[release/1.7] maximum number of threads per block for sm_86 is 1536 #45924
Conversation
Summary: according to https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#features-and-technical-specifications Pull Request resolved: pytorch#45889 Reviewed By: albanD Differential Revision: D24131188 Pulled By: ngimel fbshipit-source-id: 31d3038f7b1bc403751448c62b19609573c67a49
|
tbh, I don't see how it can improve 3080/3090 performance unless we compile for 8.6 cc (which I don't think we are going to do, relying on 8.0 binary compatibility, cc @malfet). |
albanD
added
the
triaged
|
sm_86 is enabled in cuda 11.1, so it depends on which cuda version will be compiled for the release. |
Codecov Report
@@ Coverage Diff @@
## release/1.7 #45924 +/- ##
============================================
Coverage 68.38% 68.38%
============================================
Files 406 406
Lines 52267 52267
============================================
+ Hits 35741 35742 +1
+ Misses 16526 16525 -1
Continue to review full report at Codecov.
|
|
Hi @xwang233! Thank you for your pull request. We require contributors to sign our Contributor License Agreement, and yours needs attention. You currently have a record in our system, but we do not have a signature on file. In order for us to review and merge your code, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA. If you have received this in error or have any questions, please contact us at cla@fb.com. Thanks! |
|
Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Facebook open source project. Thanks! |
cherry pick of #45889
This may improve the performance on Ampere GeForce 3080 and 3090.