FX Color compression: AA only

[Joe DeFuria]

I'm only posting this because I know that in some thread, somewhere on this board, I believe Hellbinder stated that despite many of the claims being made of nVidia's color compression being "on all the time", that nVidia admitted that's it's only enabled in conjunction with AA. (Same as ATI.) Someone asked for a source, but it couldn't be found.

I just stumbled on this recent interview:

http://www.3dcenter.org/artikel/2003/03-05_english.php

9) For the GeForceFX, NVIDIA has developed a new color compression technology that can compress color information up to a ratio of 4:1. Is color compression used in all cases, or only with activated anti-aliasing? If it is used even for non-anti-aliased modes, does this mean that the color compression will also work on textures that are not already compressed using, for example, DXTC? Specifically, are textures created by render-to-texture compressed as well, using color compression?

NVIDIA: The GeForceFX colour compression is activated only for anti-aliased rendering and only applies to the frame buffer. The GeForceFX uses DXTC and S3TC texture compression for textures.

 

[Hellbinder]

Yes it was me to state that. And it was me who got blasted by the regulars here (as usual) for saying it.

Not suprisingly, i am right again.

 

[K.I.L.E.R]

]Yes it was me to state that. And it was me who got blasted by the regulars here (as usual) for saying it.

Not suprisingly, i am right again.

That's because at the time it was stated by some PR junkie that it was doing 4:1 compression ALL the time, even when no AA was applied.
The choice came down to, believe:
HB or a PR rep.

With the way you carry on sometimes, it's no wonder some people would rather believe the PR guy over you.

Apologies, if it may seem offensive. It's just my personal opinion.

 

[Dave Baumann]

Actually - the statement about it made to B3D was by the NV30 product manager, that statement above was by a PR rep.

 

[K.I.L.E.R]

Guess the PR rep was looking at the R300 papers instead of the FX ones.

-joke, if someone doesn't get it.

 

[antlers]

Actually - the statement about it made to B3D was by the NV30 product manager, that statement above was by a PR rep.

Which goes to show that even product managers make mistakes, especially when they are trying to spin their product positively.

{Edit]

Actually, I wouldn't be surprised if color compression was operating all the time on the NV30. However, the important consideration is that it will only have a non-negligible impact on bandwidth consumption in AA, because it only compresses identically colored pixels in 2x2 blocks. When people heard framebuffer compression that was not tied to AA, they seemed to imagine some magic lossless compression technique that would somehow validate NVidia's decision to use a 128-bit bus on the NV30, and there just ain't no such animal.

 

[jjayb]

If this is true, it's just one more thing that Nvidia has mislead people about with the nv30. Yes Russ, we know, what matters in the end is performance. We know you don't feel being mislead is important.

 

[RussSchultz]

Its comforting to know my opinion is important enough to drag into every conversation.

It makes me feel all warm and fuzzy.

(p.s. its considered bad form to throw digs at people who aren't even involved in your conversation)

 

[3dcgi]

I'm wondering is Quadro FX has full time color compression. If they implemented a differential scheme like Z uses it would work well with gourand shading. If so then GFFX would support it as well, it's just not enabled.

 

[Dave H]

An interesting test case for this might be a game that uses cell shading. Should provide plenty of opportunity for color compression without MSAA.

 

[Simon F]

An interesting test case for this might be a game that uses cell shading. Should provide plenty of opportunity for color compression without MSAA.

Or Quake.. where everything is brown.

Seriously though, there doesn't seem a lot of point optimising for such a small class of applications. The chance of getting repeated colours in a 2x2 grid in a typical game is fairly small.

 

[Dave H]

An interesting test case for this might be a game that uses cell shading. Should provide plenty of opportunity for color compression without MSAA.

Or Quake.. where everything is brown.

Seriously though, there doesn't seem a lot of point optimising for such a small class of applications. The chance of getting repeated colours in a 2x2 grid in a typical game is fairly small.

True, but if the optimization comes for free due to the way color compression is already implemented, why not use it? Incidentally, I think cell shading may not turn out as small a niche as you might think; it's getting quite popular on consoles already, and the upcoming Zelda game looks to be the sort of hugely innovative and successful game that pushes the industry along new directions. There hasn't been a lot of cell shading crossover into PC gaming yet, but I bet that'll change.

 

[Dave Baumann]

I've long since been of the impression that GFFX's 1600x1200 performance has been quite strong - surely if any resolution is likely to show the benefits of colour compression it would be this one, not least because of the bandwidth constraints but because if any resolution is going to be likely to have 2x2 colour blocks its going to be this.

 

[Simon F]

True, but if the optimization comes for free due to the way color compression is already implemented, why not use it?

Because it's very unlikely to be free. With MSAA, you don't have to test whether certain colours are identical - it's inherent whenever the same polygon spans several sub-pixels.

The more general case, OTOH, would require tests of the RGBA values which, while cheap, still costs gates. Once you have any shading (and it wouldn't surprise me if some cell techniques do, say, Gouraud shading) then you have rather small chance of a repeated colour. Is it worth the effort? Of course, you could store deltas etc, but it's a case of rapidly diminishing returns.