New to the forums? Click here to read the "How To" Guide.

Developer? Click here to go to the Developer Forums.

PC VR - My Future Projections for 4K - 8k - 16k

SlaghtonSlaghton Posts: 47
Brain Burst
edited September 2015 in General
So earlier today I was browsing threads and I came across one asking what he would need for 16k vr gaming. He posted a link with AMD stating that 16k resolution (likely per eye) is whats needed to replicate retinal resolution. I had time to kill so I thought I'd do some math and rough estimating.

(I might try to make a tldr version but then I leave out a lot of data benchmarks. If I had more benchmark data I could make better estimations for future fps scaling) The tables got screwed up. I'll have to edit.

Lazy TLDR:
Current software,drivers.
Low Quality 2k x 2k per eye (4k total) vr gaming doable with 4x 7970's in crossfire for a game similar to Bioshock Infinite

Low Quality 4k x 4k per eye (8k total) vr gaming doable with 4x 980 ti's in sli for a game similar to Bioshock Infinite. (I need more data to confirm. I can't find enough people doing 4k benchmarks with their 980 ti in lowest settings.)

Low Quality 8k x 8k per eye (16k total) vr gaming doable with 4x theoretical N500's (see bottom thread) in perfect sli for a game similar to Bioshock Infinite in year 2020.




The game i'll use for my benchmark examples will be Bioshock Infinite because it was a decent looking game that was not heavily cpu dependent. Using a 7970 right off the bat the game gets about 294 fps at a resolution 1920x1080 low quality settings. Now i'm going to scale this all the way up to a 16k resolution.

7970
Bioshock Infinite Low Quality
Res
pixels
frames
difference fps
Difference res
1920x1080
2073600
294
nul
null
2560x1440
3686400
178
60.54421768707483%---56.25%+

Below is projected numbers based on above data.

3840x2160
8294400
85.03594
52.227%
55.56%+
7680x4320
33177600
22.853408875
73.125%
75%+
15360x8640
132710400
6.1432
73.11910869139429%---75%+

As you can see, with 1 7970 at lowest settings you could achieve non vr 4k(ignoring vram) but after that the fps plummets below acceptable thresholds. At ultra settings + DOF with a 7970 at 2560x1440 you will get an avg of 42fps compared to 178 with low settings. At 4k resolution with ultra settings you would net about 20 theoretical fps (compared to the similar 280x that gets 22 fps real benchmark data).

In quad crossfire 7970 at 7680x4320,lowest settings, and with 100% performance ratio per added card you get 91 theoretical fps. I use 100% because i've seen benches of cards getting almost double performance in sli with current software/drivers. That's 4k x 4k per eye for vr on lowest settings with last gens hardware. Optimized for VR you could possibly see higher fps but low settings already cuts out a lot of stuff. I believe 90 fps is the current target for CV1 so that would net you 4k VR gaming though there would be dips since the avg fps is 91 but the lowest possible fps based off benchmarks could put you at nauseating 71fps dips during action.

So, with a 7970 x4 + directx12 + low settings + perfect crossfire performance ratio's per card you could achieve somehwat playable 4k vr gaming in non intensive scenes. 2k x 2k would be a safe bet.


Now lets talk 8k - 16k and moving over to the green side for these next benchmarks cause i'm not biased :P. Over the past years for Nvidia, the performance increases avg out to around 25%+ every year over the past 6 years for their gpus. I believe there is diminishing returns showing but i'm just going to be hopeful and do these next projections at 25% increase per year on avg for the next 5 years. (It'll likely be less unless we see some nice boosts from going to 14nm-16nm,revamped architecture, etc)


Bioshock Infinite Benchmarks at 1920x1080 Ultra Settings.

Year
Series Fps
%Change
Release Date
2010S----480----38
March 26, 2010
2010S----580----55
0.3090909090909091
November 9, 2010
2012
680----76
0.2763157894736842
March 22, 2012
2013S----780----94
0.1991570073761855
May 2013
2013S----780ti--120
0.2091666666666667
November 7, 2013
2014
980---137
0.1240875912408759
Sep 2014
2015
980ti--166
0.1746987951807229
Jun 2015

Projected at a 25% increase per year.
2016
N100----207
0.1980676328502415
Jun 2016
2017
N200----259
0.2007722007722008
Jun 2017
2018
N300----324
0.2006172839506173
Jun 2018
2019
N400----405
0.2
Jun 2019
2020
N500----506
0.199604743083004
Jun 2020



