[GH-ISSUE #406] Internode comm hangs on B200 #276

New issue

Open

opened 2026-10-26 09:57:49 +00:00 by bsakaguchi · 2 comments

bsakaguchi commented

2026-10-26 09:57:49 +00:00

Owner

Originally created by @cryoco on GitHub (Sep 15, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/406

The DeepEP version is a84a248(April version) due to lacking condition to enable IBGDA(requesting rebooting).
When testing tests/test_internode.py on 2 or more nodes of B200, it hangs after the first test.

Is this version viable on B200? Or is there any modification needed?

Originally created by @cryoco on GitHub (Sep 15, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/406 The DeepEP version is a84a248(April version) due to lacking condition to enable IBGDA(requesting rebooting). When testing tests/test_internode.py on 2 or more nodes of B200, it hangs after the first test. <img width="935" height="57" alt="Image" src="https://github.com/user-attachments/assets/9453dae0-85ad-46d8-a474-fadddfb1f710" /> Is this version viable on B200? Or is there any modification needed?

bsakaguchi commented

2026-10-26 09:57:50 +00:00

Author

Owner

@sphish commented on GitHub (Sep 24, 2025):

I’m not sure whether the version you mentioned is compatible with the B200, but the latest release has been validated by the vllm/sglang community, you could give it a try.

@sphish commented on GitHub (Sep 24, 2025): I’m not sure whether the version you mentioned is compatible with the B200, but the latest release has been validated by the vllm/sglang community, you could give it a try.

bsakaguchi commented

2026-10-26 09:57:50 +00:00

Author

Owner

@cryoco commented on GitHub (Sep 24, 2025):

I’m not sure whether the version you mentioned is compatible with the B200, but the latest release has been validated by the vllm/sglang community, you could give it a try.

Is there a way to utilize the latest version(normal kernels require IBGDA) without rebooting the system? Our access to the cluster is limited.

@cryoco commented on GitHub (Sep 24, 2025): > I’m not sure whether the version you mentioned is compatible with the B200, but the latest release has been validated by the vllm/sglang community, you could give it a try. Is there a way to utilize the latest version(normal kernels require IBGDA) without rebooting the system? Our access to the cluster is limited.

bsakaguchi referenced this issue

2026-10-27 21:03:15 +00:00

[PR #276] [MERGED] Use TMA to optimize internode dispatch. #497