Vulkan API – Feedback, Known Issues, and FAQ

how much vram is needed to run ultra settings with motion blur and dynamic reflections/mirrors

 

10.9 GB VRAM is used on my RX 6800 I just got with everything on max @ 3440x1440 with T-series on WCUSA and exhanced mirrors enabled

 

What is your average frame rate?

 

Hello, with Vulkan I'm getting some pretty weird results regarding performance that I don't have with Direct3D.
My specs are the following:
NVIDIA GeForce RTX 3050 Laptop GPU
11th Gen Intel Core i7 - 11800H (2.3GHz)
16GB of RAM

With Vulkan my FPS drop drastically in big maps like Italy or WCU but increase in small maps like Gridmap.
In Italy, my GPU usage is around 90 - 100% all the time with some drops to 50 - 70% for a few moments and I'm getting an average of 16 FPS.

My CPU, in the other hand, is laying at 20% of usage
What's interesting though, is that there's a constant GPU Bottleneck as well as some spontaneous CPU and Game Engine bottleneck that keep flashing in the graph.


Any help would be appreciated!

 

Settings? May be a VRAM limit.

 

What annoys me is that in Direct3D I get 10 times better performance....

 

Ah, it all makes sense now... It turns out, that for some reason, task manager incorrectly reports GPU usage when running BeamNG with Vulkan. Task manager says ~5% (no way it's that low, lol), but Nvidia's performance overlay says 97-100%. No wonder BeamNG still says my GPU is the bottleneck...

 

Isn't this issue not as bad on dx11 ? what makes vulkan worse on this ?

 

This makes sense, but what's weird is that my VRAM isn't maxed out. It's only about half used while the GPU usage on task manager is low, and the GPU usage on the Nvidia performance overlay is 100%. I checked for the amount of VRAM being used by looking at both task manager, and BeamNG's GPU Memory Profiling, which showed the same amount.

 

Ah, okay, that makes sense. Thank you for explaining

 

It’s a bit out of context, what Tool would you recommend for monitoring resources on windows rather than taskmgr?

 

I experienced an amdgpu crash, similar to before where all graphical applications die and the window manager restarts, but in a slightly different way this time. This time everything was working great, no graphical artifacts, good frame rate, and then after quite some time, probably 45min-1hr it crashed. I was not keeping track of VRAM usage at the time, but my GPU has 32G of dedicated memory if that's important. Kernel log is below. I'm on Linux running in Proton using the "-gfx vk" launch option. Specs are also below.

Code:

kernel: amdgpu 0000:2f:00.0: amdgpu: [gfxhub] page fault (src_id:0 ring:24 vmid:2 pasid:32800, for process BeamNG.drive.x6 pid 243826 thread   --- Render -- pid 243893)
kernel: amdgpu 0000:2f:00.0: amdgpu:   in page starting at address 0x000080003f800000 from client 0x1b (UTCL2)
kernel: amdgpu 0000:2f:00.0: amdgpu: GCVM_L2_PROTECTION_FAULT_STATUS:0x00201431
kernel: amdgpu 0000:2f:00.0: amdgpu:          Faulty UTCL2 client ID: SQC (data) (0xa)
kernel: amdgpu 0000:2f:00.0: amdgpu:          MORE_FAULTS: 0x1
kernel: amdgpu 0000:2f:00.0: amdgpu:          WALKER_ERROR: 0x0
kernel: amdgpu 0000:2f:00.0: amdgpu:          PERMISSION_FAULTS: 0x3
kernel: amdgpu 0000:2f:00.0: amdgpu:          MAPPING_ERROR: 0x0
kernel: amdgpu 0000:2f:00.0: amdgpu:          RW: 0x0
kernel: amdgpu 0000:2f:00.0: amdgpu: [gfxhub] page fault (src_id:0 ring:24 vmid:2 pasid:32800, for process BeamNG.drive.x6 pid 243826 thread   --- Render -- pid 243893)
kernel: amdgpu 0000:2f:00.0: amdgpu:   in page starting at address 0x000080003f800000 from client 0x1b (UTCL2)
kernel: amdgpu 0000:2f:00.0: amdgpu: GCVM_L2_PROTECTION_FAULT_STATUS:0x00000000
kernel: amdgpu 0000:2f:00.0: amdgpu:          Faulty UTCL2 client ID: CB/DB (0x0)
kernel: amdgpu 0000:2f:00.0: amdgpu:          MORE_FAULTS: 0x0
kernel: amdgpu 0000:2f:00.0: amdgpu:          WALKER_ERROR: 0x0
kernel: amdgpu 0000:2f:00.0: amdgpu:          PERMISSION_FAULTS: 0x0
kernel: amdgpu 0000:2f:00.0: amdgpu:          MAPPING_ERROR: 0x0
kernel: amdgpu 0000:2f:00.0: amdgpu:          RW: 0x0
kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx_0.0.0 timeout, signaled seq=4710067, emitted seq=4710069
kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process BeamNG.drive.x6 pid 243826 thread   --- Render -- pid 243893

Code:

OS: Arch Linux x86_64
Host: MS-7D53 1.0
Kernel: 6.6.12-1-lts
Resolution: 3840x1600
WM: sway
CPU: AMD Ryzen 9 5950X (32) @ 3.400GHz
GPU: AMD ATI Radeon PRO W6800
Memory: 64223MiB

 

I've been using Vulkan myself for a while, because 1080TI can't handle directx, 75Hz minimum and 1440p in more complex maps. Vulkan does it much better, but there is a downside.
Directx spills pixels more nicely. Vulkan works as if there is no anti-aliasing completely. A way to get better pixel shuffling in certain areas is to use CK Graphics Settings in, for example, version 2.3 with function HigherQuaklity GTAO. This won't fix jagged lines, but it will help get a visibly better effect.
However, this is burdened, as I noticed at my place, with 10% more power demand on the graphics card.

https://drive.google.com/file/d/19mMKCL_dm7mecPZ0TeG5VoPb5PAmlNP_/view?usp=drive_link
https://drive.google.com/file/d/1lsi8C9VUn0C6t2HoOqb8Itf9HZs1MDc0/view?usp=drive_link

 

Oh I see! Do you think this could be improved for the release of Vulkan or will I have to stay with Direct3D?
Thank you so much anyway!