Renderdoc captures aren't portable. They can only be replayed on a system with a GPU of the same architecture and the same driver.
Yes, drawing 10000 objects when looking in one direction is slower than drawing 3000 objects when looking elsewhere. This is normal and has happened in literally every 3D game ever made that doesn't have an FPS cap.
Does not happen with Native DX11
DXVK isn't native D3D11, and AMD's new driver is expected to outperform our implementation in almost all situations. You're also running a Vulkan driver that is known to perform poorly with DXVK especially in GPU-bound scenarios.
DXVK isn't native D3D11
Yes. I know. My point is, this performance degredation only happens with DXVK. In fact, I've never seen such a big performance difference between DXVK and native directX in any game. Usually the difference was much smaller.
and AMD's new driver is expected to outperform our implementation in almost all situations.
I wouldn't bet on AMD backporting their driver improvements for RDNA1 gpus
You're also running a Vulkan driver that is known to perform poorly with DXVK especially in GPU-bound scenarios.
So you are saying, vulkan on windows is inferior to vulkan on linux?
So you are saying, vulkan on windows is inferior to vulkan on linux?
On AMD GPUs: Yes, absolutely. On Nvidia GPUs it's practically identical.
Thanks for the info.
Nothing extracted yet.
With the PC Performance Test (built-in benchmark) during specific sections, the FPS drops massively, from around 70 to around 40.
OS: WIndows 10 21H2
GPU: AMD Radeon RX 5700 - Drivers 22.5.2 WHQL
CPU: AMD Ryzen 9 3900X
The following graphic settings were used:
Note: Does not happen with Native DX11 and disabling Gameworks Enhanced Rain & Light Shafts does NOT make a difference.
Here is a RenderDoc capture