ATi and SSAA

[Ante P]

http://w1.461.telia.com/~u46115110/ATi1.jpg

http://w1.461.telia.com/~u46115110/ATi2.jpg

[jb]

yea but in the last review at driverhaven didn't ATI say they were not going to enable SSAA ???

[Ante P]

yea but in the last review at driverhaven didn't ATI say they were not going to enable SSAA ???

My point exactly.

[Beatles]

yea but in the last review at driverhaven didn't ATI say they were not going to enable SSAA ???

My point exactly.

Just drop it Uriel.

[Nagorak]

Who cares? This is like comparing a 320 kbps MP3 to an uncompressed Wav file. Theoretically the wav sounds better. In reality, I don't care who you are, you will not be able to tell the difference between the two files (128 kbps MP3, yes, 320 kbps, no).

When playing a game you won't be able to see the difference in SSAA and MSAA unless you're looking for it and not actually playing.

[T2k]

It's about promises...

[Hyp-X]

When playing a game you won't be able to see the difference in SSAA and MSAA unless you're looking for it and not actually playing.

It depends on whether there are any trees in the game...

[martrox]

It's about promises...

No, actually, it's about prioritys........

[jb]

well for some older games like HL/CS then I can get a better picture image with SSAA and not kill frame rate. ATI in their paper says SSAA. They say the card does support it. I should have it. Or they should remove it from their documentation.

[Sage]

damnit if they would enable JGSSAA i would grab a R300 in a moment because after owning a Voodoo5 i CAN tell the difference!!! My GF4 still doesnt look nearly as good as my Voodoo... until then, ATi can go whack off in the canadian tundra for all i care (as can nVidia and anyone else whos AA still SUCKS after years of time for improvement)

[SanGreal]

Who said it isnt supported?

:roll:

[Babel-17]

http://www.rage3d.com/board/showthread.php?s=&threadid=33658992


[Zardon] Will The Catalysts for the 9700 support both multisampling and supersampling anti-aliasing in the near future, such as detailed here http://mirror.ati.com/vortal/r300/educational/main.html ?

[Ben] We have investigated this possibility, however we believe that sample-for-sample, the RADEON 9700 & 9500 series multisampling implementation, provides better performance and image quality than any supersampling method. So we currently have no plans to support supersample anti-aliasing on these products."

[SanGreal]

http://www.rage3d.com/board/showthread.php?s=&threadid=33658992


[Zardon] Will The Catalysts for the 9700 support both multisampling and supersampling anti-aliasing in the near future, such as detailed here http://mirror.ati.com/vortal/r300/educational/main.html ?

[Ben] We have investigated this possibility, however we believe that sample-for-sample, the RADEON 9700 & 9500 series multisampling implementation, provides better performance and image quality than any supersampling method. So we currently have no plans to support supersample anti-aliasing on these products."

Humm, I dont know what happened to the rest of my post, but my point was that it isn't implemented.. that doesnt mean the feature isnt supported by the core which is all the images claim
:oops:

eidt: it seems clicking the smiley buttons on the left deletes the line of text the cursor is on :rolleyes:

[andypski]

Any application writer can get supersampled AA on a Radeon 9700 today - it is a simple matter to create a larger render buffer and then scale it with filtering to the back buffer. Hence supersampling is supported by the drivers and hardware and in fact has been since day one.

The hardware is also fully capable of doing a mix of multisampling and supersampling, although exposing this support through the API can be more complex. It seems a bit sensationalist to me to claim that a feature is not supported when in fact it is.

The issue is clearly not the support for supersampling per se. but the act of exposing it through the control panel for legacy applications in the same fashion as multisampled AA is today - feature requests like this are always given appropriate consideration and the difficulties and advantages of implementation are weighed up by the team. We do want to make our customers happy whenever possible ;)

So, please let's keep things reasonable - I hate to see it implied that something sinister is going on when in fact it isn't. :)

Thanks
- Andy.

[Evildeus]

Any application writer can get supersampled AA on a Radeon 9700 today - it is a simple matter to create a larger render buffer and then scale it with filtering to the back buffer. Hence supersampling is supported by the drivers and hardware and in fact has been since day one.

