AMD (RX 6750 XT): are the remaining performance issues from #2058 planned to be fixed? #2819

Open
opened 2026-11-12 15:49:07 +00:00 by github-sync · 0 comments
Owner

@gabrielfnatic1542-commits opened on GitHub · 2026-11-12 15:24 UTC · original

Hi! Following up on #2058 (closed and merged into #993).

Setup: RX 6750 XT (RDNA2), Ryzen 5 3600, 16 GB RAM, v2.1.0 with RX_VK_COMPACT=1, 1080p output, cap 90.

VRAM is fixed with compaction, thanks. What's still left on AMD:

  1. Low FPS: Hyrule Field 24–40, Castle Town 16–22. Per frame ~37 ms of CPU replay/submit (draw 23 ms, descriptors 10 ms) + ~22 ms GPU, with ~3000 draws. The same game runs well on an RTX 4060, so it looks like the AMD path needs optimization (task/mesh shaders on the compute fallback, 11 programs "not translated: a network left out on this card").
  2. A 41 s stall when entering an area: working set dropped from 11.9 GB to 5.9 GB (private ~29.9 GB on 16 GB RAM), which looks like paging.

Is there a plan or timeline to fix these on AMD? If not, I'm thinking of forking Hookshot and trying to fix the AMD path myself. Any pointers on where to start would be appreciated.

> [@gabrielfnatic1542-commits](https://github.com/gabrielfnatic1542-commits) opened on GitHub · 2026-11-12 15:24 UTC · [original](https://github.com/NotRexed/Hookshot/issues/2819) Hi! Following up on #2058 (closed and merged into #993). Setup: RX 6750 XT (RDNA2), Ryzen 5 3600, 16 GB RAM, v2.1.0 with RX_VK_COMPACT=1, 1080p output, cap 90. VRAM is fixed with compaction, thanks. What's still left on AMD: 1. Low FPS: Hyrule Field 24–40, Castle Town 16–22. Per frame ~37 ms of CPU replay/submit (draw 23 ms, descriptors 10 ms) + ~22 ms GPU, with ~3000 draws. The same game runs well on an RTX 4060, so it looks like the AMD path needs optimization (task/mesh shaders on the compute fallback, 11 programs "not translated: a network left out on this card"). 2. A 41 s stall when entering an area: working set dropped from 11.9 GB to 5.9 GB (private ~29.9 GB on 16 GB RAM), which looks like paging. Is there a plan or timeline to fix these on AMD? If not, I'm thinking of forking Hookshot and trying to fix the AMD path myself. Any pointers on where to start would be appreciated. <!-- github-sync:item:2819 -->
Commenting is not possible because the repository is archived.
No milestone
No assignees
1 participant
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Hookshot#2819
No description provided.