VR
Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.
You are currently reading a thread in /g/ - Technology
You are currently reading a thread in /g/ - Technology
When you go from current VR displays ~2K to the 16K needed for "retina" level resolution, the number of pixels increases by a factor of 64. Will we see GPU performance scale by 64x in our lifetime?
I think its far easier to make games not so "realistic", graphics at least.
>>53802074
I think it's more likely that, before we reach that point, we'll just figure out a good way of adapting the render resolution to the retina's orientation.
Reminder, we don't need “retina” resolution across the entire eye, we only need it for about 1° of the eye's FoV.
For peripheral vision you could easily render at a thousandth of the scale without losing any information.
>>53802074
yes without question
>>53802074
Sure. The problem is though that it doesn't really scale like that. The more pixels you render, the more details you need to render, otherwise it will look absurd. Also, the limiting factor for 16k is not really sheer performance of the GPU, but throughput of every interface.
>>53802100
right, but the ultimate dream is to get a realistic (photoreal) 3D environment like in The Matrix
>>53802111
foveated rendering
does anyone have a promising eye tracking solution?
>>53802111
>I think it's more likely that, before we reach that point, we'll just figure out a good way of adapting the render resolution to the retina's orientation.
No, we won't. The eye far to sensitive to changes, ironically ESPECIALLY in the peripheral area. So the change of details needs to be smooth and fast (REALLY fast). In order to do that you basically need to render the whole scene anyway, otherwise you will have aliasing effects that are extremely irritating.
It's a good idea that many people like to feel smart around here, but it's not feasible realistically, not at all.
>>53802145
> REALLY fast
temporal resolution != spacial resolution
>>53802111
To back this up: Normal (daytime) vision is primarily done through cone cells, of which the retina has about 6 million in total.
This means that, as a hard upper biological limit, you would only need the capacity to compute 6 million distinct points of light to fully recreate the normal daytime vision of a human being.
To put that in perspective, that's about as many as a single 1080p display.
The difference between a 1080p display and the human retina is that the points are arranged very differently. A normal display has pixels spaced out equally on a rectangular matrix, whereas the retina has the vast majority of them packed tightly into a region of only a few degrees (~0.15mm radius on the retina).
Our current GPUs are basically throwing away like 99% of the work. Sometimes you can't get away from this (e.g. when you're recording game footage to present to an audience later), but HMDs are literally tied to a single user's eyes.
>>53802160
Your point is?
>>53802143
>>53802145
Already being done
http://uploadvr.com/smi-eye-tracking-foveated-rendering-exclusive/
>>53802176
the point is that your argument is wrong
>>53802164
we are definitely wasting a shit-ton of bandwidth by rendering on external rectangular displays. how long before we hook the signal directly to our optic nerve? :) seriously, direct neural stimulation is when we'll see real VR!
>>53802194
I do not believe that this is working correctly, for so many reasons.
>>53802197
Go fuck yourself
>>53802074
Had both a Rift and a vive on my head. Screen door effect is to pronounced for me
>>53802318
same here. gonna wait for 4k HMDs. hoping that in a couple of years nvidia's volta will be able to sustain a minimum 90 fps on AAA games.
Game play is better than graphics
>>53802318
I assume you mean Rift CV1 and Vive Pre or consumer version?
I've tried both. The SDE is actually very very minimal compared to the Rift DK1 and DK2. The resolution is still a bit low but it's playable.
>>53802383
Different people have different sensitivities to stuff like that. Stop trying to force whatever you like on other people by talking about it as if it wasn't completely subjective. Especially people with a lot of experience with visuals and 3d will be much more sensitive to the limitations of VR devices.
>>53802383
Rift was the CV1, friend backed it and got it. Screen door was to much for me. As for the Vive, couldn't tell you what version, it was 2 weeks ago and someone at the pen and paper shop by me had one set up. Screen door effect was obvious to me but i do admit i'm kinda sensitive to low res stuff. All I got to play was tech demos and mini games, most "game" thing I got was this kinda hack and slash game on the Rift, seemed okay but it seems like it would of worked fine with a monitor.
>>53802496
Are you sure you're actually talking about SDE and not just resolution? SDE is related to pixel fill (amount of space between pixels), not the resolution itself.
>>53802074
Probably, but it'd be a meme
They'd be better off adding eye tracking in some way, don't need no meme resolution in your peripheral vision
>>53802521
I couldn't say, i'm not super up on the tech since i'm not interested in it, well not at least till it's all sorted out. It just looked to me like it was to low of a resolution for something that close to my face. With that effect I couldn't get myself into anything, will say I didn't get motion sick even though on the Vive even though they loaded something up that was supposed to make you sick. Coolest thing I saw was on the Vive, it was a game top simulator of some sort. The idea being you could any game you wanted. While neat I would just rather play real D&D or something with friends around me.
>>53802521
> SDE is related to pixel fill (amount of space between pixels), not the resolution itself.
i never realized that. good point.
>>53802074
Foviated Rendering allows VR (current gen) to be less graphically demanding than running a game at 1080p.
Once we get eye tracking (gen2 VR), needing a gaming PC will no longer be a thing.
