SFR & AA

[Geo]

Does SFR reduce your memory usage on each card? Or is that data for "the other 1/2 of the screen" still taking up space on the card that is not showing that portion of the screen? What I'm wondering is if SFR would allow you to double the AA samples with the same memory requirements as 1/2 the samples on a single card implementation?

That might be a nice selling point. . .it seems to me, looking at the frame rates the GT/Ultra are hitting in SLI, that (and, yes, I know this is subjective) there is a tremendous amount of FPS overkill or CPU limitation in most situations. . .unless you find another way to use all that memory and horsepower to improve IQ above the top-end settings that current single-card implementations offer.

 

[ChrisRay]

Does SFR reduce your memory usage on each card? Or is that data for "the other 1/2 of the screen" still taking up space on the card that is not showing that portion of the screen? What I'm wondering is if SFR would allow you to double the AA samples with the same memory requirements as 1/2 the samples on a single card implementation?

That might be a nice selling point. . .it seems to me, looking at the frame rates the GT/Ultra are hitting in SLI, that (and, yes, I know this is subjective) there is a tremendous amount of FPS overkill or CPU limitation in most situations. . .unless you find another way to use all that memory and horsepower to improve IQ above the top-end settings that current single-card implementations offer.

I have heard some strange hypothesis from Ailuros in the past about using each card to take 1 cycle 2xAA and somehow combine them. But maybe that was just him daydreaming about a single cycle 4xAA . But since Nvidias AA isnt programmable I'd doubt it. I do vaguely remember hearing that SFR might grant a slim amount of new memory. But I havent seen anything that would leave me to believe its doing that. Keep in mind my experimental SLI rig has been 2 6600GTs and its been a most interesting and fun experiment. If there's a way to test any of this. I'd be happy to oblige.

Chris

 

[Dave Baumann]

I have my doubts that there are any memory foot print advantages - usually render memory is allocated when the resolution is set (vide restarts) and because SFR has a variable frame size per board for every frame it likely still has to capture a full frame buffer's of memory for each board.

 

[Ailuros]

I had a brainfart actually when thinking about how VSA-100 handled load balancing. You might got there double the AA performance with two chips, yet at a sacrifice of not scaling any geometry. As I said a rather bad brainfart.