Skip to content

examples/64: relax exact CC 8.9 gate to inclusive check - #3586

Open
SAR2652 wants to merge 1 commit into
NVIDIA:mainfrom
SAR2652:ada-fp8-arch-fix
Open

examples/64: relax exact CC 8.9 gate to inclusive check#3586
SAR2652 wants to merge 1 commit into
NVIDIA:mainfrom
SAR2652:ada-fp8-arch-fix

Conversation

@SAR2652

@SAR2652 SAR2652 commented Sep 5, 2026

Copy link
Copy Markdown

Fixes #3524

Example 64 gated on compute capability exactly 8.9, rejecting newer
architectures despite the kernels compiling and running correctly. This
uses the inclusive form already present in example 58
(major < 8 || (major == 8 && minor < 9)), and updates the error
message to match.

Verified on sm_120a, CUDA 13.0, CUTLASS 4.8.0:

cmake .. -DCUTLASS_NVCC_ARCHS=120a
make 64_ada_fp8_gemm_grouped

Grouped GEMM (CUTLASS) with mode kDeviceOnly:
    9300 total threadblock tiles.
    Grouped Runtime: 6.32188 ms
    Grouped  GFLOPs: 54070.7
Passed

Signed-off-by: Sarvesh Relekar <sar265278@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Example 64 (Ada FP8 grouped GEMM) refuses capable GPUs via exact CC 8.9 gate

1 participant