R520 will support 8x AA?

[Jawed]

Using a 16x16 sparse sampled grid? I'm not thinking about "temporal AA" or a SuperAA mode, I'm thinking about the basic grid used for MSAA.

Is a 16x16 grid "compatible" with 2x, 4x and 6x AA modes? The reason I ask is that R3xx style AA, using a 12x12 grid seems more "sparse" than 16x16 - since 12 isn't a power of 2. So would a 16x16 grid actually mean that 2x, 4x and 6x AA modes look inferior on R520 than on older ATI GPUs with a 12x12 grid?

Jawed

 

[Jawed]

Thinking about it, since a 12x12 grid already supports 8xAA, there is not necessarily any need to make the grid 16x16... (R3xx ... R4xx support 12x MSAA)

I suppose it's more a matter of the maximum AA level desired. e.g. if you wanted to SuperAA two R520s, each with 8xAA, a 12x12 grid would run out of sample positions, and a 16x16 grid wouldn't look "sparse". So perhaps a 20x20 sample grid would be more suitable...

Anything to ease the boredom...

Jawed

 

[no-X]

Isn't possible, that there will be no fixed grid? It would be good for use with SuperAA (8 per VPU, 16 samples total, enough to come out with stochastic sampling).

 

[KimB]

You'd need a 24x24 grid to support 6x and 8x, minimum.

 

[Ailuros]

Isn't possible, that there will be no fixed grid? It would be good for use with SuperAA (8 per VPU, 16 samples total, enough to come out with stochastic sampling).

I won't say it's impossible, but the multi-GPU market is so small that it's questionable that it's actually worth the effort. Implementing a stochastic or semi-stochastic algorithm shouldn't be that easy according to my understanding.

 

[Dave Baumann]

You'd need a 24x24 grid to support 6x and 8x, minimum.

Sorry?

 

[no-X]

I won't say it's impossible, but the multi-GPU market is so small that it's questionable that it's actually worth the effort. Implementing a stochastic or semi-stochastic algorithm shouldn't be that easy according to my understanding.

I know it's not easy to implement (and I'm not sure if its generaly better than sparse TAA), but 3DLabs solutions use it for a long time. I don't need stochastic sampling as such, I want no fixed grid and st. sampling like a possibility (may be later)

You'd need a 24x24 grid to support 6x and 8x, minimum.

Sorry?

Chalnoth probably meant, that positions of samples wouldn't be so good, if the grid will be smaller

 

[Ailuros]

Sorry?

Besides the obvious brainfart, I'd guess that analytical algorithms might get implemented before we see such high grid values.

 

[Dave Baumann]

Chalnoth probably meant, that positions of samples wouldn't be so good, if the grid will be smaller

A 12x12 grid is already supporting 12x AA firaly optimally.

 

[tEd]

I voted no.

I don't think it's impossible but i think it wouldn't make much sense to do it. The quality increase from 4x to 6x is already small with resolutions 1024*768 and higher and from 6x to 8x would be even smaller at the point where it is not really worth the transistors IMO.

Maybe it would be better to invest in a good compression scheme for AA so it doesn't take as much space as it does today.

[sarcasm]
Besides it seems devs. can't count higher than 4 anyways as more games which support AA(which is nice and all) don't have a higher AA option than 4

 

[Jawed]

I agree that the quality difference would be small - in screenshots.

What I find particularly disappointing about ATI's 4x/6x MSAA is the amount of shimmer/crawl with movement. Anything to ameliorate that is worth having, in my view...

Jawed

 

[Ailuros]

I know it's not easy to implement (and I'm not sure if its generaly better than sparse TAA), but 3DLabs solutions use it for a long time. I don't need stochastic sampling as such, I want no fixed grid and st. sampling like a possibility (may be later)

Stochastic algorithms aren't losless afaik and that's one of the reasons why high sample densities are required.

We've looked at things such as randomly altering the programmable pattern, but the low frequency noise introduced was worst than the improvement in the sampling position. Unless you have 32~64 subsamples, introducing random variations is not good. So we are looking at other solutions and algorithms. Current and future users will be happy with our solutions. Stay tuned.

http://www.3dcenter.de/artikel/2003/11-06_english.php

 

[Ailuros]

I voted no.

I don't think it's impossible but i think it wouldn't make much sense to do it. The quality increase from 4x to 6x is already small with resolutions 1024*768 and higher and from 6x to 8x would be even smaller at the point where it is not really worth the transistors IMO.

Shouldn't it be expected that the difference between 4x and 6xAA is small? From 2x to 4xAA the EER doubles in values (2*2 vs. 4*4), which isn't the case with 6xAA and it's 6*6 EER.

However an 8*8 EER is twice as good - in theory at least- than 4*4.

 

[tEd]

Shouldn't it be expected that the difference between 4x and 6xAA is small? From 2x to 4xAA the EER doubles in values (2*2 vs. 4*4), which isn't the case with 6xAA and it's 6*6 EER.

However an 8*8 EER is twice as good - in theory at least- than 4*4.

in theory maybe but....

at lower resolution the difference is certainly bigger but at higher resolutions i play the difference is often not noticable while playing

 

[Ailuros]

in theory maybe but....

If you don't see a difference in real-time, then why do you see a difference in real-time between 2x and 4xAA?

If you enable 16xS via a third party application on a recent GeForce (which does deliver an 8*8 EER) the difference on poly edges/intersections is of course visible compared to 4xRGMS (besides the obvious difference in texture quality).

***edit:

at lower resolution the difference is certainly bigger but at higher resolutions i play the difference is often not noticable while playing

Higher than 1600?

 

[tEd]

If you don't see a difference in real-time, then why do you see a difference in real-time between 2x and 4xAA?

If you enable 16xS via a third party application on a recent GeForce (which does deliver an 8*8 EER) the difference on poly edges/intersections is of course visible compared to 4xRGMS (besides the obvious difference in texture quality).

I can see a difference from 4x to 6x if i force me to look at the edges all the time but that's not the case when i actually playing i game. Often it really depends on the game/contrast/colors/speed also. Mathematically it may still make a bigger difference but the overall experience it doesn't appear so.

All i can say from my 2+ years experience(having a card supports 6x) playing games at least at a res. 1024*768 or higher on 21" monitor 4x to 6x smaller than 2x to 4x.

 

[Ailuros]

All i can say from my 2+ years experience(having a card supports 6x) playing games at least at a res. 1024*768 or higher on 21" monitor 4x to 6x smaller than 2x to 4x.

That's what I actually said. The difference between 4x to 8x will be as big as 2x to 4x.

Naturally you'd be better off with anything =/>1280*960 on a 21" inch monitor.

I can see a difference from 4x to 6x if i force me to look at the edges all the time but that's not the case when i actually playing i game. Often it really depends on the game/contrast/colors/speed also.

It depends on way too many factors. If you have thin wires hanging in the air for instance, you'll most likely notice a difference without having to concentrate on it.

More is always better obviously and it should be separated from what is actually sensible or possible for a specific timeframe, considering bandwidth or memory footprint restrictions.

 

[tEd]

Naturally you'd be better off with anything =/>1280*960 on a 21" inch monitor.

If i can , i do

 

[KimB]

That's what I actually said. The difference between 4x to 8x will be as big as 2x to 4x.

No, it won't. Yes, you're doubling the sample density, but that doesn't mean you're doubling the benefit. The benefit of AA is the removal of situations where aliasing is noticeable. Many more situations are removed in the lower range of AA than in the higher range (e.g. 16x sparse would be virtually indistinguishable from 32x sparse).

 

[ChrisRay]

Shouldn't it be expected that the difference between 4x and 6xAA is small? From 2x to 4xAA the EER doubles in values (2*2 vs. 4*4), which isn't the case with 6xAA and it's 6*6 EER.

However an 8*8 EER is twice as good - in theory at least- than 4*4.

Are you sure that judging the EER is going to be a practical means of judging quality?