Files
Threadwell/archive/threads/NOTimothyLottes/2026-07-11-rethinking-vulkan-queue-choice/index.md
T
2026-08-25 16:26:52 -04:00

2.1 KiB

title, type, source, source_url, author, handle, post_id, date, archived, draft, tags, description, in_reply_to
title type source source_url author handle post_id date archived draft tags description in_reply_to
Rethinking Vulkan queue choice. archive twitter https://x.com/NOTimothyLottes/status/2075769880092037309 NOTimothyLottes NOTimothyLottes 2075769880092037309 2026-07-11 2026-08-23 false
archive
twitter
NOTimothyLottes
Rethinking Vulkan queue choice.

Source

Thread

1/ @NOTimothyLottes ^2075769880092037309

Rethinking Vulkan queue choice. Traditional double-buffered games might do something like the code below. Find the queue that does {presentation,graphics,compute}. First off this seems like a lot of unnecessary work when one could just choose queue 0 and be done with it ...

Branches: archive/threads/NOTimothyLottes/2026-07-11-rethinking-vulkan-queue-choice/2026-07-11-konrad_kubacki_-am-i-having-a-stroke-what-is-this-font-it-feels

2/ @NOTimothyLottes ^2075770634349527454

AMD on Linux (RADV) and Windows. Queue 0 just works. But note both platforms support doing present on queue 1. So on AMD I'm likely going to render via compute on queue 0 [to reduce chance of preemption] and present on queue 1. And have both decoupled (for front buffer rendering)

3/ @NOTimothyLottes ^2075771407510773906

NVIDIA is more complex. They put a DMA engine in queue 1 slot, and compute engine in queue 2 slot. Some sometimes queue 2 slot can or cannot do present (varies by driver). So maybe present on 0 (gfx queue), and dispatch on queue 2 by default for NV.