• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

TR: The PlayStation VR2 would be better off without eye tracking

Warnen

Don't pass gaas, it is your Destiny!
If history repeats just wait for version 2 in less then a year anyways.

Sadly my vr days are done. Glasses won’t fix my issues, Getting old sucks.

I’ll miss my days with super hot.
 
Last edited:

Rudius

Member
In the teardown you can see that the modules are very small and I doubt they are the most expensive part of the headset.
sem-titulo-png.181793



Eye tracking is the most important feature of PSVR2, since it allows ports of PS5 games to VR in good quality. Developers of Gran Turismo Sport and Ace Combat 7 said they wanted to bring the full games to PSVR, but could not do so due to lack of power.

If Sony needed to cut costs a better way to do it would be to use SDR LCD screens like every other headset. It would suck compared to OLED HDR, but at least we would not lose games due to VR being much heavier to run then the same thing in flat.
 
Last edited:

Thirty7ven

Banned
Yeh, it's really weird, on one hand he says its used for foveated rendering then says Call of the Mountain only uses it for navigation. In other words, he's another one who has heard a few things, put two and two together and got five.

It’s pervasive in the games media. The amount of times I’ve stopped reading an article or quit a podcast because people don’t know the details of what they are talking about… Most of these so called professionals are more worried in engaging with their audience on Twitter and creating content than actually knowing in detail what they are talking about. Don’t give me bullshit about deadlines either, either this is what you do or it’s not.
 

Rudius

Member
That's a dumb fucking take

The cameras for eye tracking are peanuts compared to optics-diplays. That's why Quest 3 is rumored to still sport LCD panels, but with pancake lenses at least. Will probably be higher priced than Quest 2, anything tech nowadays are $$$.

I actually wish Sony had even made it more expensive to have wireless as an option (and support PCVR damnit)
Exactly.

And eye-tracking is more important for PSVR2 then for Quest 3 for this reason: even though Quest 3 will be much weaker, it is a platform in itself, so the games can be made to whatever power is available; PSVR2, on the other hand, is part of the PS5 ecosystem, so it is important for it to be able to get ports of flat games. In my opinion the two biggest games it has are ports: GT7 and RE Village.
 
Failed at doing what?
Failing at making foveated rendering with eye tracking work apparently. He doesn’t believe in it, because he didn’t see the massive performance gains. Yet here we are, as genius as he sometimes can be, he’s proven wrong. Apparently foveated rendering with eye tracking works for Sony with a massive 3.5 times faster render times if done well.
 
Exactly.

And eye-tracking is more important for PSVR2 then for Quest 3 for this reason: even though Quest 3 will be much weaker, it is a platform in itself, so the games can be made to whatever power is available; PSVR2, on the other hand, is part of the PS5 ecosystem, so it is important for it to be able to get ports of flat games. In my opinion the two biggest games it has are ports: GT7 and RE Village.
Also, funny that all those people are talking about pancake lenses being the next big thing. Do they just ignore the fact that you lose 80% of your light from the screen vs Fresnel lenses? 80 f*cking percent. (Gotta admit the headset itself looks cooler and smaller but who cares if you’re the actual user, you don’t see the headset yourself when you’re using it).

And I love those comments about: “a shame psvr has fresnel lenses” bla bla bla “God Rays and other negative things”. No reviewer though saw them, because they’re not there. They even tried to see them but they’re not there. And why not? Watch the mini documentary on how Sony created custom fresnel lenses just for this device.

So not only they have 100% light from the oled hdr screen, but also mitigated common fresnel lenses shortcomings. But hey, informing oneself seems to be too difficult nowadays so you just have to copy biased media outlets that don’t even have their facts straight.
 
Last edited:

Rudius

Member
Isn't it true that there are some plus sides to the lenses besides cost? Don't they help with the HDR or color or something? I feel like I read there's some benefit alongside the OLED HDR-ness besides just being cheap (and not as good). Either way, it's still a pretty good package/step up from the prior model - from everything I've read.
Pancake lenses block some 80% of the light (if not more). If PSVR2 used it the HDR highlights would look as dim as a weak SDR screen. I think some colors would suffer too, but I'm not sure about that.

Also, as far as I know, there are no Pancake lenses as large as the fresnel on PSVR2, so the FoV would probably be smaller, although the lenses on Pico 4 appear to be close in size (and people complain about it being too dim).
 
