Skip to content

[Bugfix] Exclude SM107 (Rubin) from support_deep_gemm() #58022

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
meena-at-work wants to merge 1 commit into from
Closed

[Bugfix] Exclude SM107 (Rubin) from support_deep_gemm() #58022

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 9 additions & 2 deletions vllm/platforms/cuda.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -717,10 +717,17 @@ def support_static_graph_mode(cls) -> bool:

@classmethod
def support_deep_gemm(cls) -> bool:
"""Currently, only Hopper and Blackwell GPUs are supported."""
"""Currently, only Hopper and Blackwell GPUs are supported.

Within the SM100 family, only SM100 and SM103 are supported: the
vendored DeepGEMM/CuTe kernels are not built for native sm_107, so
`is_device_capability_family(100)` (which matches any 10.x) would
wrongly admit it and crash with CUDA_ERROR_ASSERT at kernel load.
"""
return (
cls.is_device_capability(90)
or cls.is_device_capability_family(100)
or cls.is_device_capability(100)
or cls.is_device_capability(103)
or cls.is_device_capability_family(120)
)

Expand Down
Loading