deepseek-ai / deepseek-ai/DeepEP

[Question] about EPv2 hybrid mode get_theoretical_num_qps

Open
#631 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Cuda
Stars
10.1k
Forks
1.4k
Avg merge
2d 17h
Merged PRs (30d)
3

Description

hybrid mode set num_qps to num_sms * 16 + 1, But it seems only `kNumChannelsPerSM` were used and `kNumMaxChannelsPerSM` is 8, So is num_sms * 8 enough?
```
const auto [qp_idx, sharing_mode] = comm::get_qp_mode 0)>(
sm_idx, (warp_idx - kNumNotifyWarps) % kNumChannelsPerSM, warp_idx < kNumNotifyWarps);
const auto gin = handle::NCCLGin(nccl_dev_comm, nccl_window, qp_idx, sharing_mode);
```

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by tracing get_theoretical_num_qps and the hybrid-mode path that computes num_qps, then compare it with get_qp_mode and NCCLGin in the shown snippet. Determine whether kNumChannelsPerSM or kNumMaxChannelsPerSM defines the required queue-pair count, and document the result with a focused test or reproducible calculation if the repository provides one.

Written by the indexing model from the issue text.

Assessment

Domain
distributed-systems
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.