Skip to content

Add 2-GPU runners: arm64 l4×2 nightly + Windows amd64 special runners - #2090

Merged
kkraus14 merged 1 commit into
NVIDIA:mainfrom
leofang:ci-nightly-2gpu
May 15, 2026
Merged

Add 2-GPU runners: arm64 l4×2 nightly + Windows amd64 special runners#2090
kkraus14 merged 1 commit into
NVIDIA:mainfrom
leofang:ci-nightly-2gpu

Conversation

@leofang

Copy link
Copy Markdown
Member

Add 2-GPU test coverage on two new fronts:

arm64 l4×2 (nightly-only)

Per the runner team's request, arm64 2-GPU runners should only be used in nightly CI due to ARM64 L4 capacity concerns. These are added as nightly-standard entries in ci-nightly.yml, which downloads wheels from the latest successful main CI run and executes the standard test suite (pathfinder, bindings, core) — no rebuild needed.

  • Python 3.14 + 3.14t (free-threaded), CUDA 13.2.1, local CTK
  • Runs at 2:17 AM UTC daily alongside the existing PyTorch/numba-cuda nightly jobs

Windows amd64 2-GPU (PR CI)

Windows 2-GPU runners have no capacity restrictions, so these are added as special runners in the regular pull-request matrix (mirroring the existing Linux amd64 2-GPU special runners).

  • t4×2 (Python 3.14, CUDA 13.2.1, local CTK, TCC)
  • h100×2 (Python 3.14, CUDA 13.2.1, wheels, MCDM)

Other changes

  • Updated Windows test job name template to show GPU count (e.g. GPU t4 (x2) (TCC)), matching the Linux format.

Closes#276, closes#1700.

-- Leo's bot

arm64 l4×2 runners are restricted to nightly-only use per the runner
team (ARM64 L4 capacity concerns). Add them as nightly-standard entries
in ci-nightly.yml so they run the standard test suite against wheels
from the latest successful main CI run.
Windows amd64 2-GPU runners (t4×2 TCC, h100×2 MCDM) are added as
special runners in the regular PR CI matrix, mirroring the existing
Linux amd64 2-GPU special runners.
Also update the Windows test job name to show GPU count (x2) for
multi-GPU entries, matching the Linux job name format.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@copy-pr-bot

Copy link
Copy Markdown
Contributor

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@github-actionsgithub-actionsBot added the CI/CD CI/CD infrastructure label May 15, 2026
@leofang

Copy link
Copy Markdown
MemberAuthor

/ok to test e3b0c71

@leofangleofang self-assigned this May 15, 2026
@leofangleofang added the P0 High priority - Must do! label May 15, 2026
@leofangleofang added this to the cuda.core next milestone May 15, 2026
@github-actions

This comment has been minimized.

@leofang
leofang marked this pull request as ready for review May 15, 2026 04:30
@kkraus14
kkraus14 merged commit eac59b6 into NVIDIA:mainMay 15, 2026
102 checks passed
@github-actions

This comment has been minimized.

@leofang
leofang deleted the ci-nightly-2gpu branch May 15, 2026 14:51
@leofang

Copy link
Copy Markdown
MemberAuthor

arm64 l4×2 (nightly-only)

Verified by manually triggering the nightly CI: https://github.com/NVIDIA/cuda-python/actions/runs/25915904780

@github-actions

Copy link
Copy Markdown
Doc Preview CI
Preview removed because the pull request was closed or merged.

Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CI/CDCI/CD infrastructureP0High priority - Must do!

Projects

None yet

Development

Successfully merging this pull request may close these issues.

CI: Add Windows 2-GPU runners CI: Cover testing against linux-aarch64 + 2 GPUs

2 participants

@leofang@kkraus14