Last edited:

midnightAI

Member
Failing at making foveated rendering with eye tracking work apparently. He doesn’t believe in it, because he didn’t see the massive performance gains. Yet here we are, as genius as he sometimes can be, he’s proven wrong. Apparently foveated rendering with eye tracking works for Sony with a massive 3.5 times faster render times if done well.
Well, to be fair, when talking about performance he was comparing it to Fixed Foveated Rendering, and then you arent getting massive gains in overall performance, however, the visual detail is vastly improved across the display which could also be considered when measuring performance because to do the same without ETFR you'd pretty much have to turn Fixed Foveated Rendering off.

One thing I am confused about though is him saying that using ETFR would cause the screen to be blurry because our eyes move so fast. I'm not sure if I am misreading that or not understanding what he is saying, but you don't see it being blurry because of eye saccade do long as the round trip is fast enough (our eyes need to refocus whenever they move which takes time). Now I'm not sure if this is because they haven't got it working perfectly when he worked at Meta (I have heard that the eye tracking tech in the Quest Pro is only at 30hz vs 120hz on PSVR2) or whether he is talking about something else.
 

midnightAI

Member
Pancake lenses block some 80% of the light (if not more). If PSVR2 used it the HDR highlights would look as dim as a weak SDR screen. I think some colors would suffer too, but I'm not sure about that.

Also, as far as I know, there are no Pancake lenses as large as the fresnel on PSVR2, so the FoV would probably be smaller, although the lenses on Pico 4 apear to be close in size (and people complain about it being too dim).
I think the max at the moment is around 90ish degrees FOV.

Another issue I believe is ghosting, again due to the way light bounces around inside the lens.

It's about choosing the right type of lens for what they are tryng to achieve. Sony wasn't too bothered about making the headset smaller and lighter (its already comfortable and light), instead they opted for fresnel as they wanted the best colour and brightness for HDR. Stand alone mobile like VR will want to use pancake because the headsets can be smaller and lighter. Ideally you'd want both but currently that isnt possible but they'll crack it in the coming years.
 
Well, to be fair, when talking about performance he was comparing it to Fixed Foveated Rendering, and then you arent getting massive gains in overall performance, however, the visual detail is vastly improved across the display which could also be considered when measuring performance because to do the same without ETFR you'd pretty much have to turn Fixed Foveated Rendering off.

One thing I am confused about though is him saying that using ETFR would cause the screen to be blurry because our eyes move so fast. I'm not sure if I am misreading that or not understanding what he is saying, but you don't see it being blurry because of eye saccade do long as the round trip is fast enough (our eyes need to refocus whenever they move which takes time). Now I'm not sure if this is because they haven't got it working perfectly when he worked at Meta (I have heard that the eye tracking tech in the Quest Pro is only at 30hz vs 120hz on PSVR2) or whether he is talking about something else.
Eye tracking is 240hz on PSVR2. If 30hz is accurate for the Quest Pro I understand why the image would get blurry when moving the eyes quickly. It simply can not keep up deciding render location with such a low speed.
 
I think the max at the moment is around 90ish degrees FOV.

Another issue I believe is ghosting, again due to the way light bounces around inside the lens.

It's about choosing the right type of lens for what they are tryng to achieve. Sony wasn't too bothered about making the headset smaller and lighter (its already comfortable and light), instead they opted for fresnel as they wanted the best colour and brightness for HDR. Stand alone mobile like VR will want to use pancake because the headsets can be smaller and lighter. Ideally you'd want both but currently that isnt possible but they'll crack it in the coming years.
I’m sure PSVR3 and other headsets will have something vastly superior than now in another 6 years 😃
 

Rudius

Member
Well, we can be sure that Sony's engineers know better than us what the best possible trades are. And since reality is what matters, since reality is that all reviewers are finding PSVR2 image quality outstanding, that's what matters.
About that, one thing that is not appreciated about Fresnel lenses are God rays, but that has been almost completely resolved by Sony's Fresnel patent. Also, Fresnel lenses are significantly better than pancake in the amount of light that they allow to pass, allowing the use of OLED screens, traditionally less bright than LCDs. One more thing is that Fresnel lenses cause less pupil swim, which is one of the causes for motion sickness.
Those are things that in VR are significantly more important that avoiding a negligible amount of MURA in some dim situations (and that you notice only if you look for it, or if you are particularly sensible to it). Perfect lenses maybe don't exist, and the chice depends by what you want to realize and the overall cost. One sure thing is that all the headset has been engineered in order to have the best possible experience at the cheaper cost possible, not to use various high tech components only to ruin everything using bad lenses.
They have a small sweet spot for the eyes, so to be able to see the whole image in focus, it's necessary to be sure to set everything correctly, but once it's done, the image is more than great. And luckily, the setup process is very simple and quick right thanks to eye tracking.
Take a look at this video of all the possible controllers they tested before creating the Dualshock 4.

ps4-controller-prototypes-8982374979847037094-1200x676.jpg



Member this one?
ps4_controller_orbis.0.jpg



For PSVR2 they almost certainly tried different screens (OLED, LCD, mini LED LCD), lenses (fresnel, pancake, aspheric), different resolutions, eye-traking on and off, FoV etc., before settling for what was the best bang for the buck as an overall package.
 

midnightAI

Member
Eye tracking is 240hz on PSVR2. If 30hz is accurate for the Quest Pro I understand why the image would get blurry when moving the eyes quickly. It simply can not keep up deciding render location with such a low speed.
Wow, even better then :)

The major benefit of Quest though for eye tracking would be that it doesn't need to go down a wire (or wireless) if its all done within the headset, still, speed of the eye tracking is very important, why they dont just use Tobii, who figured this out ages ago, I do not know (cost maybe? which would be strange seeing as their headset already costs $1500, sorry, $1100)
 

ABnormal

Member
It’s pervasive in the games media. The amount of times I’ve stopped reading an article or quit a podcast because people don’t know the details of what they are talking about… Most of these so called professionals are more worried in engaging with their audience on Twitter and creating content than actually knowing in detail what they are talking about. Don’t give me bullshit about deadlines either, either this is what you do or it’s not.
Or even worse, when it's plain obvious that there's an attempt to create some drama and stir discussions with the primary purpose of gaining clicks, rather than being informative. When I see that kind of lack of dignity, I just cut off the source. It doesn't come from passion, it's just bad marketing.
 

Soosa

Banned
Stupid bastard.

So without eye tracking it would cost what, 599€ instead of 639€?

To make the price significantly cheaper, it would have to be like 399€ max.

600€/$ is not much, people should stop crying and save money or stop buying shit they cant afford.
 

ABnormal

Member
Eye tracking is 240hz on PSVR2. If 30hz is accurate for the Quest Pro I understand why the image would get blurry when moving the eyes quickly. It simply can not keep up deciding render location with such a low speed.
30 Hz eye tracking is non enough for foveated rendering. To be enough it has ALWAYS to precede the starting point of the frame rendering pipeline, and moreover, for VR to be comfortable, the latency between head tracking and the actual display of the corresponding image must be within 20 milliseconds, otherwise it starts to create discomfort. 60 fps rendering time is around 16 milliseconds, and PSVR 2 has to be able to use foveated rendering for games that are 90 or 129 frames per second native (and the last ones have 8 millisecond frame rendering, so eye tracking must be quicker than that).
30 Hz eye tracking can be used only to control menùs or similar things.
 

SF Kosmo

Al Jazeera Special Reporter
You can't refute a real-world application with an older, theoretical projection. The fact is eye-tracking in PSVR2 does work and does seem to offer some real performance gains. I don't think adapting RE8 into VR at the fidelity they have would have been possible otherwise.

And that latter point is really key, and I think it's the reason why this was a must-have feature for Sony. Sony realizes that, if VR is going to survive and grow and be commercially viable, it's going to need to leverage cross-format game; games that support both VR and non-VR versions. Being able to lower the performance demands of VR versions makes it a lot easier to adapt VR games to 2D and vice versa while maintaining more similar assets/performance targets. I think developing true AAA VR exclusives is simply too big an ask for most companies right now, but hopefully we get more really strong VR ports like RE8.
 

Fafalada

Fafracer forever
In that case, he would be right, it offers quaite a small performance advantage.
While it's difficult to give an exact number as lots of variables can influence this - lens-matching (which is what 'fixed foveated' really does) tends to be around 2x more efficient (in terms of pixel-compute) than standard rendering on the current crop of headsets (mostly anything released in last 6 years).
So it depends on how we define 'small' - on PC it has indeed not had a lot of traction (partially because of lack of standard ways to implement it - until VRS, you had to implement almost as many versions of it as there were GPUs on the market + not to mention paths for different HMDs - and while VRS is now standard on modern GPUs - if you pick 'just that' - nothing older gets optimizations (which would arguably need it more)).
But for closed-boxes like consoles or Quest - it's absolutely critical to use and by now it should be pretty much standard for every release.
 

ABnormal

Member
While it's difficult to give an exact number as lots of variables can influence this - lens-matching (which is what 'fixed foveated' really does) tends to be around 2x more efficient (in terms of pixel-compute) than standard rendering on the current crop of headsets (mostly anything released in last 6 years).
So it depends on how we define 'small' - on PC it has indeed not had a lot of traction (partially because of lack of standard ways to implement it - until VRS, you had to implement almost as many versions of it as there were GPUs on the market + not to mention paths for different HMDs - and while VRS is now standard on modern GPUs - if you pick 'just that' - nothing older gets optimizations (which would arguably need it more)).
But for closed-boxes like consoles or Quest - it's absolutely critical to use and by now it should be pretty much standard for every release.
Well, regarding FFR, its results cannot be quantified until the extent it is used is known. We could have a case where it is used only a bit at the very external margin, or a case where it's used till near the central area of the screen. And the extent of resolution and detail reduction is also another variable. So it's literally impossible to establish a fixed value. At best, it could be estimated the average use of it.
Anyway, when eye tracked foveated rendering is avaliable, FFR should be banned altogether: it ruins the image where used, and you see it when you look around the screen. So naturally FFR is used only marginally, otherwise the image would be ruined beyond acceptability. ETFR allows the reduction of resolution and detail to be in a far more wide area of the screen.
 

Fafalada

Fafracer forever
Well, regarding FFR, its results cannot be quantified until the extent it is used is known.
Optimal value exists, which I already described - it's called 'lens-matching' - ie. redistribute pixel density based on lens-distortion of the target headset. When done this way, there are no IQ trade-offs - it's equivalent quality to 'standard rendering' at that relative cost.

While indeed - you can arbitrarily go more/less aggressive - that just goes into realm of 'rendering resolution % scaling' - ie. it's not an optimization but an IQ slider at that point.

Anyway, when eye tracked foveated rendering is avaliable, FFR should be banned altogether: it ruins the image where used, and you see it when you look around the screen.
As I point out above - not when it's used for its intended purpose.
Even with Eye-tracking, having your renderer respect lens-geometry will be an additional efficiency boost. In ideal world GPUs would allow us to render using non-linear projection, and this would be a completely solved problem - but for now we still have to live with hacks like VRS (which is already far more robust than some of the older methods employed in the past).
 

ABnormal

Member
Optimal value exists, which I already described - it's called 'lens-matching' - ie. redistribute pixel density based on lens-distortion of the target headset. When done this way, there are no IQ trade-offs - it's equivalent quality to 'standard rendering' at that relative cost.
Optical distortion software correction has been employed since the first VR unit, because it's necessary (otherwise the rectangular-shaped frame would appear fish-eye-distorted in the headset), but it's not FFR (the number of rendered pixels is the same). I understand tough that FFR can be measured somehow around it. In that case it would be quite marginal from a benefit standpoint.
 
Last edited:
More shit about it being expensive? Bore me fucking later. It’s not expensive when you compare it to headsets that provide a similar quality experience like the Index.

People will fucking cry about anything won’t they? Meanwhile, I’ll be having a blast on Wednesday in my ‘expensive’ VR headset that could ‘do without’ eye tracking.
 
More shit about it being expensive? Bore me fucking later. It’s not expensive when you compare it to headsets that provide a similar quality experience like the Index.

People will fucking cry about anything won’t they? Meanwhile, I’ll be having a blast on Wednesday in my ‘expensive’ VR headset that could ‘do without’ eye tracking.
And you will have 2000 dollars left on your bank account vs another comparable set, to buy some sweeeeeeet games.
 

dyergram

Member
It’s priced to make a profit in a time of inflation and to keep making a profit if inflation gets worse. Also if it was $200 cheaper these idiot reviewers would still make the exact same point because you would still need to buy a PlayStation 5.
 

sachos

Member
I mentioned to the author his mistake on twitter, he replied to me saying this "I mentioned that it uses foveated rendering. I meant in terms of interactive gameplay features" as in it did not use the eye tracking for gameplay but he is wrong there too since they use eye tracking for aim assist plus from what i've read in some reviews the NPC use the eye tracking to keep eye contact with you? And if you keep contact for too long they look away.
 

