Slashdot Mirror
Google To Reveal 'World's Highest Resolution OLED-On-Glass Display' For VR Headsets (roadtovr.com)
An anonymous reader writes: Last year at SID Display Week 2017, Google's VP of VR/AR teased a "secret project" that the company was working on -- a VR-optimized OLED panel capable of 20 megapixels per eye -- which was being undertaken with "one of the leading OLED manufacturers." This year, the schedule for SID Display Week 2018 indicates that Google plans to reveal its made-for-VR panel on May 22nd, which it calls the "world's highest resolution (18 megapixel, 1443 ppi) OLED-on-glass display." The company plans to detail the display in a presentation at the event, which will be co-presented with engineers from LG, suggesting the identity of the second partner on the project. Ideal for VR, the 4.3-inch panel is capable of 120Hz refresh rate and is expected to have a resolution of some 5,500 by 3,000, representing a massive leap over today's leading VR panels which offer 1,600 by 1,440 resolutions at 90Hz.
I've seen way too much evidence of Google's seemingly psychotic behavior with respect to committing to a project and then killing that project off like it was a cockroach.
When a credible company brings such a product to market, I'll consider it. In the mean time, I'm quite certain I will live very well indeed without this stuff.
Is a mid sized desktop display maybe 22" with a ratio of 16:10 and 4K resolution. IBM made such a device 15 years ago but you can't get anything close today.
You are an IDIOT.
All I did was state a preference based on Google's past behavior, and you conclude that I am an idiot ?
At this point the evidence is far more overwhelming that it is YOU who are the idiot.
I would buy a VR headset with that! It might solve the motion sickness and eyestrain issues with typical VR headsets.
Since everything that could be used to drive these panels will be mining coins, nobody can use these at home.
" No one is going to prison, and we also don't golf " -Trump / Ponce 2020
Current VR displays cover about a 200 degree field of view. 20/20 vision is defined as the ability to distinguish a line pair spaced 1 arc-minute apart, so 2 pixels per arc-minute. So this corresponds to (200 degrees) * (60 arc-minutes/degree) * (2 pixels/arc-minute) = 24,000 pixels. You need a display that's 24,000 pixels wide for it to display a 200 degree field of view and have the individual pixels not be discernible to the eye. So this display will be a little more than 1/5 of the way there.
Put another way, the angular resolution of this new VR headset will be (5500 pixels) / (200 degrees) = 27.5 pixels per degree. That's about the angular resolution of a 50" 1080p HDTV viewed from 31" away. Or a 24" 1080p monitor viewed from 15" away. The pixels will still be painfully obvious.
Which is about the date hardware will be around to actually run a 5k+ VR headset...
I can just barely see the pixels in the Sony Playstation VR headset, which is 1080p, especially when moving my head slowly or holding it at a slight angle.
A 4K screen would probably be overkill for such a small problem (3840 x 2160 = 8.2 Mpixels).
A 20 Mpixel screen would just be a waste of technology - and money, no doubt.
A Sony Playstation VR headset has about 90 degrees FOV. To cover your entire field of view, you need at least 180 degrees in all directions, meaning at least 4x the number of pixels. But you need several times HD resolution to match human acuity. 18 Mpixels is probably still not quite enough, though it will be a huge improvement over what we have.
Good luck getting the GPU hardware powerful enough to drive it; let alone a method to deliver it.
I expect that we'll never actually see headsets with panels this high-resolution. By the time video cards can drive VR software at this resolution, we'll have laser retinal displays. The main problem holding these back in the past were rainbow artifacts appearing during fast eye saccades, but waveguides have recently been devised which should prevent this (can't find a source ATM).
Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
I can just barely see the pixels in the Sony Playstation VR headset, which is 1080p, especially when moving my head slowly or holding it at a slight angle.
A 4K screen would probably be overkill for such a small problem (3840 x 2160 = 8.2 Mpixels).
A 20 Mpixel screen would just be a waste of technology - and money, no doubt.
The only metric that matters in VR displays is Pixels Per Degree of arc (PPD) with 60 being very roughly limit of what people can see.
PSVR has a crummy 100 degree FOV yielding following PPD in each resolution category.
PSVR = 14.45
4k = 44
20MP = 64
If you were to increase FOV to 180 to better match vision.
PSVR = 8
4k = 24.5
20MP = 35.55
Not only is 20MP not a waste of technology it's not nearly enough. It's less than half resolution of an iPhone display held at a distance of 1 FT.
first: PSVR uses a different pixel layout to minimize screen door effect, but at the cost of resolution.
second: Resolution is a very important next step in VR: not only is it bound to reduce SDE further and also allows for bigger FOV; it allows us to see detail farther away than just a few meters. At the moment, racing games and flight sims suffer the most from this gen of VR because of that.
great you have those, but both have crappy lenses, and not the best displays. The PSVR has a completely different display and different lenses, which makes it, even at lower resolution, better than the Vive. (but ofcourse the vive does have other plusses).. the 20MP (per eye) is really overkill even for VR, let's not forget, current highend GPU's already have trouble even pushing the current VR displays with nice visuals, even with foveated rendering a 1080 won't be enough to power even one display, let alone two (with the visuals people tend to expect using such a resolution).
But the question and problem is, how many pixels CAN the human eye see? Just like 4K televisions you won't be able to see them all.
- Depends on how wide those pixels are spread. More precisely, which angle of your view they each occupy.
The 4k television is just a rectangle in front of you.
Ideal VR headset would cover nearly everything in front of your face, so no matter which direction you turn your eyes, there's still picture on the screen.
So these 5.5k pixels are going to get spread across your whole field of vision, i.e.: at least 180, or ideally a tiny bit more (from when you're looking sideways).
So overall, a VR pixels is going to look several times bigger than the pixels on the TV screen.
- Depends on which part of your field of view.
They eye's retina and brain's visual cortex (and Convolution Neural Net which are inspired by it) works by comparing neighboring signal.
So the distance between neighboring receptors will decide how fine the details you can see.
Receptor density varies a lot across the region of the eye.
Some region have very closely packed photoreceptor (fovea in the middle), so you can distinguish very fine pixels.
Worst region have receptors very wide appart, such as the blind spot (a hole in the retina where the optical nerves exits the eyeball). There you can miss giant swaths of visual space if the details fall inside the blind spot, in between photo receptors.
- Depends on the geometry of the VR optics.
Unlike older VR headset (say eMagine's 3D Visor. Or the old school VFX1 of old time), modern VR headset don't use over complicated optics to make perfect projection of the screen, but just the simplest possible stuff to keep the image in focus (and compensate in software - shader software on the GPU).
The image end up distorted. But in a pillow shape that actually compensate the above.
In the middle of the VR image (i.e.: where the high resolution eye's fovea will be looking most of the time), a lots of pixels are compressed into the image, you get a better resolution.
In the side of the VR image, the image is heavily distorted and fewer pixels are stretched over a wider field. By luck, peripheral vision, with its lower spatial resolution, would be looking there most of the time.
So whenever you're looking straight ahead, the optical properties of the VR headset compensate a bit for the variable resolution of the retina.
Basically, due to point 3, the way the pixels are stretched according to point 1 is partially compensating point 2.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Why do I need a laptop with one of these , a bluetooth keyboard/trackpad, and my existing cell phone's CPU? Sure, big battery, but my travel bag is about to get much lighter.
My God, it's Full of Source!
OUTSIDE_IP=$(dig +short my.ip @outsideip.net)
This is not false, but it is not quite true either.
How often do you look as far left as possible? Rarely. We only actively look around directly within about 30-50% of our full view capability, except in the rare cases when we're trying to look at something without others knowing (ie, side-eye). So only the central visual field needs this level of pixel density to be nearly indistinguishable from real life's resolution. The visual fidelity could drop to 1/2 or even 1/4 outside this and you'd barely notice.
Which is exactly what ends up happening on modern VR headset (everything since the Occulus) which tend to use as simple as possible optics just to keep the image in focus (as opposed to older VR headset which used more complex optics to give perfect rectangular picture floating distorsion free in front) and use software compensation (shaders on the GPU to pre-distort the image in the opposite direction).
Due to this pillow-shape distortion, more pixels are used in the middle of the image (where the eye looks most of the time) than the side of the image (where the peripheral vision is looking most of the time).
In addition, we can render even fewer pixels with eye tracking. This has already been successfully tested on current equipment with eye tracking and foveated rendering... rendering the center at full resolution, but increasingly fewer pixels per inch as you go away from the center of vision. And it already workes very well, quartering the rendering power needed. With a massive full-vision FOV, it would reduce rendering by needs by 20-40 times.
