[QST] Build for sm100 Blackwell GPUs

[phantaurus]

What is your question?
Hello!

I noticed this:
"Note: The NVIDIA Blackwell SM100 architecture used in the datacenter products has a different compute capability than the one underpinning NVIDIA Blackwell GeForce RTX 50 series GPUs. As a result, kernels compiled for Blackwell SM100 architecture with arch conditional features (using sm100a) are not compatible with RTX 50 series GPUs."

When building cutlass examples, I tried both DCUTLASS_NVCC_ARCHS="100a" and DCUTLASS_NVCC_ARCHS="100".
When setting it to "100", examples such as 70_blackwell_gemm disappeared from the Makefile.

Does this mean that non-datacenter sm_100 Blackwell GPUs do not have the new TensorCore features? If so, do they fall back to Hopper? Can I use the hopper tensorcore examples to get max TFLOPS on sm_100 GPUs? Or does this mean cutlass currently only support sm_100a TensorCore operations?

Thank you so much!

[thakkarV]

CUTLASS 3.8 only supports sm100a aka datacenter Blackwell chips for now. Support for sm120a will land in a future release.

[phantaurus]

@thakkarV Thank you so much for your reply! What about DRIVE Thor which has a compute capability of 10.1? Will they be supported along with sm120a?

[hwu36]

What about DRIVE Thor which has a compute capability of 10.1? Will they be supported along with sm120a?

No, sm101a cannot run on sm120a.

[thakkarV]

All kernels we have released in 3.8 can be run on 100 or 101. Just compile with the right CUTLASS_NVCC_ARCH set. SM120a support will come in a future release

[leimingyu]

RTX5080 and RTX 5090 is sm120a ? what is the expected release date?

[huyi1102]

All kernels we have released in 3.8 can be run on 100 or 101. Just compile with the right CUTLASS_NVCC_ARCH set. SM120a support will come in a future release

Hi, can Cutlass 3.8 run on Drive Thor? If so, how can I set it up? I compiled GEMM-related kernels but encountered an error during execution: "Failed to query occupancy". Here is my execution process:
cmake .. -DCUTLASS_NVCC_ARCHS=100a -DCUTLASS_ENABLE_CUBLAS=OFF -DCUTLASS_ENABLE_CUDNN=OFF -DCUTLASS_ENABLE_TESTS=OFF -DCUTLASS_UNITY_BUILD_ENABLED=ON -DCUTLASS_LIBRARY_KERNELS=tensorop*gemm_f16
./tools/profiler/cutlass_profiler --kernels=gemm

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[github-actions]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.

[vgoklani]