WoW performance worse with lower render scale vs lower resolution

doitsujin / dxvk

Vulkan-based implementation of D3D8, 9, 10 and 11 for Linux / Wine

zlib License

12.95k stars 832 forks source link

WoW performance worse with lower render scale vs lower resolution #3811

Open Espionage724 opened 8 months ago

Espionage724 commented 8 months ago

My main resolution is 1920x1080. If I set the in-game rendering scale option to 1279x719, performance improves a tiny bit. If I set the in-game resolution to 1280x720 and the rendering scale to 100% of that, performance improves massively.

I didn't test 720p on Windows, but 1080p 100% ran good that I didn't need to lower the render scale.

Software information

WoW Classic (install notes)
DirectX 11 Legacy, Triple Buffering unchecked, lowest Graphics quality (high textures), Windowed Fullscreen
Fedora Workstation 39 (GNOME on Wayland; WoW from Xorg looks choppy despite high FPS)

System information

GPU: Intel UHD 630
Driver: Mesa 23.3.3
Wine version: 8.19 Staging
DXVK version: https://github.com/doitsujin/dxvk/commit/1568c263fbec4c8d5f8f9286e0be40af7fdfb9a6

doitsujin commented 8 months ago

Don't think this is entirely unexpected? Render scale options will generally render the main scene at a lower resolution but the UI (and potentially some post-processing) will still run at 1080p, as opposed to running the entire game at 720p.

Will only consider this a bug worth looking into if DXVK behaviour is substantially different from Windows on AMD/NV HW (i.e. drivers that are generally known to not have weird performance issues on Linux).

SuperSamus commented 8 months ago

(Sorry if this is a stupid question) Could it be possible that in the Linux case, OP is being bottlenecked by VRAM?

doitsujin commented 8 months ago

Well we're talking old-ish integrated graphics here, so I think we can rule that out.

Reason why I don't really care too much is because these don't have the greatest track record in terms of DXVK performance.

Espionage724 commented 8 months ago

(Sorry if this is a stupid question) Could it be possible that in the Linux case, OP is being bottlenecked by VRAM?

With lspci it seems my GPU has 256MB by default, but I can raise that to 2G with some EFI var I think related to Aperture size. I'm not entirely sure how VRAM allocation works on pre-Xe Intel. I didn't consider this and will try it out though.