(01-18-2014, 11:40 AM)kinkinkijkin Wrote: [ -> ]Memory bandwidth'll not do much for you if you can't hold the whole frame in RAM.
Seconded the admin89´s explanation. If bandwith doesn´t do a lot, unnecessary millions of dolars wouldn´t been spended just to take that up.
Well, I never realised how small a frame was in memory...
Nevermind my garbage about VRAM, then.
Yeah.
3840 x 2160 x 3 bytes per pixel = 24MB for a color buffer using standard 24 bit RGB. At super high resolutions pixel shading is always the bottleneck. And usually those end up being bottlenecked by ROP throughput and memory bandwidth at those resolutions.
Titannous Wrote:Do you know if 4xSSAA also increases CPU load?
No. But 4xSSAA combined with 4x IR or auto window size (which in your case is an even higher IR) will produce an insane internal resolution that no card on the market should be able to handle.
Have you tried using 4xMSAA in the openGL backend with auto window size? If that's still too demanding than FXAA is the way to go.