Super-Sampling AA

Ailuros said:
I can understand sparse or rotated grid SSAA, but we've already hybrid Super-/Multisampling modes since NV2x. Albeit any portion of SS will have a bigger fillrate impact, I don't think that the hybrid modes limit to just 1024, depending on application.
NVidia has hybrid sampling, but without specific hardware support. And the 1024 limit applies to ATI and S3.

Based on the above, does anyone have a clue how Quantum3D is or might be reaching 16xRGSS with their Clarity algorithm?
Maybe.


S3 supports 2x, 5x and 9x but not 4x. And they don't seem to have special hardware support for it ;)
 
Xmas,

Could you explain what you mean by 'specific hardware support' and also 'special hardware support' with reguard to AA?
 
You can do any FSAA by jittering and rendering the whole scene multiple times.
 
PeterAce said:
Xmas,

Could you explain what you mean by 'specific hardware support' and also 'special hardware support' with reguard to AA?
What I mean is the hardware being able to render the scene with a single pass using a non-ordered grid, supporting sample masking. Taking sample spacing into account for LOD/gradient calculation. Having dedicated downsampling hardware including gamma adjustment. And not wasting framebuffer space on parts of the scene that are not visible in the final output.
 
Morgoth the Dark Enemy said:
Does SSAA work with FP buffers?In theory it should

I think it will just be twice as slow with buffers that use a twice as wide data format. And the latency will increase as well.
 
this is still a "good" idea.
I absolutely wants at least 2x AA (always enabled on my ti4200).
if I had a R520 (I guess I'll wait two or three years to afford it :)) I would be willing to "waste" half the power to do 2x AA with Far Cry HDR for example.
(you're already "wasting" lots of power to do FP blending in the first place, yet it's playable on 6800GT)
 
At the moment SSAA with FP Buffers will only work if the App does ist. The driver can not do it because there is no FP16 Backbuffer format.
 
I was merely asking if SSAA as a technique would work with FP buffers, and it seems it can.I wouldn`t mind a pseudo-stochastic algorithm implemented in a future architecture, not at all :D
 
Back
Top