Skip to content

Comments

feat: add cuda-nvcc to runner image for CUDA kernel compilation#22

Closed
Gabby Romano (gromano-ltx) wants to merge 2 commits intomainfrom
feat/add-cuda-nvcc
Closed

feat: add cuda-nvcc to runner image for CUDA kernel compilation#22
Gabby Romano (gromano-ltx) wants to merge 2 commits intomainfrom
feat/add-cuda-nvcc

Conversation

@gromano-ltx
Copy link
Contributor

Summary

Adds cuda-nvcc-12-2 and cuda-cudart-dev-12-2 packages to enable
compilation of CUDA kernels on H100 runners.

Changes

  • Install cuda-nvcc-12-2 (CUDA compiler)
  • Install cuda-cudart-dev-12-2 (development headers)
  • Set PATH and LD_LIBRARY_PATH for CUDA 12.2

Why

The q8_kernels_ltx package requires nvcc to compile .cu files.
The current base image only has CUDA runtime, not the dev toolkit.

Impact

  • Image size: ~3.39 GB (+200-300 MB)
  • Affects 6 runners:
    • h100
    • h100-ml-engine
    • l4
    • comfy-h100
    • h100-8g
    • xora-core-h100

Testing

docker build -t cuda-arc-runner:test . --platform linux/amd64
docker run --rm cuda-arc-runner:test nvcc --version
# Cuda compilation tools, release 12.2, V12.2.140

Adds cuda-nvcc-12-2 and cuda-cudart-dev-12-2 packages to enable
compilation of CUDA kernels (e.g., q8_kernels_ltx) on H100 runners.

Also sets PATH and LD_LIBRARY_PATH for CUDA 12.2.
This reverts commit 8cdd7863d55ad0074cd5923827b59b8ce36d805d.
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.

2 participants