Unity 6 - Spatial-temporal post-processing (STP)

[Scott_Arm]

Just sad to see this:

https://twitter.com/x/status/1756908874311819360

Good thread by the way. Hopefully Unity does not give up on STP. Nice to see a different method for anti-aliasing and upscaling hitting the market. Hopefully they'll have someone with the capability to optimize as outlined in the thread, and they don't just turf it now that he's gone.

 

[MfA]

Was STP interpolation or extrapolation like Intel ExtraSS?

 

[Frenetic Pony]

View attachment 10782

wtf is Unity doing?

They went from a CEO that had already failed before in the games industry to one that has no idea what the games industry is about.

Here we have board that doesn't know and doesn't care what the industry the company is in, they just want returns and they'll fire people until they get them.

 

[tuna]

Whitehurst is a smart guy and I think he can figure out the game engine business (which is different from the games industry).

Unity has to downscale and focus on its strengths if they want to become profitable again.

 

[MfA]

As for the last statement, it may have been viable in the past to employ some sort of heuristic (multiple fragments/depth discontinuities/etc) or edge detection algorithm to handle MSAA pixels but with Nanite we can't rely on such clever methods to skip much of the work anymore ...

Bit of a necro, but I stumbled on this blog which suggests stochastically redistributing G-buffer samples from a quad when you get two surfaces inside a pixel. Interesting way to retain regularity.

You could probably do something similar in nanite, use a Z-buffer at 4x resolution with 30 bit for depth and a 2 bit index to select a sample from a quad in the visibility buffer. Keep the visibility buffer at 1x resolution, but add a coverage mask. Goes from 64 bit per pixel to 128 bit per pixel, but you get MSAA (approximately).