Skip to content

Conversation

@limou102
Copy link
Contributor

This test simulates P2P transmission between nodes in pipeline parallelism without any intra-node transmission.

The test connects GPUs with the same local_rank from each node in a ring topology.

for example (num_nodes=4, num_gpus_per_node=8, ringsize=4):

ring0 : gpu0-->gpu8-->gpu16-->gpu24-->gpu0
ring1 : gpu1-->gpu9-->gpu17-->gpu25-->gpu1
...
ring7 : gpu7-->gpu15-->gpu23-->gpu31-->gpu7

Each ring operates independently.
Every GPU sends data to the next GPU in the ring while simultaneously receiving data from the previous GPU.

image

image

@lhzhang333
Copy link
Collaborator

LGTM

Copy link
Contributor

@wenxie-amd wenxie-amd left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@wenxie-amd wenxie-amd merged commit 1812c5d into main Apr 21, 2025
2 checks passed
@limou102 limou102 deleted the dev/limou/preflight_ring_p2p branch June 4, 2025 03:46
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.

5 participants