deepseek-ai/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP.git synced 2026-11-05 00:47:32 +00:00
Code Issues 308 Projects Packages Wiki Activity

[GH-ISSUE #394] Unable to saturate NVL BW w/ intranode on GB200 #265

New issue
Open
opened 2026-10-26 09:57:39 +00:00 by bsakaguchi · 1 comment
bsakaguchi commented 2026-10-26 09:57:39 +00:00
Owner
Copy link

Originally created by @goelayu on GitHub (Sep 5, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/394

Are there any configuration knobs that need to be tuned to enable full bandwidth on GB200 nodes (800Gb/s)?

On a single host, w/ the intrahost kernel I get:

  Best config BW (GB/s) Latency (us)
Dispatch SMs 24, NVL chunk 16 169 690
Combine SMs 24, NVL chunk 16 135 861
Originally created by @goelayu on GitHub (Sep 5, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/394 Are there any configuration knobs that need to be tuned to enable full bandwidth on GB200 nodes (800Gb/s)? On a single host, w/ the intrahost kernel I get:   | Best config | BW (GB/s) | Latency (us) -- | -- | -- | -- Dispatch | SMs 24, NVL chunk 16 | 169 | 690 Combine | SMs 24, NVL chunk 16 | 135 | 861
bsakaguchi commented 2026-10-26 09:57:40 +00:00
Author
Owner
Copy link

@sphish commented on GitHub (Sep 10, 2025):

Since we currently do not have GB200 hardware, we have not performed any optimizations for GB200.
Additionally, according to community feedback, the low-latency mode is more friendly to GB200 and easier to use.

<!-- gh-comment-id:3273856742 --> @sphish commented on GitHub (Sep 10, 2025): Since we currently do not have GB200 hardware, we have not performed any optimizations for GB200. Additionally, according to community feedback, the low-latency mode is more friendly to GB200 and easier to use.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
hybrid-ep
chore/remove-clang-format-yapf
v2.5
nvDev
try-fix-cu128
mori-ep
antgroup-opt
compat-old-torch
add-linter
eager-rdma
v1.2.1
Labels
Clear labels   
pull-request
Mirrored from GitHub Pull Request

No labels
pull-request
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
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#265
No description provided.