p></td>
<td><p>(nvltx__bytes_data_protocol.sum / nvltx__bytes.sum) * 100</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_receive_throughput</p></td>
<td><p>nvlrx__bytes.sum.per_second</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_total_data_received</p></td>
<td><p>nvlrx__bytes.sum</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_total_data_transmitted</p></td>
<td><p>nvltx__bytes.sum</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_total_nratom_data_transmitted</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_total_ratom_data_transmitted</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_total_response_data_received</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_total_write_data_transmitted</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_transmit_throughput</p></td>
<td><p>nvltx__bytes.sum.per_second</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_user_data_received</p></td>
<td><p>nvlrx__bytes_data_user.sum</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_user_data_transmitted</p></td>
<td><p>nvltx__bytes_data_user.sum</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_user_nratom_data_transmitted</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_user_ratom_data_transmitted</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>nvlink_user_response_data_received</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>nvlink_user_write_data_transmitted</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>pcie_total_data_received</p></td>
<td><p>pcie__read_bytes.sum</p></td>
</tr>
<tr class="row-odd"><td><p>pcie_total_data_transmitted</p></td>
<td><p>pcie__write_bytes.sum</p></td>
</tr>
<tr class="row-even"><td><p>shared_efficiency</p></td>
<td><p>smsp__sass_average_data_bytes_per_wavefront_mem_shared.pct</p></td>
</tr>
<tr class="row-odd"><td><p>shared_load_throughput</p></td>
<td><p>l1tex__data_pipe_lsu_wavefronts_mem_shared_op_ld.sum.per_second</p></td>
</tr>
<tr class="row-even"><td><p>shared_load_transactions</p></td>
<td><p>l1tex__data_pipe_lsu_wavefronts_mem_shared_op_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>shared_load_transactions_per_request</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>shared_store_throughput</p></td>
<td><p>l1tex__data_pipe_lsu_wavefronts_mem_shared_op_st.sum.per_second</p></td>
</tr>
<tr class="row-odd"><td><p>shared_store_transactions</p></td>
<td><p>l1tex__data_pipe_lsu_wavefronts_mem_shared_op_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>shared_store_transactions_per_request</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>shared_utilization</p></td>
<td><p>l1tex__data_pipe_lsu_wavefronts_mem_shared.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-even"><td><p>single_precision_fu_utilization</p></td>
<td><p>smsp__pipe_fma_cycles_active.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-odd"><td><p>sm_efficiency</p></td>
<td><p>smsp__cycles_active.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-even"><td><p>sm_tex_utilization</p></td>
<td><p>l1tex__texin_sm2tex_req_cycles_active.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-odd"><td><p>special_fu_utilization</p></td>
<td><p>smsp__inst_executed_pipe_xu.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-even"><td><p>stall_constant_memory_dependency</p></td>
<td><p>smsp__warp_issue_stalled_imc_miss_per_warp_active.pct</p></td>
</tr>
<tr class="row-odd"><td><p>stall_exec_dependency</p></td>
<td><p>smsp__warp_issue_stalled_short_scoreboard_per_warp_active.pct + smsp__warp_issue_stalled_wait_per_warp_active.pct</p></td>
</tr>
<tr class="row-even"><td><p>stall_inst_fetch</p></td>
<td><p>smsp__warp_issue_stalled_no_instruction_per_warp_active.pct</p></td>
</tr>
<tr class="row-odd"><td><p>stall_memory_dependency</p></td>
<td><p>smsp__warp_issue_stalled_long_scoreboard_per_warp_active.pct</p></td>
</tr>
<tr class="row-even"><td><p>stall_memory_throttle</p></td>
<td><p>smsp__warp_issue_stalled_drain_per_warp_active.pct + smsp__warp_issue_stalled_lg_throttle_per_warp_active.pct</p></td>
</tr>
<tr class="row-odd"><td><p>stall_not_selected</p></td>
<td><p>smsp__warp_issue_stalled_not_selected_per_warp_active.pct</p></td>
</tr>
<tr class="row-even"><td><p>stall_other</p></td>
<td><p>smsp__warp_issue_stalled_dispatch_stall_per_warp_active.pct + smsp__warp_issue_stalled_misc_per_warp_active.pct</p></td>
</tr>
<tr class="row-odd"><td><p>stall_pipe_busy</p></td>
<td><p>smsp__warp_issue_stalled_math_pipe_throttle_per_warp_active.pct + smsp__warp_issue_stalled_mio_throttle_per_warp_active.pct</p></td>
</tr>
<tr class="row-even"><td><p>stall_sleeping</p></td>
<td><p>smsp__warp_issue_stalled_sleeping_per_warp_active.pct</p></td>
</tr>
<tr class="row-odd"><td><p>stall_sync</p></td>
<td><p>smsp__warp_issue_stalled_barrier_per_warp_active.pct + smsp__warp_issue_stalled_membar_per_warp_active.pct</p></td>
</tr>
<tr class="row-even"><td><p>stall_texture</p></td>
<td><p>smsp__warp_issue_stalled_tex_throttle_per_warp_active.pct</p></td>
</tr>
<tr class="row-odd"><td><p>surface_atomic_requests</p></td>
<td><p>l1tex__t_requests_pipe_tex_mem_surface_op_atom.sum</p></td>
</tr>
<tr class="row-even"><td><p>surface_load_requests</p></td>
<td><p>l1tex__t_requests_pipe_tex_mem_surface_op_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>surface_reduction_requests</p></td>
<td><p>l1tex__t_requests_pipe_tex_mem_surface_op_red.sum</p></td>
</tr>
<tr class="row-even"><td><p>surface_store_requests</p></td>
<td><p>l1tex__t_requests_pipe_tex_mem_surface_op_st.sum</p></td>
</tr>
<tr class="row-odd"><td><p>sysmem_read_bytes</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_read * 32</p></td>
</tr>
<tr class="row-even"><td><p>sysmem_read_throughput</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_read.sum.per_second</p></td>
</tr>
<tr class="row-odd"><td><p>sysmem_read_transactions</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_read.sum</p></td>
</tr>
<tr class="row-even"><td><p>sysmem_read_utilization</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>sysmem_utilization</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>sysmem_write_bytes</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_write * 32</p></td>
</tr>
<tr class="row-odd"><td><p>sysmem_write_throughput</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_write.sum.per_second</p></td>
</tr>
<tr class="row-even"><td><p>sysmem_write_transactions</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_write.sum</p></td>
</tr>
<tr class="row-odd"><td><p>sysmem_write_utilization</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>tensor_precision_fu_utilization</p></td>
<td><p>sm__pipe_tensor_op_hmma_cycles_active.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-odd"><td><p>tensor_precision_int_utilization</p></td>
<td><p>sm__pipe_tensor_op_imma_cycles_active.avg.pct_of_peak_sustained_active (SM 7.2+)</p></td>
</tr>
<tr class="row-even"><td><p>tex_cache_hit_rate</p></td>
<td><p>l1tex__t_sector_hit_rate.pct</p></td>
</tr>
<tr class="row-odd"><td><p>tex_cache_throughput</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>tex_cache_transactions</p></td>
<td><p>l1tex__lsu_writeback_active.avg.pct_of_peak_sustained_active + l1tex__tex_writeback_active.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-odd"><td><p>tex_fu_utilization</p></td>
<td><p>smsp__inst_executed_pipe_tex.avg.pct_of_peak_sustained_active</p></td>
</tr>
<tr class="row-even"><td><p>tex_sm_tex_utilization</p></td>
<td><p>l1tex__f_tex2sm_cycles_active.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-odd"><td><p>tex_sm_utilization</p></td>
<td><p>sm__mio2rf_writeback_active.avg.pct_of_peak_sustained_elapsed</p></td>
</tr>
<tr class="row-even"><td><p>tex_utilization</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>texture_load_requests</p></td>
<td><p>l1tex__t_requests_pipe_tex_mem_texture.sum</p></td>
</tr>
<tr class="row-even"><td><p>warp_execution_efficiency</p></td>
<td><p>smsp__thread_inst_executed_per_inst_executed.ratio</p></td>
</tr>
<tr class="row-odd"><td><p>warp_nonpred_execution_efficiency</p></td>
