Multi-GPU VR is only ~3% faster here but why is it working at all?

Hi All, maybe I'm very behind on this, it's from 2018. Was doing some late night trawling of GPU tech and came across this:
Source: https://www.youtube.com/watch?v=gFfQorILRf4


ED is at time code 4:47-5:14. Assuming the average fps is aggregated for the clip, frame time is slightly down ( and up in other games ) but, bizarrely, ave fps is up by 3.3% for ED. Multi-GPU ( except on games designed for it ) doesn't do anything, we know this. So, my question would be, why is it working at all? The pattern in the various game tests, seems pretty consistent, +/- 1%, which, to me, moots a 'within tolerance' point but it shouldn't work on any level.

Not that I'm about to rush out and buy a couple of 3090s... If it was 10% I might since my credit card is empty and it looks like I'm not going on hols for a few years...

This next chap, at 8:11, underscores that modern titles like ED do not support multi-GPU VR because, as we already know, almost no-one uses VRWorks so, as I asked... why is there this tiny performance bump? It should have been making no difference whatsoever.
Source: https://www.youtube.com/watch?v=xaGK_ytNQfc
 
I've said this all along and nobody believed me. When I played it with SLI 670s and 970s the main frame rate pretty much stayed put in VR but it didn't drop frame rates as much and was a smoother game. You have to connect the VR headset to the same card as your monitor for it to work or it's a mess. I'm guessing everyone who reported on it not working plugged the headset into one card and the monitor into the other.

I played Elite with SLI from the first week of VR on 670s on DK1, through DK2 and Vive, until GTX 980Ti's came out where I moved to a single card.
 
Last edited:
You appear to be vindicated (y)

Hey I don't supposed you kept any of your frame time comparisons? Be interesting to see the variation. I'm guessing you saw far fewer spikes.
 
ED is at time code 4:47-5:14. Assuming the average fps is aggregated for the clip, frame time is slightly down ( and up in other games ) but, bizarrely, ave fps is up by 3.3% for ED.

How are they duplicating the tests between configuration? If it's done manually, then 3.3% could easily be within margin of error.

This next chap, at 8:11, underscores that modern titles like ED do not support multi-GPU VR because, as we already know, almost no-one uses VRWorks

Are we certain that traditional AFR cannot be forced in VR? VRWorks and other VR specific multi-GPU APIs are about splitting the work per eye, but two GPUs should hypothetically still be able to leverage AFR or SFR with two view ports the same way it does with one.

I've certainly seen reports of people seeing improvement with SLI and/or Crossfire in VR. Haven't tested it myself though.
 
I'm certain of nothing, although I don't agree with you about the error. However that's just my personal opinion rather than a qualified assessment and I do take your point. I suspect I'm being subjective.

I do recall, when I tested it on a pair of 970s with my Vive, both GPUs were heavily utilised but I just wasn't seeing any improvement in actual fps. I didn't check frametimes so I've no idea about that.
 
AFR and traditional SLI can give you better framerates but it increases head tracking latency which is heavily detrimental to the VR experience.
 
I'm seeing 77% on GPU1 & 11% on GPU2.
Can I set it better than below?
I have an Oculus Rift S connected.

1604435615117.png


1604435654762.png
 
Last edited:
I had issues using SLI for VR, so have turned it off. I mostly use multiple GPUs for data processing in CUDA-friendly programmes anyway, so can live without it while in VR space. When/if the summer gets too hot here and I leave the VR headset off again (if I can live without it), I'll turn SLI back on and enjoy a bit of added grunt as compensation for having to play in flatland.

:D S
 
Because developers haven't coded for VR SLI basically. They can program one GPU per eye, but they don't.

SLI will boost a little in 2D but not in VR.
There are many games which run slower in SLI VR believe it or not.

Which is why SLI has been dropped on most cards.v 3090 is the only new version with it.

Shame, I love the idea of it...
Shame developers never made good use of it.




Hopefully DLSS 2.1 or 3.0 comes soon. 2.1 will use DLSS cores to boost VR supersampling. If game learned by Nvidia. 3.0 apparently may work on any game using TAA and will not need to be learned by nvidia
 
Because developers haven't coded for VR SLI basically. They can program one GPU per eye, but they don't.

SLI will boost a little in 2D but not in VR.
There are many games which run slower in SLI VR believe it or not.

Which is why SLI has been dropped on most cards.v 3090 is the only new version with it.

Shame, I love the idea of it...
Shame developers never made good use of it

I use my second GPU for PhysX when in SLI. No frame rate change, and getting things to run smooth gets a bit fiddly. But it appears to result in enough GPU grunt that chaos can erupt on-screen without noticable reduction in performance.

:D S
 
Back
Top Bottom