The hardware is also fully capable of doing a mix of multisampling and supersampling, although exposing this support through the API can be more complex. It seems a bit sensationalist to me to claim that a feature is not supported when in fact it is.

The issue is clearly not the support for supersampling per se. but the act of exposing it through the control panel for legacy applications in the same fashion as multisampled AA is today - feature requests like this are always given appropriate consideration and the difficulties and advantages of implementation are weighed up by the team. We do want to make our customers happy whenever possible ;)

So, please let's keep things reasonable - I hate to see it implied that something sinister is going on when in fact it isn't. :)

Thanks
- Andy.
Then do not ad for it in this way or make it available. If the card support it, i whould have the power to choose in the control panel which one i want.

It's just misconducting, and i'm sure that an american lawer could find a way to sue Ati :lol:

[Dave Baumann]

It's just misconducting, and i'm sure that an american lawer could find a way to sue Ati

Why? Did they advertise it as being in the control panel?

[Basic]

No, but they advertised a jittered super sampling with color compression that could give up to 6x the performance than smoothvision (1.0).

Application driven render to higher resolution and then downfiltering will give OGSS, and won't use the color compression.

So you can't use the SS features in smoothvision 2.0. If the hardware has a feature, but it isn't available through drivers, and there's no plans to add it, then I wouldn't call it sensationalistic to say that it isn't supported. Because for the end user, it isn't.

Whether it's good prioritation to do other optimizations first due to limited driver writer resourses is a different question.

[Dave Baumann]

It strikes me as a mixup between PR and engineering, since the documentation talked only of colour compression with SSAA, and of higher Z compression with MSAA. Due to the nature of the process MSAA is the one thats going to benefit most from colour compression (and Z buffer compression) and SSAA will probably only benefit from Z buffer compression.

[Ante P]

It seems a bit sensationalist to me to claim that a feature is not supported when in fact it is.

I just went by what ATis Vice President of Software said. "We currently do not support super sampling".

Anyways I as well as most other potential consumer would assume it's a feature that the user can control since they do not at all differentiate between the three modes in that presentation when it comes to how it will be made availible.

When was the last time you bought a MOBO which claimed to have FSB steps in 1 MHz intervals only to realize that the fine print said "only if some third party software developers decides to write software to control it, we don't let you control it in our BIOS"?

[Babel-17]

Perhaps we can all agree that it would be both pleasing and conducive to harmony if ATI would post a substantive and encompassing clarification. :)

If people as intelligent as the ones who post on these boards find it hard to readily fathom the sense of the matter think of how at a loss the general public must be. It behooves the more connected members of this board to gently prod ATI to action regarding this slight contremps.

Lol, my original post looked quite different. Ah, the calming influence of meditation. :)

[Ante P]

Is a software developer able to make his application use rotated grid, jittered, super sampling with color compression?
If not when what you said pretty much doesn't apply.

If the document just said "Super Sampling" and nothing else then sure, you could always brush it off by saying that you can render at a higher res and then downfilter it. (Personally I'd still say it would be a bit misleading.)

I'll stop bickering now hehe, I just want to see the feature (preperly, as in by user control) supported.
(It's not about sensationalism or flaming ATi or anything like that, I just want the feature the document says my card supports: nothing more, nothing less.)

[Dave Baumann]

Is a software developer able to make his application use rotated grid, jittered, super sampling with color compression?

IMO, the documenation is just plain wrong. Even if SSAA were available via 'SMOOTHVISION' there wouldn't be color compression anyway, because SSAA doesn't lend itself to compression as well as MSAA -- especially since each pipeline on Radeon 9700 would be producing one SSAA sample whereas each pipe produces up to 6 MSAA samples.

SSAA does get the benefit of up to 24:1 Z compression though (in 6X mode), regardless of how its enabled.

[Dave Baumann]

Does anyone have the original 9700 White paper / document that was sent out at the launch? I appear to have lost mine.

[Ante P]

Does anyone have the original 9700 White paper / document that was sent out at the launch? I appear to have lost mine.

I have it on the CD they sent out if that's what you mean.
I'll have a look around, perhaps I threw it in the bin.

