Benchmark Result Discussion

[Hedgepig]

@Hedgepig, I have a 5600X and a 3060 (no Ti for me .. boo) and was getting lower FPS than you in desktop mode, so it seems the Ti part certainly makes a difference, I think the 3070 and 3080 people also get better performance. I don't think the CPU is a "bottleneck" as such. Think of it more as a half and half. Speeding either half up certainly speeds the whole result, but you get diminishing returns if you don't reduce each half sort of equally. An awful CPU offsets a LOT of expense on a high end GPU where you would get better results spending less money on a slightly better CPU (if I explained that well, sorry otherwise).
So think of it less as a bottleneck and more of a balancing act with cost as a major factor.

I was like, no 3070/3080 = boo. But, as is the case with motorbikes-- you can have a seriously fast bike, but there's always guys and women who'll have faster bikes than you. Fun literary fact: did you know, the poet, Sylvia Plath's, daughter, Frieda, rode a Suzuki Hayabusa? Suzuki Hayabusa - Wikipedia My bike only had half the Hayabusa's BHP but it was still so much fun to ride. (Miss riding, so bad) So, some of the experience is subjective.

Relating to VAM, experimentally, I've run scenes with multiple characters, and these take the FPS down to 30-40 frames, it's only around there that things slow down to the point, for me, it breaks the immersion in the scene. And, that's got to be my subjectivity coming into play. For others, who are more cognitively sensitive to low FPS, 30-40 FPS might be a problem for them. Hell, I started using VAM on a crappy moped-of-a-PC, with every boxed ticked in Give ME FPS. I was only getting 15FPS, but I still loved it. When I got the 3060, it was like the gates of heaven and hell had simultaneously opened up an fused into a single coherent reality, and I was like... OMFG...

 

[geo_gan]

My results.

Desktop

Oculus Rift S

I notice the resolution here seems to be about 1.2/1.3 times the default Rift S panel resolution for some reason

Can someone explain the "Internal Resolution" displayed in VR results?

On MacGruber VR benchmark he used Rift CV1 which has a per-eye panel resolution of 1080×1200... so VAM should render at 2160x1200... but result said resolution was 2688x1600

On my benchmark for Rift S which has a per-eye panel resolution of 1280×1440... so VAM should render at 2560x1440... but result said resolution was 3296x1776

In both the "Resolution Scale" was 1x

I don't think I have set any super-sampling anywhere

This appears to be about a 1.3x scaling in all dimensions - why is this??

 

[MacGruber]

I notice the resolution here seems to be about 1.2/1.3 times the default Rift S panel resolution of 2560×1440. I don't think I have set any super-sampling anywhere?!?

Totally normal, I think most/all VR headsets render a bit larger field of vision so they can more quickly adapt to your head rotation by showing you the right section....much like a VR video where you can look around a bit. There is a reason VR eats so much resources

 

[geo_gan]

Totally normal, I think most/all VR headsets render a bit larger field of vision so they can more quickly adapt to your head rotation by showing you the right section....much like a VR video where you can look around a bit. There is a reason VR eats so much resources

I found answer here https://venturebeat.com/2019/06/02/oculus-rift-s-has-a-well-hidden-resolution-setting/

• Prioritize Quality: 1648-by-1776
• Prioritize Performance: 1504-by-1616

Seems I was losing a lot of performance with this setting stuck on "Prioritize Quality" for years.... here is new VR run with the setting at "Prioritize Performance"
Notice the internal resolution now matches above numbers!

It looked to me that the benchmarks would not go over 80FPS while running - is this a limit of Oculus driver? I definitely had ASW off (ran your script even). I notice avg and min1% gone up but the max1% dropped from over 160FPS to just about 83FPS

 

[MacGruber]

It looked to me that the benchmarks would not go over 80FPS while running - is this a limit of Oculus driver?

VSync for the 80Hz display kicks in, can't be turned off apparently. But there is no point in rendering more frames than you can display....unless you want to benchmark

 

[geo_gan]

VSync for the 80Hz display kicks in, can't be turned off apparently. But there is no point in rendering more frames than you can display....unless you want to benchmark

May need to make the benchmarks a bit more difficult then, since all GPUs over about 3070 will hit this hard limit in VR and all end up with same numbers... would be like everyone running the 3DMark Mobile benchmark!

Maybe need to add another character at least?

Probably best to set top end of scale for 3090 performance (ie make it so even a 3090 doesn't even hit 80FPS in Rift S in VR)

 

[KamiKamui]

I thought my computer was a beast, but it is apparently no match for VaM while running the HP Reverb G2...