<td><p>smsp__thread_inst_executed_per_inst_executed.pct</p></td>
</tr>
</tbody>
</table>
<dl class="footnote brackets">
<dt class="label" id="fn2"><span class="brackets">2</span><span class="fn-backref">(<a href="#id16">1</a>,<a href="#id17">2</a>)</span></dt>
<dd><p>Sector reads from reductions are added here only for compatibility to the current definition of the metric in nvprof. Reductions do not cause data to be communicated from L2 back to L1.</p>
</dd>
</dl>
</section>
<section id="event-comparison">
<span id="nvprof-event-comparison"></span><h3><span class="section-number">4.6.4. </span>Event Comparison<a class="headerlink" href="#event-comparison" title="Permalink to this headline"></a></h3>
<p>For nvprof events, the following table lists the equivalent metrics in NVIDIA Nsight Compute, if available. For a detailed explanation of the structuring of PerfWorks metrics, see <a class="reference external" href="../ProfilingGuide/index.html#metrics-structure">Metrics Structure</a>.</p>
<p>Metrics starting with <em>sm__</em> are collected per-SM. Metrics starting with <em>smsp__</em> are collected per-SM subpartition. However, all corresponding nvprof events are collected per-SM, only. Check the <a class="reference external" href="../ProfilingGuide/index.html#metrics-guide">Metrics Guide</a> for more details on these terms.</p>
<table class="table-no-stripes docutils align-default" id="id29">
<caption><span class="caption-text">Table 12. Events Mapping Table from CUPTI Events to PerfWorks Metrics for Compute Capability &gt;= 7.0</span><a class="headerlink" href="#id29" title="Permalink to this table"></a></caption>
<colgroup>
<col style="width: 32%" />
<col style="width: 68%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>nvprof Event</p></th>
<th class="head"><p>PerfWorks Metric or Formula (&gt;= SM 7.0)</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>active_cycles</p></td>
<td><p>sm__cycles_active.sum</p></td>
</tr>
<tr class="row-odd"><td><p>active_cycles_pm</p></td>
<td><p>sm__cycles_active.sum</p></td>
</tr>
<tr class="row-even"><td><p>active_cycles_sys</p></td>
<td><p>sys__cycles_active.sum</p></td>
</tr>
<tr class="row-odd"><td><p>active_warps</p></td>
<td><p>sm__warps_active.sum</p></td>
</tr>
<tr class="row-even"><td><p>active_warps_pm</p></td>
<td><p>sm__warps_active.sum</p></td>
</tr>
<tr class="row-odd"><td><p>atom_count</p></td>
<td><p>smsp__inst_executed_op_generic_atom_dot_alu.sum</p></td>
</tr>
<tr class="row-even"><td><p>elapsed_cycles_pm</p></td>
<td><p>sm__cycles_elapsed.sum</p></td>
</tr>
<tr class="row-odd"><td><p>elapsed_cycles_sm</p></td>
<td><p>sm__cycles_elapsed.sum</p></td>
</tr>
<tr class="row-even"><td><p>elapsed_cycles_sys</p></td>
<td><p>sys__cycles_elapsed.sum</p></td>
</tr>
<tr class="row-odd"><td><p>fb_subp0_read_sectors</p></td>
<td><p>dram__sectors_read.sum</p></td>
</tr>
<tr class="row-even"><td><p>fb_subp1_read_sectors</p></td>
<td><p>dram__sectors_read.sum</p></td>
</tr>
<tr class="row-odd"><td><p>fb_subp0_write_sectors</p></td>
<td><p>dram__sectors_write.sum</p></td>
</tr>
<tr class="row-even"><td><p>fb_subp1_write_sectors</p></td>
<td><p>dram__sectors_write.sum</p></td>
</tr>
<tr class="row-odd"><td><p>global_atom_cas</p></td>
<td><p>smsp__inst_executed_op_generic_atom_dot_cas.sum</p></td>
</tr>
<tr class="row-even"><td><p>gred_count</p></td>
<td><p>smsp__inst_executed_op_global_red.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed</p></td>
<td><p>sm__inst_executed.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_fma_pipe_s0</p></td>
<td><p>smsp__inst_executed_pipe_fma.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_fma_pipe_s1</p></td>
<td><p>smsp__inst_executed_pipe_fma.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_fma_pipe_s2</p></td>
<td><p>smsp__inst_executed_pipe_fma.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_fma_pipe_s3</p></td>
<td><p>smsp__inst_executed_pipe_fma.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_fp16_pipe_s0</p></td>