[Dave Baumann]

AHHHH! The CD :!:

[Ante P]

AHHHH! The CD :!:

lol sorry
no CD anyway, I was referring to the press kit-CDs from last summer

[Dave Baumann]

Yeah.

The documentation there only states MSAA anyway.

SMOOTHVISION 2.0 technology offers the fastest, highest quality, and most flexible anti-aliasing implementation available in a consumer product. It combines the best features of all existing solutions, while adding a few unique innovations of its own. SMOOTHVISION 2.0 anti-aliasing uses a multi-sample algorithm with 2, 4, or 6 samples per pixel and a programmable jitter table for custom sample patterns. The default sample patterns are the result of years of testing and experimentation.

[Ante P]

Yup Smoothvision2.0.pdf also has a breakdown of different versions of SSAA and then continues by saying R300 uses MSAA.
But still apparently they've been thinking about SSAA since they've mentioned twice that they though about using it as well as combinations, I guess that presentation just slipped out the door while they were pondering to implement it or not.

Still: it's there alive and kicking.
Seems strange that they don't remove or chage it.

[Basic]

I agree that color compression probably isn't going to be as good with supersampling, and even if the compression was so good, it wouldn't be the limiting factor. So I realy didn't expect any 6x speedup. I wouldn't be suprised if PR got something wrong. (They still promised it though. :wink:)

The important part is the JGSS, which can't be done on the application side. Unless with an accumulation buffer approach, but that will certainly not give any speedup compared to earlier methods.

So it's clearly a promised feature that didn't make it. Given what andypski said here it seems as it's in the hardware. But we just can't use it.

Again, it still might have been a good thing to put other things first in the todo-list.

[Ichneumon]

Is there a link to where SSAA was "Promised"
It has been talked about I know... and that ATI said they were going to look into implementing it... but I don't remember ever hearing ATI saying "We'll add SSAA support for sure in the control panel by X" or something.

I get the feeling this is another circumstance of a company saying we know you all want this, and we'll look into its feasability... and users saying oh boi ATI promised us new features!

[Dave Baumann]

I agree that color compression probably isn't going to be as good with supersampling, and even if the compression was so good, it wouldn't be the limiting factor. So I realy didn't expect any 6x speedup.

I think, due to the nature of how the colour compression works, it wouldn't operate at all with supersampling. ATI don’t compress the entire buffer, the compress the block of pixels on the way out – this is why compression only works with MSAA on, and not with normal rendering. SSAA would probably go through the normal rendering path and hence not be compressed (other then Z buffer).

GeForce FX, on the other hand, compresses the colour buffer all the time. I think they actually had it in mind that they were going to continue with the mixed super/multi sample modes.

The important part is the JGSS, which can't be done on the application side. Unless with an accumulation buffer approach, but that will certainly not give any speedup compared to earlier methods.

Or a D3D Multisample buffer approach.

[Ante P]

Is there a link to where SSAA was "Promised"
It has been talked about I know... and that ATI said they were going to look into implementing it... but I don't remember ever hearing ATI saying "We'll add SSAA support for sure in the control panel by X" or something.

I get the feeling this is another circumstance of a company saying we know you all want this, and we'll look into its feasability... and users saying oh boi ATI promised us new features!

Doesn't the first two pics speak for themselves?

[Basic]

DaveBaumann:
Are you saying that you get SS on a R9700 (with current drivers) if you set the right "MultisampleQuality" when creating the rendertarget?
What andypski said didn't sound like that to me, but I might have misunderstood him.

[Dave Baumann]

Are you saying that you get SS on a R9700 (with current drivers) if you set the right "MultisampleQuality" when creating the rendertarget?

Sorry?

[SanGreal]

Doesn't the first two pics speak for themselves?

You mean where it says "SMOOTHVISION 2.0 Supports" ?

ATi also said the r300 supports 256-chip configurations.. I dont consider that a promise to sell them.

[Ante P]

Doesn't the first two pics speak for themselves?

You mean where it says "SMOOTHVISION 2.0 Supports" ?

ATi also said the r300 supports 256-chip configurations.. I dont consider that a promise to sell them.

