Intellisample HTC and AA

[LS32]

Looking at the AA results from the nv3x gpus, I was thinking that the intellisample HTC compression technology does not work over 4x AA. It would make sense why the AA upto 4x has such a low performance hit, but as soon as you go over 4x AA the performance is practically halved or worse. Even with the 256 bit memory interface, without compression at all I would imagine that the scores would correspond to the huge loss.

Or is it just that modes above 4xAA use some chunk of SS to help their AA and it overwhelms the memory. If someone could test 6x+ AA on both cards (5800 and 5900) I think we could find this out.

[Uttar]

4XS+ uses a mix of MSAA and SSAA.
Nothing more to know, really. Big fillrate hit, and Color Compression isn't as efficient.


Uttar

[Slappi]

Originally posted by Uttar
4XS+ uses a mix of MSAA and SSAA.
Nothing more to know, really. Big fillrate hit, and Color Compression isn't as efficient.


Uttar


Uttar since you know wtf you are talking about and don't seem biased one way or the other (Plz stay this way) ..... who has the best AA and who has the best AF. Also what is better on the nVidia GPUs 4X AA or 4XS AA or to make it even easier WHAT IS THE BEST AA MODE for nVidia cards?

Thanks.

[StealthHawk]

Originally posted by Slappi
Uttar since you know wtf you are talking about and don't seem biased one way or the other (Plz stay this way) ..... who has the best AA and who has the best AF. Also what is better on the nVidia GPUs 4X AA or 4XS AA or to make it even easier WHAT IS THE BEST AA MODE for nVidia cards?

Thanks.

Everything is relative. It really depends on what you mean by "best." Best IQ without regards for speed? Best speed? Best mix?

4xS clearly has better quality than 4x for example, but the performance hit is a lot higher. You have to determine whether or not the hit(which is basically double 4x's) is worth it or not.

I haven't looked at reviews that compared the r9800Pro's FSAA quality with the gfFX5900's yet, but from what I've heard and seen, the gfFX5900's FSAA still isn't as good as ATI's quality-wise. But it's a lot faster now.

[Slappi]

Originally posted by StealthHawk
Everything is relative. It really depends on what you mean by "best." Best IQ without regards for speed? Best speed? Best mix?

4xS clearly has better quality than 4x for example, but the performance hit is a lot higher. You have to determine whether or not the hit(which is basically double 4x's) is worth it or not.

I haven't looked at reviews that compared the r9800Pro's FSAA quality with the gfFX5900's yet, but from what I've heard and seen, the gfFX5900's FSAA still isn't as good as ATI's quality-wise. But it's a lot faster now.


Heh.... maybe I should say the best overall AA quality and the best sweetspot AA mode.

[Uttar]

ATI has the best AA, mostly thanks to Gamma Correction, but also thanks to better sample patterns.

The difference in quality increases as the modes increase.
So the quality difference between 2x AA on nVidia & 2x AA on ATI is smaller than the difference between 4x AA on nVidia & 4x AA on ATI.

On the edge quality, the difference between 6xS ( nV ) and 6x ( ATI ) is even bigger: ATI looks a LOT better.
However, 6xS got better texture quality thanks to the SSAA, but the performance hit is a LOT bigger too.


With the NV30, ATI had the best AF, mostly quality-wise, and performance hits were roughly equal.
With the NV35, the quality now seems equal, in fact, you might even say it's BETTER on the NV35 in some cases ( smoother transistion in Performance by using some Trilinear, as good on all angles, ... )

So, with NV35 vs R350:
AF: Roughly equal, slight advantage for NV35
AA: Huge advantage for R350

The best mode, quality-wise, between 4x & 4xS is obviously 4xS.
6xS & 8x(S) obviously looks better, but the difference is very small and, IMO, not worth the performance hit.
4xS still got worse edge quality than ATI's 4x, though. But it got slightly better texture quality, not fully compensating it IMO.

So, if you got performance to waste, use 4xS instead of 4x. But most of the time, it ain't worth it.
I'm interested in seeing the quality of 4xS & 8x Performance AF, though. I guess I'll see once I'll get my card when it finally is available here in Belgium ( or I could always order in another European country )


Uttar

[StealthHawk]

Originally posted by Uttar
So, if you got performance to waste, use 4xS instead of 4x. But most of the time, it ain't worth it.
I'm interested in seeing the quality of 4xS & 8x Performance AF, though. I guess I'll see once I'll get my card when it finally is available here in Belgium ( or I could always order in another European country )


Uttar

Heh, would that be 8x "High Performance" AF or 8x "Performance" AF ;)

Gotta love how nvidia keeps changing the naming schemes :mad: I really think for the NV35 they should have dropped Aggressive entirely. There's no way a high end card should ever have to look that bad.

Also, do 6xS and 8xS look better than they did with NV30? I honestly couldn't tell the difference between those two modes and 4xS in NV30 reviews, but in Mike's NV35 review 6xS and 8xS are noticeably better than 4xS. Although the performance hit is still probably not worth it except in old games. Same OG modes as before? Or are they using the RG modes now?

I guess I'll probably find out later when I read other reviews :D

[Slappi]

Originally posted by Uttar
ATI has the best AA, mostly thanks to Gamma Correction, but also thanks to better sample patterns.

The difference in quality increases as the modes increase.
So the quality difference between 2x AA on nVidia & 2x AA on ATI is smaller than the difference between 4x AA on nVidia & 4x AA on ATI.

On the edge quality, the difference between 6xS ( nV ) and 6x ( ATI ) is even bigger: ATI looks a LOT better.
However, 6xS got better texture quality thanks to the SSAA, but the performance hit is a LOT bigger too.


With the NV30, ATI had the best AF, mostly quality-wise, and performance hits were roughly equal.
With the NV35, the quality now seems equal, in fact, you might even say it's BETTER on the NV35 in some cases ( smoother transistion in Performance by using some Trilinear, as good on all angles, ... )

So, with NV35 vs R350:
AF: Roughly equal, slight advantage for NV35
AA: Huge advantage for R350

The best mode, quality-wise, between 4x & 4xS is obviously 4xS.
6xS & 8x(S) obviously looks better, but the difference is very small and, IMO, not worth the performance hit.
4xS still got worse edge quality than ATI's 4x, though. But it got slightly better texture quality, not fully compensating it IMO.

So, if you got performance to waste, use 4xS instead of 4x. But most of the time, it ain't worth it.
I'm interested in seeing the quality of 4xS & 8x Performance AF, though. I guess I'll see once I'll get my card when it finally is available here in Belgium ( or I could always order in another European country )


Uttar


Thanks Uttar! It's sure nice when you don't have the time to research all this stuff yourself to get a good non-biased answer that fast. :thumbsup:

[pat777]

HCT=High Cheating Technology
That's what it stands for if your using 3Dmark 2003.
R350 has better anti-aliasling and equal AF IMHO.

[Richthofen]

no AF is inferiour because it depends on the angle.

[Morrow]

Without trying to be biased, the R3x0 has indeed much better FSAA (as Uttar said due to better sample patterns) and the nv2x and nv3x have much better AF (not angle depending) when considering quality as factor.

I'm too busy right now to look up which cards have the least performance hit by enabling FSAA or AF but fortunately Uttar did already a good job concerning this :)

[Typedef Enum]

The 'much better' part regarding AF is, IMHO, not completely justified.

The small tradeoff that ATI makes is extremely rare to detect, unlike the poor Antialiasing modes found in nVidia solutions. I realize I'm talking about 2 different animals, but I'm trying to counter people who might want to say, "Hey, you're saying that they made tradeoffs in their filtering...so, nVidia did the same with their AA modes as well."

While there is some degree of truth to that, my opinion is that the poor nVidia AA modes are much more obvious to my eye than anything relating to ATI's Anisotropic Filtering. To me, jaggies just stick out like a sore thumb.

Of course, it doesn't hurt that ATI's filtering takes such a low framerate hit as well. The bottom line is this...If I were the one calling the shots on designing the basic requirements for a 3D chip, I would much rather see filtering a-la ATI, and totally awesome, no frills Antialiasing modes.

I can only hope that nVidia's NV40 will finally put to use some of that 3dfx IP that's been collecting dust for a couple of years. I realize that NV40 wouldn't incorporate the exact same techniques as found on the VSA-100 chip...but they could, at the very least, dump all of their current modes and provide some significantly better ones using their experienced ex-3dfx engineering talent.

[CaptNKILL]

*zips up flame suit*

..... its gonna get hot in here..... REAL hot...

:rant: :firedevil :rant:

[StealthHawk]

Originally posted by Typedef Enum
Of course, it doesn't hurt that ATI's filtering takes such a low framerate hit as well. The bottom line is this...If I were the one calling the shots on designing the basic requirements for a 3D chip, I would much rather see filtering a-la ATI, and totally awesome, no frills Antialiasing modes.

With 43.51+ drivers AF on NV30 takes about the same or even lower performance hit than ATI's AF does.

[Typedef Enum]

I'm not confident that the low hit you're referring to is an across-the-board thing.

[StealthHawk]

Originally posted by Typedef Enum
I'm not confident that the low hit you're referring to is an across-the-board thing.

Well, that's what the game benchmarks in the previews showed, but you could be right ;)

[Ninja Prime]

Yeah, aren't those the same drivers that have the 3dmurk AF lower quality cheat....?