<td><p>smsp__inst_executed_pipe_fp16.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_fp16_pipe_s1</p></td>
<td><p>smsp__inst_executed_pipe_fp16.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_fp16_pipe_s2</p></td>
<td><p>smsp__inst_executed_pipe_fp16.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_fp16_pipe_s3</p></td>
<td><p>smsp__inst_executed_pipe_fp16.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_fp64_pipe_s0</p></td>
<td><p>smsp__inst_executed_pipe_fp64.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_fp64_pipe_s1</p></td>
<td><p>smsp__inst_executed_pipe_fp64.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_executed_fp64_pipe_s2</p></td>
<td><p>smsp__inst_executed_pipe_fp64.sum</p></td>
</tr>
<tr class="row-odd"><td><p>inst_executed_fp64_pipe_s3</p></td>
<td><p>smsp__inst_executed_pipe_fp64.sum</p></td>
</tr>
<tr class="row-even"><td><p>inst_issued1</p></td>
<td><p>sm__inst_issued.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_read_sector_misses</p></td>
<td><p>lts__t_sectors_op_read_lookup_miss.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_read_sector_misses</p></td>
<td><p>lts__t_sectors_op_read_lookup_miss.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_read_sysmem_sector_queries</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_read.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_read_sysmem_sector_queries</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_read.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_read_tex_hit_sectors</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_read_lookup_hit.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_read_tex_hit_sectors</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_read_lookup_hit.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_read_tex_sector_queries</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_read.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_read_tex_sector_queries</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_read.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_total_read_sector_queries</p></td>
<td><p>lts__t_sectors_op_read.sum + lts__t_sectors_op_atom.sum + lts__t_sectors_op_red.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_total_read_sector_queries</p></td>
<td><p>lts__t_sectors_op_read.sum + lts__t_sectors_op_atom.sum + lts__t_sectors_op_red.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_total_write_sector_queries</p></td>
<td><p>lts__t_sectors_op_write.sum + lts__t_sectors_op_atom.sum + lts__t_sectors_op_red.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_total_write_sector_queries</p></td>
<td><p>lts__t_sectors_op_write.sum + lts__t_sectors_op_atom.sum + lts__t_sectors_op_red.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_write_sector_misses</p></td>
<td><p>lts__t_sectors_op_write_lookup_miss.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_write_sector_misses</p></td>
<td><p>lts__t_sectors_op_write_lookup_miss.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_write_sysmem_sector_queries</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_write.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_write_sysmem_sector_queries</p></td>
<td><p>lts__t_sectors_aperture_sysmem_op_write.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_write_tex_hit_sectors</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_write_lookup_hit.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_write_tex_hit_sectors</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_write_lookup_hit.sum</p></td>
</tr>
<tr class="row-odd"><td><p>l2_subp0_write_tex_sector_queries</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_write.sum</p></td>
</tr>
<tr class="row-even"><td><p>l2_subp1_write_tex_sector_queries</p></td>
<td><p>lts__t_sectors_srcunit_tex_op_write.sum</p></td>
</tr>
<tr class="row-odd"><td><p>not_predicated_off_thread_inst_executed</p></td>
<td><p>smsp__thread_inst_executed_pred_on.sum</p></td>
</tr>
