[GH-ISSUE #90] Consultation on the maximum number of nodes supported by internode #83

Closed
opened 2026-10-26 09:53:58 +00:00 by bsakaguchi · 2 comments
Owner

Originally created by @yeliang2258 on GitHub (Mar 25, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/90

From the code, does DeepEP internode currently support a maximum of 32 nodes and up to 256 GPUs ?

Originally created by @yeliang2258 on GitHub (Mar 25, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/90 From the code, does DeepEP internode currently support a maximum of 32 nodes and up to 256 GPUs ?
Author
Owner

@LyricZhao commented on GitHub (Mar 26, 2025):

Normal kernels supports EP2/4/8/16/32/64/144/160.

No hard limit for low-latency kernels, but it may depends on the number of experts and SMs. Generally speaking, for V3/R1, low-latency kernels support up to EP320 (tested).

<!-- gh-comment-id:2753124722 --> @LyricZhao commented on GitHub (Mar 26, 2025): Normal kernels supports EP2/4/8/16/32/64/144/160. No hard limit for low-latency kernels, but it may depends on the number of experts and SMs. Generally speaking, for V3/R1, low-latency kernels support up to EP320 (tested).
Author
Owner

@yeliang2258 commented on GitHub (Mar 26, 2025):

Got it, thank you very much for your reply

<!-- gh-comment-id:2753329896 --> @yeliang2258 commented on GitHub (Mar 26, 2025): Got it, thank you very much for your reply
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#83
No description provided.