Skip to content

Add peer access support #479

Closed

Description

@vzhurba01

In a multi-GPU setting, right now it's not possible to memcpy directly between the two devices. A couple of checks would have to be done by cuda.core as part of enabling this:

  • cuDeviceCanAccessPeer
  • cuCtxEnablePeerAccess

Activity

  1. added
    featureNew feature or request
    cuda.coreEverything related to the cuda.core module
    on Feb 28, 2025
  2. added
    P1Medium priority - Should do
    and removed
    triageNeeds the team's attention
    on Mar 3, 2025
  3. added this to the cuda.core beta 4 milestone on Mar 3, 2025
  4. leofang commented on Mar 5, 2025

    @leofang
    Member

    This issue is actually very tricky because the modern way of exposing peer access support is again tied to the mempool/VMM APIs, such as cuMemPoolSetAccess/cuMemSetAccess, same as IPC (#103). We'll need a coherent, unified, and consistent solution here.

  5. added
    P0High priority - Must do!
    and removed
    P1Medium priority - Should do
    on Oct 21, 2025
  6. moved this from Todo to Done in CCCLon Dec 2, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

P0High priority - Must do!cuda.coreEverything related to the cuda.core modulefeatureNew feature or request

Projects

  • Status
    Done

Relationships

None yet

Development

No branches or pull requests

Issue actions