<tr class="row-even"><td><p>pcie_rx_active_pulse</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>pcie_tx_active_pulse</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>prof_trigger_00</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>prof_trigger_01</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>prof_trigger_02</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>prof_trigger_03</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>prof_trigger_04</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>prof_trigger_05</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>prof_trigger_06</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-odd"><td><p>prof_trigger_07</p></td>
<td><p>n/a</p></td>
</tr>
<tr class="row-even"><td><p>inst_issued0</p></td>
<td><p>smsp__issue_inst0.sum</p></td>
</tr>
<tr class="row-odd"><td><p>sm_cta_launched</p></td>
<td><p>sm__ctas_launched.sum</p></td>
</tr>
<tr class="row-even"><td><p>shared_load</p></td>
<td><p>smsp__inst_executed_op_shared_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>shared_store</p></td>
<td><p>smsp__inst_executed_op_shared_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>generic_load</p></td>
<td><p>smsp__inst_executed_op_generic_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>generic_store</p></td>
<td><p>smsp__inst_executed_op_generic_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>global_load</p></td>
<td><p>smsp__inst_executed_op_global_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>global_store</p></td>
<td><p>smsp__inst_executed_op_global_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>local_load</p></td>
<td><p>smsp__inst_executed_op_local_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>local_store</p></td>
<td><p>smsp__inst_executed_op_local_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>shared_atom</p></td>
<td><p>smsp__inst_executed_op_shared_atom.sum</p></td>
</tr>
<tr class="row-odd"><td><p>shared_atom_cas</p></td>
<td><p>smsp__inst_executed_op_shared_atom_dot_cas.sum</p></td>
</tr>
<tr class="row-even"><td><p>shared_ld_bank_conflict</p></td>
<td><p>l1tex__data_bank_conflicts_pipe_lsu_mem_shared_op_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>shared_st_bank_conflict</p></td>
<td><p>l1tex__data_bank_conflicts_pipe_lsu_mem_shared_op_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>shared_ld_transactions</p></td>
<td><p>l1tex__data_pipe_lsu_wavefronts_mem_shared_op_ld.sum</p></td>
</tr>
<tr class="row-odd"><td><p>shared_st_transactions</p></td>
<td><p>l1tex__data_pipe_lsu_wavefronts_mem_shared_op_st.sum</p></td>
</tr>
<tr class="row-even"><td><p>tensor_pipe_active_cycles_s0</p></td>
<td><p>smsp__pipe_tensor_cycles_active.sum</p></td>
</tr>
<tr class="row-odd"><td><p>tensor_pipe_active_cycles_s1</p></td>
<td><p>smsp__pipe_tensor_cycles_active.sum</p></td>
</tr>
<tr class="row-even"><td><p>tensor_pipe_active_cycles_s2</p></td>
<td><p>smsp__pipe_tensor_cycles_active.sum</p></td>
</tr>
<tr class="row-odd"><td><p>tensor_pipe_active_cycles_s3</p></td>
<td><p>smsp__pipe_tensor_cycles_active.sum</p></td>
</tr>
<tr class="row-even"><td><p>thread_inst_executed</p></td>
<td><p>smsp__thread_inst_executed.sum</p></td>
</tr>
<tr class="row-odd"><td><p>warps_launched</p></td>
<td><p>smsp__warps_launched.sum</p></td>
</tr>
</tbody>
</table>
</section>
<section id="filtering">
<span id="nvprof-filtering"></span><h3><span class="section-number">4.6.5. </span>Filtering<a class="headerlink" href="#filtering" title="Permalink to this headline"></a></h3>
<ul>
<li><p><strong>Filtering by kernel name</strong></p>
<p>Both nvprof and NVIDIA Nsight Compute CLI support filtering which kernels’ data should be collected. In nvprof, the option is <code class="docutils literal notranslate"><span class="pre">--kernels</span></code> and applies to following metric collection options. In NVIDIA Nsight Compute CLI, the option is named <code class="docutils literal notranslate"><span class="pre">--kernel-regex</span></code> and applies to the complete application execution. In other words, NVIDIA Nsight Compute CLI does not currently support collecting different metrics for different kernels, unless they execute on different GPU architectures.</p>
