Page MenuHome

Cycles: Preliminary Cuda 11 build support
AbandonedPublic

Authored by Ray molenkamp (LazyDodo) on Wed, Jun 17, 11:00 PM.

Details

Summary

Basic cmake changes, there's nothing surprising in here.

Cuda 11 dropped sm_30 and added sm_80 the
appropriate kernels will be disabled depending
on the cuda version found.

Given there has been little to no testing of
cuda 11, I left in the warning about only
10.0 and 10.1 being supported.

Diff Detail

Repository
rB Blender
Branch
tmp_cuda11 (branched from master)
Build Status
Buildable 8651
Build 8651: arc lint + arc unit

Event Timeline

Ray molenkamp (LazyDodo) requested review of this revision.Wed, Jun 17, 11:00 PM
Ray molenkamp (LazyDodo) created this revision.
  • missed a change in kernel_config.h
This revision is now accepted and ready to land.Fri, Jun 19, 12:24 PM
  • Merge remote-tracking branch 'origin/master' into tmp_cuda11
  • fix optix still using the 30 target
Ray molenkamp (LazyDodo) requested review of this revision.Sat, Jun 20, 7:53 PM

During final testing bumped into some optix issues, it does not like sm_30 anymore, and anything below 52 gives deprecation warnings, given we only support optix on the 900+ cards i think 52 should be alright?

@Patrick Mours (pmoursnv) / @Brecht Van Lommel (brecht) do either of you have a strong opinions here?

I think we should bump the OptiX target to at least sm_50, if not sm_52 (for the purpose of CUDA 11) in general. It cannot run on earlier cards anyway and having it at sm_30 misses some (albeit small) optimization potential (since most optimization is handled in the PTX->binary driver step). I wanted to do that a while ago, but seamingly forgot about it.

Changing to sm_52 is fine with me.