Performance regressions on the AMD Windows driver are somewhat expected.
Meanwhile it should be faster on Linux and on every other driver.
Yes, this is known, and will not be addressed. All, and I really mean all other drivers that we've tested recent changes on have seen performance improvements if anything, AMD's Windows driver (at least on RDNA1/2) is the odd one out, but it also doesn't get any feature updates anymore (so it is already hitting slow paths) and any sort of current profiling tools also aren't supported on these GPUs anymore so I frankly just don't care.
Switch to Linux if you want a good D3D9 experience, otherwise stick to older DXVK or native DX9.
Reallly amd fuck up his drivers on windows. i dont know why them no supports the compatibility for this project, when they lack optimal compatibility for old dx9 games or opengl on purpose.
ok I'll give linux a chance, cause I'm afficionate of playing old games.
@cfelitti
What CPU do you have?
P.S: 12 logical threads seems more reasonable, since most CPUs are at least 6/12, and should still solve address space issues.
@Digger1955 I'm at the bottleneck limit, I guess... i7 6700, 16 GB of RAM, 1 TB SSD. disable all c states, disable meltdown and spectre, tweak a little volt, and i disable the p states on the gpu. Anyway the gpu amd driver sucks. Call of duty 2003 stutters, dx9 games stutters, and relative modern games works fine. Just there is not will from them correct this or support compatiblity on windows to work dxvk.
Anyway the latest worflow build that says something like fog, works again like it should.
@cfelitti
Before trying Linux (although it is inevitable at some point), try DXVK-GPLALL. DXVK-GPLALL 2.6.7-1 should work better in your specific situation.
Nothing extracted yet.
Like i say, the performance regression is notorious, testing with the commit 660d382. i testing with the previous commit "version" and work flawlessly, i always remove the dxvk cache before testing to if i see stuttering, and im not having almost stuttering. but when i remove the previous cache for testing the new commit i see agressive stuttering.
In this regard, the new commit stutters more when compiling the shader and creating the cache than the previous version. If further testing is required, I'll make an apitrace and provide the logs.
System information