AMD vs nvidia

Help!
1

Today i tested AMD card vs nvidia on vrchat and we’ll.. this is very bad.

I tested 2 quite low end but usual cards

A AMD rx5700xt

And a gtx1070

Those have the same amount of VRAM and are quite cheap.

But are a good choice for gaming.

On paper the 5700xt should be around the double of raw power of the 1070.

But in use both cards are at exactly the same fps, around 60 to 70 in the room i tested.

There is clearly a bias toward the Nvidia card.

So i tried to hide the hardware with the AMD card with a few driver spoofing trick

And found out the card was now running at the expected 110/120 fps .

There is clearly hardware detection and a “AMD fix” ruining vrchat performance.

My trick sadly work only one time and go after a reboot so, any ideas?

1 Like
2

Did you perform the test in VR or on Desktop and did you have any NVidia optimizations enabled…

AMD tend to also act differently with different CPU’s (prefers their own AMD CPU’s over Intel and can actually give a 20-80% better performance if the GPU and CPU are both AMD)

How much RAM and what kind if RAM does your testbench have, which is also a major factor as NVidia uses more RAM as before processing into VRAM)

3

intel cpu (2683v4 6th gen xeon cpu) amd graphic card (5700xt)
32gb ram ddr4 windows 11 24h2.
tested on desktop and vr
i know vr is pretty bad on amd due to bad driver optimization and slow shader execution of amd cards.
windows and linux (Bazzite deck).
i found this issue with unity too

This affect AMD gpu and integrated graphics only.

It is a bug found in unity engine itself.

This can lead to +50% more frame in desktop mode.

Sadly it doesn’t affect vr mode.

But this will work on every flat screen unity games.

Source

so yea it seems to be a combinaison of multiples things.

4

I have some questions about the test environment.

  • What world were you loading?
  • Were you alone in the instance, or were others present?
  • Was this done in desktop or in VR? If desktop, did you have vsync and/or variable framerate (freesync/gsync/etc.) enabled?
  • Did you have any cameras or mirrors open? Was this the same on both setups?
5

avatar testing chamber
alone to avoid any variation
desktop, variable refresh rate, no cap or vsync
one mirror, both the same on both setup.

1 Like
6

Okay yeah that sounds like an incredibly fair test.

Although, Avatar Testing Chamber is a very well-optimized world. Have you tried doing a similar comparison using a world that isn’t so optimized? It’d be interesting as a comparison, like maybe something VRChat is doing is accidentally enabling vsync.

7

The world is not that optimized.
It has bloom,ambient occlusion and a video player running in the background.
And to avoid any vsync issues, the screen used is set to 165hz as refresh rate.