[GH-ISSUE #392] Will DeepEP work on GB200 NVL72? #267

Open
opened 2026-10-26 09:57:39 +00:00 by bsakaguchi · 1 comment
Owner

Originally created by @pbelevich on GitHub (Sep 3, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/392

on 72-GPU NVLink domain? And should it be used without IB?

cc @amanshanbhag

Originally created by @pbelevich on GitHub (Sep 3, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/392 on 72-GPU NVLink domain? And should it be used without IB? cc @amanshanbhag
Author
Owner

@sphish commented on GitHub (Sep 5, 2025):

The current low-latency kernels supports GB200 and you can refer to the usage in SGLang.
Since the current normal kernels limits the number of GPUs per node to 8, the normal mode doesn't support GB200 yet. We will remove this limitation in a future refactor.

<!-- gh-comment-id:3257114074 --> @sphish commented on GitHub (Sep 5, 2025): The current low-latency kernels supports GB200 and you can refer to the usage in SGLang. Since the current normal kernels limits the number of GPUs per node to 8, the normal mode doesn't support GB200 yet. We will remove this limitation in a future refactor.
Sign in to join this conversation.
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepEP#267
No description provided.