NVIDIA / NVIDIA/nccl-tests

Two A100 nodes cannot reach ideal all-reduce performance

Open
#175 4 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Cuda
Stars
1.7k
Forks
406
PR merge metrics
No merged PRs in 30d

Description

Hello,
We ary trying to test all-reduce performance for A100 of two nodes, eyery node the configuration is as follows:
A100 * 8, IB 200Gb * 4 (dual port),
image

We tested three scenarios:

  1. ib_bw_write of NIC within two nodes
image
  1. singleNode, ./build/all_reduce_perf -b 8 -e 4G -f 2 -g 8, result: 232.54GB/s
  2. nccl all-reduce between two nodes only 86GB/s。There is a big gap compared to the theoretical value(100 GB/s), why is this?
    log:
    test.log

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the attached test.log and the reported all_reduce_perf command; compare the single-node and two-node measurements with the A100 and InfiniBand topology described. Use the all_reduce_perf entry point to investigate the reported gap. Done means a reproducible diagnosis and a validated two-node performance result.

Written by the indexing model from the issue text.

Assessment

Domain
distributed-systems, networking, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.