Notes:
-RAM is clocked to 3600MHz and is CL16.
-CPU not running OC. FCLK is set to 1800MHz (My CPU's infinity fabric).
-GPU is the Asus ROG Strix OC model. Factory OC.

 

[KamiKamui]

VR mode with the Reverb G2. If you were thinking of getting a 3090 to drive a pixel count like the G2, maybe save up and wait for the next generationView attachment 71663

I am struggling to understand how your system runs 13 FPS better. Essentially the only difference between our GPUs is the VRAM. Does VaM fully saturate your VRAM?

 

[Bob Nothing]

Turn off 2x scale! Just run at 1x - that 2x VR resolution on a CV1 is crazy

Question for all - how did you enter name in results picture? Mine just says Anonymous!

Thanks for the tip, I'll try that out! I don't think I manually set it to that?? I must have I guess, moved my NVME from my old build into this new 3090 rig and the launched it up. Need to tweak this rig, it's just 'out of the box' at the moment.

 

[Bob Nothing]

I am struggling to understand how your system runs 13 FPS better. Essentially the only difference between our GPUs is the VRAM. Does VaM fully saturate your VRAM?

Yes, it really does. I have a 2090 that has 24GB of VRAM and I've exceeded 12GB of used VRAM and I haven't even really stress tested it yet. (Key word being yet, I plan to do all sorts of horrible things to it.)

 

[KamiKamui]

Yes, it really does. I have a 2090 that has 24GB of VRAM and I've exceeded 12GB of used VRAM and I haven't even really stress tested it yet. (Key word being yet, I plan to do all sorts of horrible things to it.)

I see. Thanks for the info. Here I was convinced 24GB of VRAM was totally unnecessary except for people who work in Blender and other 3D creation tools...

 

[Bob Nothing]

I see. Thanks for the info. Here I was convinced 24GB of VRAM was totally unnecessary except for people who work in Blender and other 3D creation tools...

Honestly, this is a creation tool like Blender and others

 

[Jiraiya]

Well that makes me feel better about my "pointless" 12gb 3060 vram.

 

[Bob Nothing]

May need to make the benchmarks a bit more difficult then, since all GPUs over about 3070 will hit this hard limit in VR and all end up with same numbers... would be like everyone running the 3DMark Mobile benchmark!

Maybe need to add another character at least?

Probably best to set top end of scale for 3090 performance (ie make it so even a 3090 doesn't even hit 80FPS in Rift S in VR)

Also I should add that I don't have a CV1, it's a Quest 2. Not sure why it shows up as a CV1 in the benchmark too.

 

[WanderingWomble]

I am struggling to understand how your system runs 13 FPS better. Essentially the only difference between our GPUs is the VRAM. Does VaM fully saturate your VRAM?

Ah, I'll go back and edit my post when I've checked my settings, but if you compare our resolutions you can see that I'm not running at 100% resolution in SteamVR. I bet that 13fps is more down to me having to calculate less pixels than you!

 

[WanderingWomble]

Ah, I'll go back and edit my post when I've checked my settings, but if you compare our resolutions you can see that I'm not running at 100% resolution in SteamVR. I bet that 13fps is more down to me having to calculate less pixels than you!

Nope, I take that back, I'm at 100% in steam and 1x in VaM.

Give your Steam settings a quick check, I bet it's just over 100%

 

[WanderingWomble]

Ok, I'm running another test, this time I've put my headset on the desk during warmup so I can watch task manager. My VRAM is indeed above 12GB sitting at 12.7 currently. 65C temp air cooled on a stock 3090 FE. CPU is at 4.16GHz. GPU is at about 87% usage (bounces a bit of course but surprisingly flat graph). 13.2GB now the first test scene has loaded. Does it need 13.2GB or has it left some stuff from the last scenes in that it could safely drop? It's not gone up again for the next set of tests. My RAM (3600 DDR4) is 18.6GB/32GB, but chrome is running too

and the results are in!

 

[KamiKamui]

Ok, I'm running another test, this time I've put my headset on the desk during warmup so I can watch task manager. My VRAM is indeed above 12GB sitting at 12.7 currently. 65C temp air cooled on a stock 3090 FE. CPU is at 4.16GHz. GPU is at about 87% usage (bounces a bit of course but surprisingly flat graph). 13.2GB now the first test scene has loaded. Does it need 13.2GB or has it left some stuff from the last scenes in that it could safely drop? It's not gone up again for the next set of tests. My RAM (3600 DDR4) is 18.6GB/32GB, but chrome is running too

and the results are in!

View attachment 72054

For some reason my resolution is 7768x3792 and yours is 6344x3100. Perhaps it is related to that pixel count difference?
My CPU stays around 4.9GHz 62C running the test and my GPU reaches 74C on stock fan curve. I am going to try to create a more aggressive fan curve to see if that makes a difference because my GPU clock does seem to fluctuate between 1965 down to 1700s MHz during the test.

