Apparently there is a hardware component to the solution. Without details to go by, here are some 4k screens of Horizon Zero Dawn that were released by Sony.
What do you guys think?
http://www.eurogamer.net/articles/2016-09-08-horizon-zero-dawn-4k-screenshots
One of the sample shots:
What is checkerboard rendering? The basic idea is, you render half the pixels in a checkerboard pattern and then reconstruct the rest. For instance, you could render the red pixels in this 2x2 grid pattern and reconstruct the green pixels. This is suggested in Valve's checkboard method in a presentation on rendering for VR. It is not yet known if this is what is suggested by Sony for PS4 Pro.
image credits to @HTupolev
There are a few known solutions for this. Rainbow Six Siege does a version of this. I believe the checkerboard pattern is individual pixels instead of 2x2 pixel groups. It uses motion vectors to reconstruct the green pixels from a previous frame of data. For a 4k output it would render at 1080p with 2xMSAA to get half the samples of 4k. See checkerboard rendering in this slide show. The final resolve is very complex and would be hard to implement as an easy solution for developers to turn on/off.
http://www.gdcvault.com/play/1022990/Rendering-Rainbow-Six-Siege
The other method I've seen linked by Digital Foundry was suggested by Valve for VR. It has 2x2 pixel groups and it interpolates the missing "green" pixels by average neighbouring pixels with weighted values. It would probably cause some blur, on top of whatever other artifacts that type of interpolation would suggest. The information begins at slide 21. http://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_Performance_GDC2016.pdf
@Globalisateur Linked a Sony patent that describes "uprendering" multimedia content. It appears to only require the current final frame buffer as a reference, and then creates a number of sub-pixel shifted copies to reconstruct a final "uprendered" image. The example in the document creates a 2x2 matrix out of each pixel with a 0.5 shift right, a 0.5 shift down and a 0.5 shift diagonal. That is my understanding. Good candidate for hardware. Does not seem to match well with "checkerboard" rendering or Jonathan Blow's vague (because of NDA) comments about PS4 Pro. May be a good starting point.
https://patents.google.com/patent/US20160005344A1/en
What do you guys think?
http://www.eurogamer.net/articles/2016-09-08-horizon-zero-dawn-4k-screenshots
One of the sample shots:
What is checkerboard rendering? The basic idea is, you render half the pixels in a checkerboard pattern and then reconstruct the rest. For instance, you could render the red pixels in this 2x2 grid pattern and reconstruct the green pixels. This is suggested in Valve's checkboard method in a presentation on rendering for VR. It is not yet known if this is what is suggested by Sony for PS4 Pro.
image credits to @HTupolev
There are a few known solutions for this. Rainbow Six Siege does a version of this. I believe the checkerboard pattern is individual pixels instead of 2x2 pixel groups. It uses motion vectors to reconstruct the green pixels from a previous frame of data. For a 4k output it would render at 1080p with 2xMSAA to get half the samples of 4k. See checkerboard rendering in this slide show. The final resolve is very complex and would be hard to implement as an easy solution for developers to turn on/off.
http://www.gdcvault.com/play/1022990/Rendering-Rainbow-Six-Siege
The other method I've seen linked by Digital Foundry was suggested by Valve for VR. It has 2x2 pixel groups and it interpolates the missing "green" pixels by average neighbouring pixels with weighted values. It would probably cause some blur, on top of whatever other artifacts that type of interpolation would suggest. The information begins at slide 21. http://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_Performance_GDC2016.pdf
@Globalisateur Linked a Sony patent that describes "uprendering" multimedia content. It appears to only require the current final frame buffer as a reference, and then creates a number of sub-pixel shifted copies to reconstruct a final "uprendered" image. The example in the document creates a 2x2 matrix out of each pixel with a 0.5 shift right, a 0.5 shift down and a 0.5 shift diagonal. That is my understanding. Good candidate for hardware. Does not seem to match well with "checkerboard" rendering or Jonathan Blow's vague (because of NDA) comments about PS4 Pro. May be a good starting point.
https://patents.google.com/patent/US20160005344A1/en
Last edited: