ModelEngine-Group / ModelEngine-Group/unified-cache-management

[Performance]: Improvements good on single npu but not obvious on multi-npu environment

Open
#657 13 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

performance
Dominant language
C++
Stars
334
Forks
119
Avg merge
1d 15h
Merged PRs (30d)
82

Description

Proposal to improve performance

Hi, we test the situation under 0% hit rate and 100% hit rate with 1 npu and 4 npu. UCM performs well on a single NPU, reducing TTFT by 35%. However, its performance is mediocre on 4 NPUs—the TTFT is almost the same as when UCM is not used. I’d like to ask what might be causing this issue.

I also wonder why TPOT decreased under 100% cache hit rate. From my understanding, UCM should only accelerate the prefill stage.

Report of performance regression

No response

Misc discussion on performance

No response

Your current environment (if you think it is necessary)
The output of `python collect_env.py`

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 by reading the issue's 1-NPU and 4-NPU observations, then run the project's performance tests with 0% and 100% cache hit rates. Use the environment details from collect_env.py and compare TTFT and TPOT with and without UCM; done means the multi-NPU behavior and TPOT change are explained or reproduced with actionable evidence.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
ai-infra-agents, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.