Skip to content

Support A3 Mega GCP clusters with GPUDirect-TCPXO #2469

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.

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

Merged
merged 19 commits into from
Apr 4, 2025

Conversation

r4victor
Copy link
Collaborator

@r4victor r4victor commented Apr 3, 2025

Closes #2447

The PR adds support for GPUDirect-TCPXO on GCP A3 Mega instances:

  • Adds extra_vpcs parameter to GCP config to configure VPC for GPU NICs.
  • Supports placement groups for GCP – fleets with placement: cluster now have instances in the same collocated placement.
  • Adds a Github Action to build a debian-based VM image with GPUDirect-TCPXO – this is now used for A3 Mega instances.
  • Adds A3 Mega specific commands to run on cloud-init that configure GPUDirect-TCPXO.
  • Adds an example on how to configure and use GCP A3 Mega with GPUDirect-TCPXO.

TODO:

  • Currently to use GPUDirect-TCPXO optimally, certain env variables need to be set up. An example shows how it can be done. The next step may be to configure all the env vars GPUDirect-TCPXO for A3 Mega automatically. Although, it needs to be researched if hardcoding them is a good idea.
  • Supporting GPUDirect-TCPX.

@r4victor r4victor requested a review from peterschmidt85 April 3, 2025 11:40
@r4victor r4victor force-pushed the issue_2447_gcp_networking branch from c123cac to 69dbb0f Compare April 4, 2025 09:44
@r4victor r4victor force-pushed the issue_2447_gcp_networking branch from 69dbb0f to d99e07c Compare April 4, 2025 09:56
@r4victor r4victor merged commit 5408601 into master Apr 4, 2025
24 checks passed
@r4victor r4victor deleted the issue_2447_gcp_networking branch April 4, 2025 11:50
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.

[Feature]: Support A3 Mega GCP clusters with GPUDirect-TCPXO
1 participant