File: cudnn_rnn_determinism.rst

package info (click to toggle)
pytorch-cuda 2.6.0%2Bdfsg-7
  • links: PTS, VCS
  • area: contrib
  • in suites: forky, sid, trixie
  • size: 161,620 kB
  • sloc: python: 1,278,832; cpp: 900,322; ansic: 82,710; asm: 7,754; java: 3,363; sh: 2,811; javascript: 2,443; makefile: 597; ruby: 195; xml: 84; objc: 68
file content (16 lines) | stat: -rw-r--r-- 641 bytes parent folder | download | duplicates (4)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
.. warning::
    There are known non-determinism issues for RNN functions on some versions of cuDNN and CUDA.
    You can enforce deterministic behavior by setting the following environment variables:

    On CUDA 10.1, set environment variable ``CUDA_LAUNCH_BLOCKING=1``.
    This may affect performance.

    On CUDA 10.2 or later, set environment variable
    (note the leading colon symbol)
    ``CUBLAS_WORKSPACE_CONFIG=:16:8``
    or
    ``CUBLAS_WORKSPACE_CONFIG=:4096:2``.

    See the `cuDNN 8 Release Notes`_ for more information.

.. _cuDNN 8 Release Notes: https://docs.nvidia.com/deeplearning/sdk/cudnn-release-notes/rel_8.html