</li>
<li><p><strong>Filtering by kernel ID</strong></p>
<p>Nvprof allows users to specify which kernels to profile using a kernel ID description, using the same <code class="docutils literal notranslate"><span class="pre">--kernels</span></code> option. In NVIDIA Nsight Compute CLI, the syntax for this kernel ID is identical, but the option is named <code class="docutils literal notranslate"><span class="pre">--kernel-id</span></code>.</p>
</li>
<li><p><strong>Filtering by device</strong></p>
<p>Both nvprof and NVIDIA Nsight Compute CLI use <code class="docutils literal notranslate"><span class="pre">--devices</span></code> to filter the devices which to profile. In contrast to nvprof, in NVIDIA Nsight Compute CLI the option applies globally, not only to following options.</p>
</li>
</ul>
</section>
<section id="output">
<span id="nvprof-output"></span><h3><span class="section-number">4.6.6. </span>Output<a class="headerlink" href="#output" title="Permalink to this headline"></a></h3>
<ul>
<li><p><strong>API trace and summary</strong></p>
<p>NVIDIA Nsight Compute CLI does not support any form of API-usage related output. No API data is captured during profiling.</p>
</li>
<li><p><strong>Dependency analysis</strong></p>
<p>NVIDIA Nsight Compute CLI does not support any dependency analysis. No API data is captured during profiling.</p>
</li>
<li><p><strong>GPU trace</strong></p>
<p>NVIDIA Nsight Compute CLI does not support any GPU trace output. Due to kernel replay during profiling, kernel executions are serialized, and start and end timestamps do not necessarily match those during application execution. In addition, no records for memory activities are recorded.</p>
</li>
<li><p><strong>Print summary</strong></p>
<p>While nvprof has several command line options to specify which summary information to print, NVIDIA Nsight Compute CLI uses further arguments to the <code class="docutils literal notranslate"><span class="pre">--print-summary</span></code> options. Profiling data can be summarized <code class="docutils literal notranslate"><span class="pre">per-gpu</span></code>, <code class="docutils literal notranslate"><span class="pre">per-kernel</span></code> or <code class="docutils literal notranslate"><span class="pre">per-nvtx</span></code> context.</p>
</li>
<li><p><strong>Kernel name demangling</strong></p>
<p>Nvprof allows users to decide between name demangling on or off using the <code class="docutils literal notranslate"><span class="pre">--demangling</span></code> options. NVIDIA Nsight Compute CLI currently always demangles kernel names in the output. In addition, the option <code class="docutils literal notranslate"><span class="pre">--kernel-regex-base</span></code> can be used to decide which name format should be used when matching kernel names during filtering.</p>
</li>
<li><p><strong>Pages</strong></p>
<p>Nvprof has no concept of output pages, all data is shown as a list or summarized. NVIDIA Nsight Compute CLI uses <em>pages</em> to define how data should be structured and printed. Those correspond to the report pages used in the GUI variant. The option <code class="docutils literal notranslate"><span class="pre">--page</span></code> can be used to select which page to show, and <code class="docutils literal notranslate"><span class="pre">details</span></code> is selected by default. All pages also support printing in CSV format for easier post-processing, using the <code class="docutils literal notranslate"><span class="pre">--csv</span></code> option.</p>
</li>
</ul>
</section>
<section id="launch-and-attach">
<span id="nvprof-launch-attach"></span><h3><span class="section-number">4.6.7. </span>Launch and Attach<a class="headerlink" href="#launch-and-attach" title="Permalink to this headline"></a></h3>
<ul>
<li><p><strong>Launching a process for profiling</strong></p>
