Can you: find out which version in particular introduced the performance regression? And also test the latest unreleased build?
IIRC my testing, playing with dxvk.enableDescriptorBuffer and d3d9.deviceLocalConstantBuffers were interesting knobs in that regard. EDIT: AFAIR even d3d9.cachedWriteOnlyBuffers could help in this regards?
IIRC my testing, playing with
dxvk.enableDescriptorBufferandd3d9.deviceLocalConstantBufferswere interesting knobs in that regard. EDIT: AFAIR evend3d9.cachedWriteOnlyBufferscould help in this regards?
Unfortunately, this did not help; it reduced the FPS even further. Now 4 benchmarks give an avg. FPS of 149,65, before it was 153.95.
As K suggested, could you try to restrict the version range? And also try the latest git?
In this post, I was referring to both v2.7.1 in form of the latest release as well as the latest build, since the figures don't really differ that much or are within the margin of error.
could you try to restrict the version range?
Yeah, I can do that.
Btw. are you able to reproduce my observations on your end?
Greetings, @Mofkos
I am DXVK-GPLALL developer. Please, can you check DXVK-GPLALL 2.6.8-2? I am interested in your feedback.
Nothing extracted yet.
Well, I'm not sure if this can be considered a bug (hence not a bug report), but it's still something worth noting I think. It appears that in v2.7.1, the FPS you get is a bit lower compared to early 2.x versions in CPU-bound scenarios (no, not because of ReBar being disabled). I tested this on the Complete Edition.
This is my system:
RTX 3080 (drivers: 596.36)
Ryzen 7 3700X
32 GB DDR4 RAM
Windows 11 (build number: 26200.8328)
The only custom setting in dxvk.conf I set is "d3d9.maxFrameLatency = 1".
I've done 6 benchmarks via the in-game benchmarking feature, 3 with v2.7.1 and 3 with v2.2, and restarted the game after each one. Resolution was Full HD. These are the results:
v2.2 FPS: 167.10, 169.58, 166.48 -> Avg: 167.72
v2.7.1 FPS: 152.23, 155.46, 154.15 -> Avg: 153.95
Meaning a reduction of 8.21% FPS on average.
I believe that this is because of a higher load on the CPU in v2.7.1, as this game is heavily CPU-bound. Avg CPU usage in v2.2 was 10.67% and in v2.7.1 it was 12.33%, even though the FPS was 8.21% lower like I said.
I also did a custom benchmark where the game was highly CPU-bound all of the time (~50% GPU usage). I just drove this straight road. I did 8 benchmarks in total, 4 for each. In every benchmark:
(all set by a mod menu)
So I made them as equal as possible and therefore the variance between them was about the same as between the normal benchmarks.
v2.2: avg FPS: 92.63, avg CPU usage: 22.25%
v2.7.1: avg FPS: 86.58, avg CPU usage: 24.25%
-> reduction of 6.53% FPS on average.
While I'm still incredibly thankful for that mod, it is still a bit disappointing to see that in 3 years of DXVK development, the performance didn't get any better but in fact even a bit worse (yeah, I know that DXVK isn't only meant for GTA 4 and that you guys are also working on ensuring compatibility with other games but still). Now the question is: Is this performance regression in the newest version normal? Is there anything I could try in the conf file in order to boost performance?