>>53802318
>>53802496
Sucks to be you. The vast majority of people say its a non-issue after the first 10 seconds, as you completely forget its there.
>>53802911
Few places I googled say it is an issue, how much of it differs from person to person. Also why so aggressive? You work for facebook or something?
>>53802935
It's some kind of issue with most VR enthusiasts. They don't get that not everyone is some impressionable neckbeard that dreams about having sex with animu characters or whatever the hell they plan to do. Pretty annoying, not only on /g/.
>>53802935
I don't think you've been here that long if you think that post is aggressive.
The resolution isn't great but I think that really depends on the game more than by person. It's good enough that the majority of people can get lost in the world despite the res. Bigger problem are things like sims that have you look or try to focus into the distance and the detail just isn't there for something like that.
>>53802957
I think it's neckbeards that focus on the faults if anything. Normal people don't care about the resolution at all.
Not related to rendering, more just a VR general comment, a while back there was a discussion about making a VR headset with an iPhone 4s display because they where so cheap, but with an extra $40 in HDMI to mipi hardware needed it wasn't really worth it.
I just realised that you can get a Zenfone 2's screen, lcd, digitizer and all, for >$20 us.
That's 5.5" at 1080p.
Add $60 for a custom mipi to hdmi + gyro board, $20 for a case and you have a >$100 DK2 equivalent.
Just a thought.
>>53802993
>Zenfone 2
>IPS screen
Not that simple. This is basically a RiftUp DK1 equivalent. Not a DK2.
You have no positional tracking or low persistence. You're going to have guaranteed headaches without the latter.
>>53802993
Yeah the displays aren't terribly expensive. All the R&D is probably expensive but give it a generation or two and VR should be much cheaper
>>53802135
when devs forgo art direction and style in favour of generically realistic graphical style, this will be a bad thing. of course its only AAA devs that will do this but still.
>>53802074
>16K
>screen is 2 inches away from face
I'm sure we'll have pixels or LEDs or whatever the fuck that are smaller than atoms.
>>53802074
>~2K to the 16K needed
2K = x
>x*2 = 2x
>2x*2 = 4x
>4x*2 = 8x
>8x*2=16x
>16x*2=32x
>32x*2 = 64x
Are you dying withing the next 10 years?
>>53802074
>Will we see GPU performance scale by 64x in our lifetime?
Why wouldn't we?
>>53804799
Assume we have a VR headset display. It's a square screen with a 16384x16384 display and 4 inches diagonal. If you do the math, you get a dotpitch of 0.0044mm. A cesium atom (the largest known atoms) is 300pm in diameter. That's 0.0000003mm.
>>53804216
They're not mutually exclusive.
>>53804216
i want my holodeck dammit!
>>53806719
> 10 years
kek!
You may not have heard that Moore's Law is coming to a screeching halt.
>>53802074
VR with goggles (or any other kind of display for that matter) is a deadend. Similarly, real physical interaction (like walking on some treadmill or plank or what not) is also doomed to fail.
>>53802074
>Will we see GPU performance scale by 64x in our lifetime?
Yes, but it's going to take a while. Moore's law is dead, we only have 3 more cycles left before carbon memetubes become necessary to reduce the size of chips further and after that only 6 or 7 more cycles until they're using transistors the size of a single atom.
>>53810330
can't we refine the manufacturing process if it's stuck at like 7nm and then start making bigger chips or is the square cube law thing stopping us from doing that?
>>53802074
More like you will see transfer speeds increase in your life time and you will rent/lease processing power from a company.
>>53810388
You're still going to be limited in how large you can make a single core without having to reduce the speed that it would run at and would have to deal with increased power consumption. It would be the same as companies going to dual GPU cards as the standard nowdays to increase power.
Given a rough interpretation of Moore's Law, a 64x increase in computational power will be achieved in around nine years.
>>53812940
Moore's Law is dead
>>53802074
that would be only 12 years, because GPU still follows Moore's rule of thumb. also foveated rendering will decrease the needed rendering at least 32x over the usual 110 FOV.
>>53802074
>>53812940
Moores law is dead though
AMD has already made it clear that multiple smaller inexpensive GPUs is the future, as opposed to one large GPU
Also low level APis and multi core utilization
Yes it will happen in the next 10 years not a lifetime.
Think 10 years ago we were on 110nm fab with 7800gt cards which came with 256mb of vram
The famed 8800gt is only 8~ years old now
Current cards are about 40 times as powerful as cards 10 years ago and with Moore's law its exponential
Ray tracing will become a trend in the next 3-5 years and we wont have to worry about it.
>>53813163
>Ray tracing will become a trend in the next 3-5 years
no
>>53810417
latency requirements won't make this approach workable for VR
>>53813163
are you posting from 1997? greetings time traveling friend!
>>53807214
because Moore's Law begin to slow down in die shrinking?
>>53815916
Between things like graphene and photonics computers there's a pretty good chance we'll see another huge leap in processing power within the next 50-60 years most of us have.
>>53812972
More's law isn't dead for GPUs. We're still seeing strong performance gains on GPUs compared to CPUs as the "moar cores" model actually works. Video processing is highly parallel compared to the general-purpose processing done on a CPU.