As proven by all the research done in VR since Occulus : the main drawback would be rendering speed/feedback loop.
Will this eyesight tracking work fast enough so the image rendering doesn't lag too much behind the eye motion ? (Other wise you'd be seeing blurred image when ever you're looking around fast, as the eye motion overshoots the region that was rendered at high resolution by the fovea tracking you're advocating).
Avoiding lags and blurs seems to be key to keeping the immersion realistic, and reducing risks of motion sickness.
Adding a eye-tracking/resolution feedback loops risks increasing these lags and blurs.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
.. the 20MP (per eye) is really overkill even for VR
20MP is not enough. See my previous message in this thread.
let's not forget, current highend GPU's already have trouble even pushing the current VR displays with nice visuals, even with foveated rendering a 1080 won't be enough to power even one display, let alone two (with the visuals people tend to expect using such a resolution).
With perfect eye tracking you need only render 1MP with any detail to reach practical limits of human vision. The rest can be a blurry mess.
Line in Youtube video linked in TFA about bandwidth of the optic nerve weighing in at a whopping 10mbit/s is a good way of thinking about the problem. GPU power is comparatively irrelevant.
on the other hand, stereo images is really one of the best situations for multi GPU rendering.
Once the geometry is processed (which can be somewhat balanced accross multi GPUs),
the rest of the rendering is different between both eyes (that's the whole point of parallax) with little inter-dependency.
You can simply subdivide the VR image into "two eyes, one eye per GPU"
(Though for objects far from eyes, the tesselation and rendering should be very similar / almost identical. Again that's due to the parallax).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Look up foveated rendering. You only need to render about 5% of the entire field in high resolution, the rest can be low res so the GPU requirements are not that high.
The area required to be high res will move around as your eye moves, so the display needs to be high res [almost] everywhere however.
What we really need for VR is 8000x4000 pixel screens, so the user can't see the pixels. The pixels per inch is irrelevant, since there is a lens between the screen and the eye anyway.
I've abandoned my search for truth; now I'm just looking for some useful delusions.
I do. But 4K 3D content is pretty hard to come by.
I've abandoned my search for truth; now I'm just looking for some useful delusions.
I think people underestimate the potential raw resolution that would be required to have a true full FOV display that would effectively be indistinguishable from reality (not even getting into color gamut issues). With a typical resolution of about one arcminute and a FOV around 160-175, you need around 10K x 10K, but only within the central 2-5 degrees. Some people can resolve down to 20 arcseconds or so, increasing it to 30K x 30K.
With one arcminute and 2 degrees, you only need to render at 120x120 within about a millisecond, which is less than 640x480@60, so you can easily go a bit larger and higher resolution without pushing any limits. 5 degrees and 30 arcseconds in a millisecond would be close to 2560x1440@120.
Note that you don't need to render at 1000 fps, the 1 millisecond is about the latency you'd want as you track the eye, so it needs to be fast but has lower power requirements than if it was rendering continuously.
what machine will power this? if you see the specs for the current best vr headsets, which don't come close to this, it's already a very powerful and expensive build.
On a long enough timeline, the survival rate for everyone drops to zero.
You might be able to get by with a much smaller hi-res area if it can be projected onto the retina using high speed MEMS mirrors, using the same tracking you'd already be doing.
Even if not, foveated rendering will significantly reduce GPU requirements which would enable extremely high resolutions with much wider FOV and very low latency.
Already solved... nearly two years ago. It hasn't been used yet because no mainstream headset has installed the eye tracking sensors needed.
According to the source you cite : the system works at 250Hz to avoid the fovea out-running the the rendering.
That is way much higher thant the 90 FPS target of Occulus Rift (designed so the *head-motion* doesn't out run display. Intertia, etc.).
It will take some time until : ...while keeping in mind that the above (throwing GPU power at high FPS) is in competition with other targets (throwing GPU power at more scene detail and more pixels).
- VR hardware catches up and makes screen that accept framerates beyond 250 FPS
- The GPU rendering hardware is powerful enough so that, once factoring the foveated rendering speed gains, the over all system can stay above 250 FPS, as close to 100% of time as possible.
But could still be implemented half-way (not perfect foveated rendering, but slightly reducing resolution of parts of the image that are so much far away that there isn't much risk of over shooting even at lower FPS).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]