VR Benchmarking Results, using FCAT from nVidia

I was asked awhile ago (forget where!) to post my Temps & Usage of GPU & CPU.

The following is taken with the same Settings as the First Post in this thread, except with 29 AI at Sau Paulo in the Enduracers Mod.
As you can see the GPU hovers in the 80s/90s and hits 100% some times, whereas CPU hovers in the 40s - the early CPU spikes are loading screens.

Surprised nothing is melting, it's 28c ambient in my house when this was taken and the PC Case was 31c...it's very sweaty in the Rift!

 

I've added the Build Numbers to the Benchmarks on Posts 2 & 3, what I'll do is wait for a few more Builds then revisit each Benchmark to see if there's any significant difference.

 

I recorded just over 42c in my office/man cave the other day. I was sat here with sweat streaming down my face as I was going through paperwork.

Yesterday, warmest day in London, I tried rF2 (in the evening) and I saw my case temperature was 43c, but both CPU & GPU are on a custom loop, GPU went up to 50+c and CPU 60+c. Luckily I have a leather/foam face mount on the CV1 I can just wipe with a cloth as it got pretty sweaty under there.

 

Only Steam super sampling works, not Oculus debugging?

 

I've stopped using oculus debug tool or oculus tray tool to set a pixel density. I'm solely relying on Steam VR Supersampling, option ticked and set at 1.5 with AA=3.

 

Exactly @Magus - if they work together, it is a bug.

 

I was intrigued by the recent discussions about Oculus PPD vs SteamVR SS settings so did some tests this morning, as you can see the results pretty much speak for themselves!

The visual difference was slightly noticeable from what I saw so that suggests to me that they are working in Tandem, so each one has it's own way of achieving its output to the HMD. This maybe subjective, it may also be a Placebo (!).

IMO it's not worth running Both for that extra bit of quality, pick one over the other and stick with it. In may case quite clearly the Oculus at 1.4 (Light Blue) makes for better performance but the Steam VR at 1.9 (Pink) was visually a bit crisper to me. Remembering SteamVR now uses Linear settings so 1.9 used to be 1.4.

Note: This does not interrupt my First Post Build Benchmarks as I always run with the same settings anyway.

 

This does raise one question I've just thought of; *If we get the Option in the new upcoming UI to alter the PPD/SS which will it read from? Oculus or SteamVR...or both?

*I say "If" but really I mean "I hope that when we get the Option...." - As this is a Must Have for VR users in game

 

This is very misleading. You should set one of the two to the default - yet you've only measured SteamVR at 0.6 and 1.9. Why not at 1.0?

The reason the light blue (O: 1.4, S: 0.6) was performing better than the purple one (O: 0, S: 1.9), is because the combined result is a lot less pixels - which is exactly why it looked worse.

In other words, for an apples to apples, the setup should be:

1) Oculus default (0?), Steam 1.9
2) Oculus 1.4, Steam 1.0

These two should be equivalent (in both performance and quality), the only difference will be the filtering algorithm.

 

Thanks for doing this Marc, but @Milopapa. We shouldn't mix those multipliers (it is really nasty if they are not mutually exclusive). What will be also interesting to see if "Enhanced filtering" in SVR impacts performance.

 

Firstly I don't have all the time in the world to test every value nor is it totally worth it, secondly the 0.6 is the lowest setting in SteamVR as I was merely testing the lowest values in both, if SteamVR 1.0 = Oculus 0 then then I'll try and fit that benchmark in when I get time.
Again I don't know the exact ins and outs of how Oculus & SteamVR code their settings, I'm merely just doing simple benchmarks of which you can do yourself as well

 

That's exactly what I'm saying. SteamVR Supersampling = 1.0 is the default, unchanged resolution.

 

I get that - trouble is, you're making recommendations based on these benchmarks and people (who have even less clue) will be using your settings - that's why I'm pointing out the mistakes.

 

Lot of confusion around SteamVR supersampling, let me try to summarize what's been said above.

In SteamVR, you have a setting under Developer options called Supersampling, which forces supersampling for all VR games (some don't support it). Supersampling is upscaling the image during the rendering process which has the side effect of appearing sharper in the headset. (This is the same as DSR in the nVidia control panel for example, except that doesn't work in VR.)

The default setting is 1.0 - this is the standard rendering pipeline.
If you change this to 2.0, then the amount of pixels rendered will double, and consequently, it will need more GPU power.


If you lower this number below 1.0, you're telling SteamVR to downsample the image - which will result in the image losing details (ie. getting blurry). If you supersample this downsampled image (eg. by enabling Oculus supersampling), you might reach the original resolution (or more) BUT you will have already lost detail! See example below.


Now I'm not sure which supersampling effect will be processed first (Oculus or Steam), but to avoid loss of detail, you should never use downsampling on either side.
Hence my recommendation of not using a combination of SteamVR=0.65 and Oculus=1.4.

 

I'm not making recommendations to anyone, as per my first post "This may help you to find your own preference in settings". So if my own findings help others achieve better results or more fluid game play then that's great, if not then that's fine and they can keep trying. As for mistakes (!), like I said, you are free to do your own Benchmarks

 

I think Milopapa's just trying to say that the way you're you're trying the settings is not even helping you since the comparisons are not really valid. Especially since you (like most of us) don't have time to test everything a million times.
This is all really good stuff Marc so thanks for starting this thread and thanks to all involved. Can't wait to have a go at all this VR madness myself.
TK

 

Post #1 Updated with June 27th with Build 1915820, marginal improvement but an improvement none the less.

I've removed Build 1879993 from the results as that was obviously erroneous after a Software failure/issue.

 

A little hard to tell but it seems the last build really did improve performance, didn't it?

Do you have the raw data? It'd be interesting to see frametime distribution.

 

Big improvement here. Start up time less than half. And it's a little quicker and clearer. The wheel seems more responsive to me as well.

 

Default player.JSON - Apart from Rolling Start, as I start the Benchmark when the car crosses the S/F Line to prevent wheel spin smoke from a standing start.
Mercedes AMG GT3 + 19 AI (AI all the same car)
Grid Position 12
Silverstone International Circuit
PP Effects = Off
SuperSampling = Default (1.0)
AA = Off
HMD Only (no Mirror on monitor)
nVidia Drivers = 387.92

-=-=-=-=-=-=-=-=-

Test #1 - Pre-Windows 10 Fall Creators Update: