This, you can't really convey how UHD and HDR look through a low bitrate stream on youtube and twitch. Just like you can't really sell VR on a live demo. Both of these things have to be experienced first hand. Sony didn't even make an effort to release the Pro footage at an acceptable bitrate and in HDR for those who have UHD/HDR sets.
Isn't that true about many things these days? People don't buy a car from a couple of videos they see on youtube, they tend to go try it or see it in person at least. Or a new TV. Probably even a new phone, a lot of the time.
I think it is more an effect of a non understood "minority law", companies should have never engaged in social media, a couple well crafteed (/useless) PR aside. They are filled by intolerant, full of themselves, spoiled, people that was given a perfect opportunity to matter...
As for the upscaling technique and what SOny does (checkboard OR foveat, or something else) I'm eager to learn the approach Sony will promote.
I'm curious to know if there is a bottom entry point or more precisely where it is: starting from where the pixel quality is considered good enough (I guess I consider the number of pixels, precision before the final resolved in the framebuffer, texture quality and filtering). Woud starting from 720P and going to 1080p produce good as good a result? I suspect not, and that it gets worse the lower the resolution drop compare the information contained (geometry, texture,etc.) before rendering.
Is this what they mean by checkerboard rendering? If so, it seems Xbox One already has it.
https://twitter.com/nesboxcom/status/778907625555714048
https://twitter.com/nesboxcom/status/778907625555714048
yup, listened to the giant bombcast, I was like what the hell is this.
As I understand it, checkerboard rendering works per pixel. It just alternates the pixel arrangement to fit the quads of the GPU. If you're looking for 2x2 pixel arrangements, you shouldn't be finding them.
Valve's checkerboard rendering paper suggested using 2x2 pixel arrangements. When the checkerboard story first broke, that's what DF pointed to. Not that it's necessarily how they're actually doing it.
Rainbow six renders at 1/4 resolution with 2xMSAA and saves two samples per pixel, if I remember correctly. It also has a fairly complicated process for reconstruction.
There's also the possibility that COD is doing something custom. Until they have the hardware talk about how it works, not idea how to tell which games are using Sony's method, and which are doing custom. Repi said on twitter that Mass Effect was custom, but wouldn't talk details.
The screenshots for Horizon ZD show pixel-level dithering, not quad-level. If reconstructing in a 2x2 box, you shouldn't get that sort of dithering. It's fairly straightforward to look at the concept and conclude that alternating quads over columns/rows and frames would give you pixel-level information for reconstruction, so there'd have to be a good reason not to do that.
Addressing this from another direction, what does @jlippo see in the video to suggest it is pixel level and not quad level?
Addressing this from another direction, what does @jlippo see in the video to suggest it is pixel level and not quad level?
Both London studios and Valve agreed, when explaining their methods, that doing the checkerboard pattern at per-pixel granularity would be preferable quality-wise, but it was a concession they did because gpus shade fragments in 2x2 groups, or, pixel quads. That might help z-compression too, but I have no idea. They probably could have used custom MSAA resolve, but its a more complicated implementation, and in their case, they were using this to just reduce resolution at screen edges that end up distorted and scaled down anyways in the final VR buffer. For a full screen planar image as in PS4 pro, a more refined checker-board method is definetly warranted, but we still don't know sony's strategy.
Well then there's definitely a discrepancy between what we see in the screenshots (4K, PS4 Pro, pixel level checkerboard artifacts) and what they claim to be doing with PSVR on PS4.
The PSVR patent:
http://www.freepatentsonline.com/20150287166.pdf
This is beyond what I can understand, too many words, too long sentences, maths.
The PSVR patent:
http://www.freepatentsonline.com/20150287166.pdf
This is beyond what I can understand, too many words, too long sentences, maths.
Checkered patterns are not rendered as such.
Samples are non-interleaved until the combination pass. (Ie. In 2xMSAA buffer.)
That was not the case with valve's and london's naive implemebtations, which they admited themselves to be early experiments with room for improvement. They did reportetly produce good enough results as is though. Check out their respective presentations.
Yes, both of them worked by dropping/masking quads.
My response was aimed for the pixel sized checkerboard.
I really doubt that there is anything new in Ps4 to bypasses quad limitation, just classic tricks to render interleaved patterns in non-interleaved way.
This generation will certainly be interesting in terms of how we think resolution and shading resolution. (spatial and temporal.)
I'm sure there will be lots of interesting tricks.
