JuliaGPU / JuliaGPU/CUDA.jl

Add stream pool to avoid cuStreamCreate overhead

Open
#2,012 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

enhancement good first issue performance
Dominant language
Julia
Stars
1.4k
Forks
281
Avg merge
1d 7h
Merged PRs (30d)
30

Description

Describe the bug

I tried adding multithreading to our code, and now we see occasional pauses due to CuStreamCreate:
Screenshot 2023-07-29 at 5 27 06 PM

Could be fixed by adding a Stream pool?

To reproduce

Available on request

Expected behavior

No pauses

Version info

Not sure exactly, I think it was Julia 1.8.5 with CUDA.jl 3.13.1 and CUDA 11.2.

Once our cluster upgrades the drivers, I can rerun with CUDA 12.

Additional context

cc @lcw

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 reproducing the multithreaded workload described in the issue and profiling the CuStreamCreate pauses with the reported Julia and CUDA versions. Determine whether a stream pool addresses the observed overhead, then verify behavior under concurrent use and with CUDA 12 when the drivers are available.

Written by the indexing model from the issue text.

Assessment

Tech stack
julia
Domain
backend, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.