Simple Q: How to calc AA Sample Fill Rate

[demonic]

Noob Q really

Going over your article about the R9700.

http://www.beyond3d.com/reviews/powercolor/radeon9700/index.php?p=3

How did you guys calculate the AA Sample Fill Rate?

Thanks

 

[demalion]

Looks like pixel fillrate * maximum number of MSAA samples (6 for the R300).

Why does this make sense? To me it is because:

With color compression and MSAA, theoretically you can get a maximum of 6 samples for every pixel written. The actual rate would depend on the efficiency of the color compression.
Also, with SSAA, color compression (as I currently envision it) wouldn't work, and if some form did, its efficiency would tend to be unpredictable (and pretty low, I believe).

I was going to linkify "MSAA" and "SSAA" above to the glossary, but it still hasn't been updated.

*cough* *ahem* Excuse me, just clearing my lazy-user-cracking-his-whip-at-the-people-who-run-the-site-again throat.

 

[demonic]

thanks demalion.

 

[Ante P]

so what does "Gs" stand for?

"GigaSamples"?

anyways, seems like an outdated specification to me
upping the memory (which doesn't really have an impact on pixel fillrate) would increase bandwidth which is important to MSAA

so if you want a measurement of theoretical AA speed just looking at the plain bandwidth would be a better idea

or am I just being a moron again?

 

[KimB]

Just looking at plain bandwidth doesn't mean nearly as much as it would have with GeForce3/4-class MSAA. Color/z compression is now in use. The only numbers worth looking at are benchmarks.

 

[Ante P]

Just looking at plain bandwidth doesn't mean nearly as much as it would have with GeForce3/4-class MSAA. Color/z compression is now in use. The only numbers worth looking at are benchmarks.

of course

but in any case bandwidth would be more telling that "AA fillrate" correct?

 

[Xmas]

but in any case bandwidth would be more telling that "AA fillrate" correct?

Either of them could be limiting in some situations, so none is a 'better' measurement. But it is better to separately give pixel fillrate and maximum number of MS samples per pipe rather than 'AA fillrate'.

 

[KimB]

Final performance, not theoretical numbers, are what matters.

 

[Arun]

Final performance, not theoretical numbers, are what matters.

Agreed. But theorical numbers are useful for developers: knowing a specific future card will probably have 5GP/s is useful to determine your game targets. And of course, you always consider something like 10% waste to be sure your game is playable


Uttar

 

[Humus]

I don't think I've ever considered any theoretical number while coding anything. I just do what I think would run reasonably good, then code it. Then I'll see if it turned out as fast as I thought. Only if it runs considerably slower than I had projected I would start to analyse it.