mirror of
https://github.com/deepseek-ai/DeepEP.git
synced 2026-11-05 00:47:32 +00:00
[GH-ISSUE #649] DeepEP v2 hybrid mode runs out of QPs with the recommended QPS configuration #384
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/DeepEP#384
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @Cinderella0709 on GitHub (May 27, 2026).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/649
Hi,
When running DeepEP v2 in hybrid mode with the recommended QPS configuration from the following code:
github.com/deepseek-ai/DeepEP@723716fc11/deep_ep/buffers/elastic.py (L206)we encounter the following error:
It seems that the process may be exhausting RDMA/NIC resources while creating QPs.
Could this issue be related to host-side RDMA driver or firmware configurations, such as BFREG/UAR limits or other mlx5 resource constraints?
Any suggestions or insights would be greatly appreciated. Thanks in advance!
@sphish commented on GitHub (Jul 7, 2026):
NCCL Gin created too many QPs, and most of them are unnecessary. NCCL team will fix this issue in the future.