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 #9] Is IBGDA necessary? #9

New issue
Closed
opened 2026-10-26 09:52:18 +00:00 by bsakaguchi · 5 comments
bsakaguchi commented 2026-10-26 09:52:18 +00:00
Owner
Copy link

Originally created by @Baibaifan on GitHub (Feb 25, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/9

hi, team

I would like to ask, is IBGDA(InfiniBand GPUDirect Async) necessary? Can the RDMA network of RoCEv2 use DeepEp normally? When installing NVSHMEM, I found that IBGDA needs to be configured.

Originally created by @Baibaifan on GitHub (Feb 25, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/9 hi, team I would like to ask, is IBGDA(InfiniBand GPUDirect Async) necessary? Can the RDMA network of RoCEv2 use DeepEp normally? When installing NVSHMEM, I found that IBGDA needs to be configured.
bsakaguchi commented 2026-10-26 09:52:20 +00:00
Author
Owner
Copy link

@wx-csy commented on GitHub (Feb 25, 2025):

IBGDA is necessary and crucial to run DeepEP. RoCE network should work, too.

<!-- gh-comment-id:2681038328 --> @wx-csy commented on GitHub (Feb 25, 2025): IBGDA is necessary and crucial to run DeepEP. RoCE network should work, too.
bsakaguchi commented 2026-10-26 09:52:20 +00:00
Author
Owner
Copy link

@Baibaifan commented on GitHub (Feb 25, 2025):

I didn't see any information about how to configure the RoCE network when installing NVSHMEM. I would like to ask if deepseek has tried this? @wx-csy

For example:

Image
<!-- gh-comment-id:2681051164 --> @Baibaifan commented on GitHub (Feb 25, 2025): I didn't see any information about how to configure the RoCE network when installing NVSHMEM. I would like to ask if deepseek has tried this? @wx-csy For example: <img width="1135" alt="Image" src="https://github.com/user-attachments/assets/64ef6500-69ef-4979-bb20-d12140545328" />
bsakaguchi commented 2026-10-26 09:52:20 +00:00
Author
Owner
Copy link

@wx-csy commented on GitHub (Feb 25, 2025):

I'm not aware whether deepseek has tried RoCE. However, the configuration and installation of NVSHMEM for RoCE should be no different. Just enable IBGDA normally.

I guess you have probably been confused by the IB in IBGDA. The IB refers to the Infiniband RDMA transport layer defined by IBTA, which is used by both Infiniband and RoCE.

<!-- gh-comment-id:2681064990 --> @wx-csy commented on GitHub (Feb 25, 2025): I'm not aware whether deepseek has tried RoCE. However, the configuration and installation of NVSHMEM for RoCE should be no different. Just enable IBGDA normally. I guess you have probably been confused by the `IB` in IBGDA. The IB refers to the Infiniband RDMA transport layer defined by IBTA, which is used by both Infiniband and RoCE.
bsakaguchi commented 2026-10-26 09:52:20 +00:00
Author
Owner
Copy link

@Baibaifan commented on GitHub (Feb 25, 2025):

Thanks, I'll try it. If possible, I still hope DeepSeek can provide a test process for the RoCE network. I think most developers will use the RoCE network. @wx-csy

<!-- gh-comment-id:2681092575 --> @Baibaifan commented on GitHub (Feb 25, 2025): Thanks, I'll try it. If possible, I still hope DeepSeek can provide a test process for the RoCE network. I think most developers will use the RoCE network. @wx-csy
bsakaguchi commented 2026-10-26 09:52:20 +00:00
Author
Owner
Copy link

@sphish commented on GitHub (Feb 25, 2025):

Agree with @wx-csy.

<!-- gh-comment-id:2681208931 --> @sphish commented on GitHub (Feb 25, 2025): Agree with @wx-csy.
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#9
No description provided.