 

[KamiKamui]

For some reason my resolution is 7768x3792 and yours is 6344x3100. Perhaps it is related to that pixel count difference?
My CPU stays around 4.9GHz 62C running the test and my GPU reaches 74C on stock fan curve. I am going to try to create a more aggressive fan curve to see if that makes a difference because my GPU clock does seem to fluctuate between 1965 down to 1700s MHz during the test.

Mystery solved thanks to a hint from @WanderingWomble. I set my resolution in SteamVR to 100% for VaM and now the FPS scaling makes sense. I updated my original post with the new VR Benchmark. Went from 38.68 to 47.68 avg FPS, compared to the 3090's 51.93 FPS.

 

[WanderingWomble]

Mystery solved thanks to a hint from @WanderingWomble. I set my resolution in SteamVR to 100% for VaM and now the FPS scaling makes sense. I updated my original post with the new VR Benchmark. Went from 38.68 to 47.68 avg FPS, compared to the 3090's 51.93 FPS.

Nice!

 

[geo_gan]

I thought my computer was a beast, but it is apparently no match for VaM while running the HP Reverb G2...

Notes:
-RAM is clocked to 3600MHz and is CL16.
-CPU not running OC. FCLK is set to 1800MHz (My CPU's infinity fabric).
-GPU is the Asus ROG Strix OC model. Factory OC.

Crazy resolution there on that G2 benchmark - no wonder the 3080Ti is having such a hard time - I'm surprised it got any decent framerates at all. This is why the mainstream headsets like from Facebook have to keep the headset resolution way lower than what the G2 is at - because 95% of people can't afford to pay for a card powerful enough to run it fast enough!

 

[WanderingWomble]

Crazy resolution there on that G2 benchmark - no wonder the 3080Ti is having such a hard time - I'm surprised it got any decent framerates at all. This is why the mainstream headsets like from Facebook have to keep the headset resolution way lower than what the G2 is at - because 95% of people can't afford to pay for a card powerful enough to run it fast enough!

we have the tech to enjoy the crazy high resolution without paying for every pixel to be rendered. with the fresnel lenses many headsets use the centre is nice and clear but the edges are blurry. so, if it's blurry, why bother rendering it perfectly? this is where Nvidias VRSS helps as it only applies the expensive anti-aliasing to the centre of the image, and it can dynamically change the size based on how much performance is spare. sounds great huh? after 2 years there are only about 30 games on the list as the developers have to jump through hoops to get a game approved and Nvidia patched out a manual way to override that list in a driver update. le sigh.

anyway, moving on eye tracking can also help by telling the GPU what area of the screen the user is looking at to employ similar techniques. the commercial vario headset with the 2 embedded screen, one with a higher pixel density than the eye can see is only 1080p pixels large, so most modern cards can push that. the problem is still pushing the raw pixel count for the rest of the screen.

so the future is looking good, both in clarity and framerate. right now however it kinda sucks, the best GPU on the market can't keep up with the clearest headsets.

 

[keiseyku]

For fun I ran this on my travel laptop in desktop modes only using an i5-8250u and an MX150 with 8gb ram. I changed settings for the second test. GPU was at 100~% load during both tests

First test:

Second Test:

I'll be trying this again on other setups.

 

[keiseyku]

Here is my secondary computer with an i7-10700k and 1660 Ti:

Didn't perform as well despite it being in desktop mode but considering it's a 1660 Ti-- I'm wondering if that was it's weak point. I doubt ram has anything to do with it. GPU was at 100~% load and CPU at 20~% no OC.

 

[TToby]

I once wrote something about supersampling

Some warnings on using the VaM Supersampling slider.

As many of you will know, the image that your graphic card is sending to your VR headset has most of the time a bigger resolution than the physical resolution of your headset screens. This has technical reasons, like for removing the distortions caused by the lenses, and so on. How much it...

hub.virtamate.com

Every headset brand defines what is 100% resolution for them. It is not neccessarily the resolution of the panel(s). For a Valve Index this is ca 2016x2240, which is roughly 1.4 times higher than the visible panel resolution of 1440x1600.
In addition to this, SteamVR automatically sets or suggests what is the standard resolution/supersampling according to your GPU. In my case with a 6900 it is automatically set to 150% resolution. For this testings I had to manually lower it down to 100%. The benchmark sets the VaM super sampling to 100%, but cant set the hmd settings outside of VaM.
So, please don't be confused about VR resolution, because thos values can only be compared within the same headset brand.
P.S.: I found the reason for the 1017 desktop resolution... for me it is the taskbar being set to always visible.