Skip to content

build(infraCommons): bump for the fat CUDA architecture list - #71

Merged
ajakhotia merged 1 commit into
mainfrom
ajakhotia/cuda-arch-fat
Aug 23, 2026
Merged

ajakhotia merged 1 commit into
mainfrom
ajakhotia/cuda-arch-fat

Conversation

@ajakhotia

Copy link
Copy Markdown
Owner

Bumps infraCommons to ajakhotia/infraCommons#55: the toolchains now compile CUDA for every architecture CUDA 13.3 supports (sm_75 through sm_121, spelled out because OpenCV rejects CMake's special arch values) instead of the 75;80 pair, so newer GPUs get native SASS rather than PTX JIT. Validated by a full local lane (24.04 clang-shared) building and installing cleanly; fat per-TU objects keep ccache growth negligible (~0.3 GiB).

Also removes the caching commentary block above the workflow's docker run, which restated what its env block already shows. Closes #70.

The toolchains now compile CUDA for every architecture the toolkit supports instead of the
75;80 pair, so newer GPUs get native SASS rather than PTX JIT. Also drops the caching
commentary block from the workflow, which restated its env block (closes #70).
@ajakhotia
ajakhotia merged commit 9337595 into main Aug 23, 2026
43 checks passed
@ajakhotia
ajakhotia deleted the ajakhotia/cuda-arch-fat branch August 23, 2026 11:05
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.

Remove the caching commentary block from docker-image.yaml

1 participant