xpra icon
Bug tracker and wiki

Opened 2 months ago

Last modified 25 hours ago

#2416 assigned enhancement

CUDA device selection by name

Reported by: Antoine Martin Owned by: Antoine Martin
Priority: major Milestone: 4.0
Component: encodings Version: 3.0.x
Keywords: Cc:

Description

Follow up for #2415: we should be able to select the list of enabled / disabled devices using their name and / or PCI bus location (ie: 0000:1A:00.0) without doing a full CUDA context initialization.

load_device can get to the device name before calling make_context (which is slow).

Change History (4)

comment:1 Changed 2 months ago by Antoine Martin

Status: newassigned

We can't skip the make_context completely in select_device because driver.mem_get_info() requires a valid context.

Maybe we can have different policies for load balancing, and one may not require memory info to do its job.

Last edited 2 months ago by Antoine Martin (previous) (diff)

comment:2 Changed 3 weeks ago by Antoine Martin

Updates:

  • r24331: can use device name or PCI id to enable or disable devices
  • r24332 + r24333: update examples (maybe we should ignore the PCI prefix? or allow a pattern match?)

comment:3 Changed 3 weeks ago by Antoine Martin

r24340 adds a load-balancing attribute and a round-robin option, the memory policy is still the default. (chooses the GPU with the most free memory)

Still needs testing.

comment:4 Changed 25 hours ago by Antoine Martin

Seems to have caused a bug: #2489.

See also initial load balancing code in #520.

Last edited 25 hours ago by Antoine Martin (previous) (diff)
Note: See TracTickets for help on using tickets.