So when they say they support Pixel Shader 2.0 that's not really a promise either I guess?
I mean they could propably software emulate it at 0.01 fps but hey I didn't see them saying it would all be done in hardware?

Or what about not providing Open GL extention for anything?
Hey the hardware still supports it.

[SanGreal]

So when they say they support Pixel Shader 2.0 that's not really a promise either I guess?
I mean they could propably software emulate it at 0.01 fps but hey I didn't see them saying it would all be done in hardware?

Or what about not providing Open GL extention for anything?
Hey the hardware still supports it.

I don't understand your analogy. The pages in question discuss the Smoothvision 2.0 technology, not the Radeon 9700. The only claim being made is that Smoothvision 2.0 supports supersampling. For all you know, ATi could release a future product incorporating Smoothvision 2.0 that implements the supersampling features, or not.. it doesn't really matter as ATi never said that any product would support supersampling, merely that the technology does.

[Ante P]

So when they say they support Pixel Shader 2.0 that's not really a promise either I guess?
I mean they could propably software emulate it at 0.01 fps but hey I didn't see them saying it would all be done in hardware?

Or what about not providing Open GL extention for anything?
Hey the hardware still supports it.

I don't understand your analogy. The pages in question discuss the Smoothvision 2.0 technology, not the Radeon 9700. The only claim being made is that Smoothvision 2.0 supports supersampling. For all you know, ATi could release a future product incorporating Smoothvision 2.0 that implements the supersampling features, or not.. it doesn't really matter as ATi never said that any product would support supersampling, merely that the technology does.

Look at the pics: what does the TITLE of the page say? (Radeon 9700)

WHERE do you find this presentation? (Radeon 9700 Pro Specifications)

(Beyond that, and this I know is insignificant so don't get me wrong, the address to the presentation says R300.)

[T2k]

It's about promises...

No, actually, it's about prioritys........

Sorry to disagree, but it's not about priorities: it's about ATI's reliability.

Edit: typo

[Basic]

DaveBaumann:
You said that you could get supersampling on a R9700 by using D3D multisampling support.

You specify what kind of multisampling you want when creating the rendertarget. It's specified by number of samples and a "MultisampleQuality" factor. The only way I could see to get supersampling would be that they have some value on the quality parameter that means supersampling instead of multisampling.

Or how would you propose that it was available?

[Dave Baumann]

Sorry Basic, I assume you can use the D3D Multisample buffer is a similar fashion to how 3dfx used the T-Buffer.

[Colourless]

Sorry Basic, I assume you can use the D3D Multisample buffer is a similar fashion to how 3dfx used the T-Buffer.

Yes you can, provided the driver says it supports it.

[Basic]

I don't say that it can't be done with the D3D multisampling interface. Just that I don't think it is implemented. You can probably get both the quality and the performance version of their multisampling, and it does indeed seem possible to add a third quality level that means supersampling. But if that level already existed, then why not add it in the driver panel? It should be a very simple task. If it isn't there it's probably either because the mode isn't implemented, or it has problems. In either case it's not likely that you're allowed to explicitly request it when programming yourself.

So unless some ATI developer says how to get smoothvision 2.0 SSAA, I think it's a safe assumption that it's not available for developers in any way.

[Dave Baumann]

I don't say that it can't be done with the D3D multisampling interface. Just that I don't think it is implemented.

Thats how SMOOTHVISION 1.0 was supposed to be implemented - if it is a development on the original process then its likely that it would contain the the original method as well. The multisample buffer is in the caps list IIRC, so I assume a developer would be able to

Why not implement it? One possability is that the performance is so bad in comparison to MSAA they they just don't want it used. Another is that there is, as you say, some hardware bugs as, after all, it was never implemented correctly in 8500 (but, is that the case if the multisample buffers are available to the developer?).

[Ante P]

I don't understand the complaining abour performance.
I used 2x (Quality) SSAA on my 8500LE in 1152x864 on many games (as well as 16x Aniso for that matter, and of course maxed in game options) with sufficient performance.
Since the 9700 Pro has greatly increased bandwidth and fillrate as well as other optimizations I can only assume that 4x FSAA would work pretty great in "mid" resolutions. 2x SSAA + 2 or 4x MSAA should be a pretty good combination too.

