dxvk-async

An attempt to improve the dxvk-async patch at https://github.com/jomihaka/dxvk-poe-hack

Improvements

• Compatible with dxvk v1.4.3 - v1.10.3

• A lot less stuttering (nearly none at all in some games) by not blocking the main thread when compiling async pipelines.

• dxvk-async-e3a63d4.patch and earlier: Async pipelines are written to the state cache. The original patch doesn't since it was made before dxvk introduced the state cache. Previously you would end up with near empty state caches when using dxvk-async, now you can have the best of both worlds.

• dxvk-async.patch uses the existing worker thread system (controlled by dxvk.numCompilerThreads) to compile async pipelines.

• dxvk-async-ab1d629.patch and earlier: A new option dxvk.numAsyncThreads to specify the number of async pipeline compiler threads. Previously half the cpu thread count was used by default, now it's the same logic as the state cache.

Caveats

• dxvk-async-ab1d629.patch and newer: Enabling dxvk-async will force disable the new dxvk option: dxvk.enableGraphicsPipelineLibrary. This is currently necessary for dxvk-async to work.

• dxvk-async-af418dc.patch and earlier: The shader cache can be around 5-10% larger.

Instructions

• Patch dxvk with dxvk-async.patch
• Set the environment variable DXVK_ASYNC=1 or use dxvk.enableAsync = true in dxvk.conf
• To compare with stock dxvk, rename or delete your state and shader caches
• Use DXVK_HUD=pipelines to see the pipeline count go up (hopefully) without stutter

Compatibility Matrix

Warnings

dxvk-async could theoretically trigger client-side anti cheats, and as such, may be risky to use inside of multiplayer games. If you have been banned because of dxvk-async, please report about it at this issue here with information about your system (like a neofetch), distribution/OS (BSD, GNU/Linux, etc.), how long it took the ban to take effect, and whether or not you were able to get unbanned.

As of 2022-07-16 we are not aware of any bans that happened due to dxvk-async.