Bioshock Infinite - Ultra Settings + DOF
980 Ti
Real bench data
1920x1080 - 166 avg fps
3840x2160 - 52.46fps

At 3840x2160 with 4 980 ti's in a realistic sli scenario you would achieve almost 90 fps. (2nd card 40%+, 3rd +20%+, 4th +5%) That's 2k x 2k VR gaming with Ultra settings. Considering the frames that real bench data showed earlier for a 7970 at lowest settings which is more suited for vr in its infancy you could be looking at 3-4x performance increase at those settings. Its almost safe to say 4k x 4k vr gaming would be doable with 4 980 ti's with current sli performance gains + low quality. If software/driver improvements + directx12 and other methods could be implemented to increase sli % + per card to a 1 to 1 ratio you could just achieve 8k x 8k at this point.


16k VR Resolution

Bioshock Infinite Ultra settings
N500
Quad Sli low settings Perfect Sli

Projected Bench Data
Resolution
Pixels
Fps
1920x1080
2073600
2024
2560x1440
3686400
885.5
3840x2160
8294400
491.95
7680x4320
33177600
122.99
15360x8640
132710400
30.7475

In the year 2020 with the projected Nvidia N500 GPU in Quad SLI with a perfect 1 to 1 performance ratio for each extra card added you would be able to achieve a total 8k resolution (4k x 4k per eye). Today's sli limitations would put 8k closer to 54 fps instead of the perfect sli 122.99 fps. Lowering settings to lowest would achieve 3-4x increase to 150-216 fps with the realistic starting fps being 54 . 16k resolution at low settings at perfect sli would be about 54fps abest realistic scenario. Perfect scenario would be 92.25-123 which would be 16k gaming (8k x 8k per eye) at lowest settings in a game not designed for vr.

Well I tried to keep all my thoughts coherent but i'll probably make a quick tldr at the top summarizing everything I typed out. If we can get perfect sli/crossfire scaling with other goodies by 2020 we could see 16k VR in low settings or 8k in higher.

EDIT: I left out foveated rendering but if this was implemented we could definitely see 16k VR by 2020 if we had the screens. Five years is a short time with screens just reaching 4k now so 8k is what I expect around then.
Ordered: November 9
Processing: 19
Shipped: ???
Delivered: 25

Comments

  • madmossymadmossy Posts: 7
    Brain Burst
    Couple of things, the introduction of HBM and HBM2 will drastically improve throughput on a GPU, coupled with the introduction of DX12 which will essentially cause SLI/XFIRE to appear as 1 GPU will also aid in better performance in multi GPU setups, Resulting in a near perfect scaling as you add extra GPU's to the system.

    I would bet money on GPU's being able to output 8k at a steady VR suitable frame rate within the next 3 years, the only question that really remains is.

    Will the bandwidth be able to keep up, probably not, that's where the real issues remain and its not something that can be done in a short period of time.
  • MrMonkeybatMrMonkeybat Posts: 640
    Brain Burst
    Something that could make higher resolutions much easier is eyetracked foveated rendering. http://research.microsoft.com/pubs/1766 ... inal15.pdf
  • jyounjyoun Posts: 232
    Is eye-tracking fast and consistent enough now though? I agree they could at least do foveated rendering... the peripheral is so low res compared to on-axis.
  • jyounjyoun Posts: 232
    Foveated rendering in Unity, 3x faster:
    https://www.youtube.com/watch?v=GKR8tM28NnQ

    Unreal folks talk about it here:
    http://www.dailymotion.com/video/x1ehtw ... ering_tech
  • MrMonkeybatMrMonkeybat Posts: 640
    Brain Burst
    jyoun wrote:
    Is eye-tracking fast and consistent enough now though? I agree they could at least do foveated rendering... the peripheral is so low res compared to on-axis.
    According to Microsoft Research no one could tell there set up from the control. The question is whether eye trackers of that quality can be be made cheap a small enough to be put in an HMD. And at what angular resolution does foveated make enough savings compared to the cost.
  • jyounjyoun Posts: 232
    Right, but without eye tracking the technique is still useful due to our current inability to focus on the edges of the screen.

    Also, IRL move your eyes to some object near you. Notice It takes a little bit of time for it to come into focus. Eye-tracking similarly doesn't need to "adjust" so quickly. When sensing eye movement you could render everything lower until movement stops, then fades in the higher resolution focus point.

    Future projections like the one in the OP doesn't seem to take any techniques like this into account, so it's pretty conservative/worse-case-scenario.
Sign In or Register to comment.