Baki

Member
https://www.trustedreviews.com/opin...ld-be-better-off-without-eye-tracking-4303024


So basically Ryan at Trusted Reviews believes that the big flaw with PSVR2 is price, and suggests that Sony could have sold the headset cheaper if they didn't include Eye-tracking/Foveated Rendering, which he claims is underutilized and not currently worth it. He also uses John Carmack to attempt to debunk Sony's claims of the benefits with Eye-Tracking, and that he doesn't believe many games will put in effort to support it, including Sony himself expressing his disappointment with Horizon: Call of the wild's implementation.

While there are some issues in his post, he does raise two points that are pretty reasonable,

1. Devs who want to make cross-platform games for the most money, may likely skip eye-tracking for compatibility since most headsets don't have it. This is especially true since the biggest software hub currently in the VR market right now is on headsets that don't have the feature.

2. Sony not having a major FP showcase of the feature at launch.

With that said, while these points are valid for now, his main argument still derives form the issue of the headsets price.

I believe that while many people on gaming forums and certain enthusiast circles are quick to shrug off the price, especially for the specs for the current headsets out at this time, I'm noticing more and more from non-hardcore gamers, to casuals bringing up the price more and more as we get closer to launch which does makes me believe it's going to be an issue that will impact sales.

Ironically, I'm not seeing this same mindset for Apple's rumored headset, and Bigscreen was mopped on the floor over there $999 reveal for their headset, granted Apple has the infrastructure to offer subsidization options, but that's still a bit hypocritical.
It was always going to be niche as a wired headset. So they might as well go all out on the tech specs. If they wanted mainstream adoption, they needed to make a direct Meta Quest competitor and support the platform with tons of 1st party games. But that’s expensive and risky. So they went for niche and expensive instead.
 
Eye tracking is 240hz on PSVR2. If 30hz is accurate for the Quest Pro I understand why the image would get blurry when moving the eyes quickly. It simply can not keep up deciding render location with such a low speed.
I didn't know the eye tracking is 240hz - I did some math in a previous post around the biological speed your eyes move and focus, but with 120hz as the assumption. It still had a few frames to work with at 120hz, but 240hz would give it plenty of time. No wonder it works so well.

30hz would be completely unusable for rendering.
 

Three

Member
But wasn't Carmak talking about fixed foveated rendering, at that time? In that case, he would be right, it offers quaite a small performance advantage.
He can't have been because he mentioned the latency at the time. Fixed foveated rendering wouldn't be affected by latency
 

ABnormal

Member
He can't have been because he mentioned the latency at the time. Fixed foveated rendering wouldn't be affected by latency
If so, that would be actually strange: he is really prepared and it's strange that he would claim those low gains. Or maybe those gains are enough to keep a 60 fps game like RE8 at the same rendering level in VR as it is in flat. After all, even GT7 is already running at 60 fps, and in their VR versione they are still running at 60, with reprojection to 120, so maybe that' enough to bring parity between the versions.
Well, the most important thing is that it works, and they already demonstrated it very well.
Is there a video of him with the timestamp where he spoke about it?
 

Fafalada

Fafracer forever
Optical distortion software correction has been employed since the first VR unit
I was not referring to optical distortion correction - that's just a post-process shader that indeed, is always there.
The problem is the scene is rendered with linear-projection, into a rectangle, because that's what GPUs accelerate. Lens-matching is redistributing samples (or pixels - depending on your hw) from the simple rectangle into something that has density where it matters.
The simplest example is 'multi-resolution render targets' where you literally have variable resolution depending on which part of the render target you're looking at. Another is VRS - where you get variable sample count (it's not as effective as MRRT - but it's easier on the hardware, and now - somewhat standardized). There's a few dozen other approaches - but I won't go listing all of them - the principle objective is the same for all of them.

but it's not FFR (the number of rendered pixels is the same). I understand tough that FFR can be measured somehow around it. In that case it would be quite marginal from a benefit standpoint.
As I note above - FFR is what we do to optimize for the fact 'a lot' of pixels get wasted with your standard linear render into a rectangle. And the difference is definitely not marginal - on average, most HMDs (for at least first 5 years or so - I stopped paying attention after) had a sweet spot requiring somewhere between 1.4-1.6x increase in pixels (on each axis) - ie. 1.96-2.56x more pixels than the native resolution of the HMD display.
That's your multiplier for what 'optimal' lens-matching/FFR can do (optimal as in - without quality loss).
 
Top Bottom