<p>In nvprof, the application to profile is passed to the tool as a command line argument. The application must be a local executable. Alternatively, you can choose to use the tool in a <em>daemon mode</em> and profile all applicable processes on the local machine (nvprof option <code class="docutils literal notranslate"><span class="pre">--profile-all-processes</span></code>). In nvprof, the decision to profile the complete process tree or only the root process is done via the <code class="docutils literal notranslate"><span class="pre">--profile-child-processes</span></code> flag. In NVIDIA Nsight Compute CLI, the <code class="docutils literal notranslate"><span class="pre">--target-processes</span></code> option is used for this.</p>
<p>NVIDIA Nsight Compute CLI has several modes to determine which application to collect data for. By default, the executable passed via the command line to the tool is started, connected to, and profiled. This mode is called <code class="docutils literal notranslate"><span class="pre">launch-and-attach</span></code>.</p>
</li>
<li><p><strong>Launching a process for attach</strong></p>
<p>In contrast to nvprof, you can choose to only launch a local executable. In this mode (<code class="docutils literal notranslate"><span class="pre">--mode</span> <span class="pre">launch</span></code>), the process is started, connected to, but then suspended at the first CUDA API call. Subsequently, there is a third mode (<code class="docutils literal notranslate"><span class="pre">--mode</span> <span class="pre">attach</span></code>) to attach to any process launched using the aforementioned mode. In this case, all profiling and output options would be passed to the attaching instance of NVIDIA Nsight Compute CLI.</p>
</li>
<li><p><strong>Remote profiling</strong></p>
<p>Finally, using <code class="docutils literal notranslate"><span class="pre">launch</span></code> and <code class="docutils literal notranslate"><span class="pre">attach</span></code>, you can connect to a launched process on a remote machine, which could even run a different operating system than the local host. Use <code class="docutils literal notranslate"><span class="pre">--hostname</span></code> to select which remote host to connect to.</p>
</li>
</ul>
<p class="rubric-h1 rubric">Notices</p>
<p class="rubric-h2 rubric">Notices</p>
<p>ALL NVIDIA DESIGN SPECIFICATIONS, REFERENCE BOARDS, FILES, DRAWINGS, DIAGNOSTICS, LISTS, AND OTHER DOCUMENTS (TOGETHER AND SEPARATELY, “MATERIALS”) ARE BEING PROVIDED “AS IS.” NVIDIA MAKES NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT, MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.</p>
<p>Information furnished is believed to be accurate and reliable. However, NVIDIA Corporation assumes no responsibility for the consequences of use of such information or for any infringement of patents or other rights of third parties that may result from its use. No license is granted by implication of otherwise under any patent rights of NVIDIA Corporation. Specifications mentioned in this publication are subject to change without notice. This publication supersedes and replaces all other information previously supplied. NVIDIA Corporation products are not authorized as critical components in life support devices or systems without express written approval of NVIDIA Corporation.</p>
<p class="rubric-h2 rubric">Trademarks</p>
<p>NVIDIA and the NVIDIA logo are trademarks or registered trademarks of NVIDIA Corporation in the U.S. and other countries. Other company and product names may be trademarks of the respective companies with which they are associated.</p>
</section>
</section>
</section>


           </div>
          </div>
          <footer>

  <hr/>

  <div role="contentinfo">
    <p>&#169; Copyright 2018-2024, NVIDIA Corporation &amp; Affiliates. All rights reserved.
      <span class="lastupdated">Last updated on Mar 06, 2024.
      </span></p>
  </div>

   

</footer>
        </div>
      </div>
    </section>
  </div>
  <script>
      jQuery(function () {
          SphinxRtdTheme.Navigation.enable(true);
      });
  </script>
 



</body>
</html>