CUDA in your pipeline
GPU runners
NVIDIA L4 and A10G runners with CUDA preinstalled, for model evaluation and kernel compilation in CI.
L4 and A10G
24 GB of VRAM, enough for evaluation suites and mid-size inference tests.
Drivers ready
CUDA 12.6 and cuDNN 9 already installed and verified.
Per-minute
No reserved capacity commitment.
GitHub does not sell GPU runners, so GPU code paths usually go untested until deployment. Per-minute GPU CI closes that gap without a standing cluster.
Common questions
- Why would I need a GPU in CI?
- So the first time a CUDA kernel runs is not in production. GPU runners let you compile kernels, run model evaluation on pull requests and verify inference latency budgets before merge.
- Which GPUs are available?
- NVIDIA L4 and A10G, both 24 GB, with CUDA 12.6 and cuDNN 9 preinstalled and verified.
- Does GitHub offer GPU runners?
- No. This is a capability with no GitHub-hosted equivalent, which is why teams either self-host or skip GPU tests entirely.
Your next build could be twice as fast, at half the price
Start free. Migrating away is the same one line, and we publish that diff too.