[Basic]

Yes, being too slow is one of the possible problems to not expose it. And then it would probably not be exposed for programmers either.

So I still say that unless some ATI developer says otherwise, it's probably not available.

[Dave Baumann]

Since the 9700 Pro has greatly increased bandwidth and fillrate as well as other optimizations I can only assume that 4x FSAA would work pretty great in "mid" resolutions.

Its texture fillrate, which is important for SSAA, it equal to 8500 per cycle, since it only has 8 bilinear texture units.

[Ante P]

Since the 9700 Pro has greatly increased bandwidth and fillrate as well as other optimizations I can only assume that 4x FSAA would work pretty great in "mid" resolutions.

Its texture fillrate, which is important for SSAA, it equal to 8500 per cycle, since it only has 8 bilinear texture units.

isn't the texel fillrate 2.2 Gtex while it's 2.6 Gtex on the 9700 Pro?
Well it's not much but at least something and coupled with the three factors mentioned earlier at least should amount to some sort of improvement..

[Bambers]

The important part is the JGSS, which can't be done on the application side. Unless with an accumulation buffer approach, but that will certainly not give any speedup compared to earlier methods.

So it's clearly a promised feature that didn't make it. Given what andypski said here it seems as it's in the hardware. But we just can't use it.

Again, it still might have been a good thing to put other things first in the todo-list.

JGSS never turned up on the 8500 either :(, except for 2xQ and a not fully optimal 4xQ pattern in the 2nd driver set which was reduced to only under d3d no fog in later drivers with OGSS (and screwed mip boundries) for everything else. :(
I think it must have a problem in hardware. Performance certainly wasn't any issue with it as it ran just the same as 2xOGSS.

[Basic]

Actually, wasn't it a mix of JGSS and OGSS?
Textures sampled with an ordered grid, but geometry with a jittered grid.

[Ante P]

The important part is the JGSS, which can't be done on the application side. Unless with an accumulation buffer approach, but that will certainly not give any speedup compared to earlier methods.

So it's clearly a promised feature that didn't make it. Given what andypski said here it seems as it's in the hardware. But we just can't use it.

Again, it still might have been a good thing to put other things first in the todo-list.

JGSS never turned up on the 8500 either :(, except for 2xQ and a not fully optimal 4xQ pattern in the 2nd driver set which was reduced to only under d3d no fog in later drivers with OGSS (and screwed mip boundries) for everything else. :(
I think it must have a problem in hardware. Performance certainly wasn't any issue with it as it ran just the same as 2xOGSS.

Even worse was the absense of MSAA. ;)
http://www17.tomshardware.com/graphic/20010814/radeon8500-03.html

"Sophisticated Multi Sampling FSAA" turned out to be "Crappy Super Sampling FSAA"

Though 2x Quality Smoothvision + 16x Aniso still looked pretty good IMHO except on vertical edges which had massive aliasing.

[Dave Baumann]

It was never meant to be MSAA, it was meant to be 'Mutli-buffered' SSAA.

[Bambers]

Install the 7206 drivers (or xp equivalent) i think they're still on atis site. Turn on 2xAA and you'll get nice vertical and horizontal AA in opengl and d3d :). Just don't expect all modern games to run without crashing :|

[dksuiko]

I really don't mind them not supporting SS at the moment. It's not like I'd use that over MSAA, anyway. But on the otherhand, I can only imagine how many pages past 20 this thread would have been if the tables were turned and it was nVidia. :)

[SanGreal]

So when they say they support Pixel Shader 2.0 that's not really a promise either I guess?
I mean they could propably software emulate it at 0.01 fps but hey I didn't see them saying it would all be done in hardware?

Or what about not providing Open GL extention for anything?
Hey the hardware still supports it.

I don't understand your analogy. The pages in question discuss the Smoothvision 2.0 technology, not the Radeon 9700. The only claim being made is that Smoothvision 2.0 supports supersampling. For all you know, ATi could release a future product incorporating Smoothvision 2.0 that implements the supersampling features, or not.. it doesn't really matter as ATi never said that any product would support supersampling, merely that the technology does.

Look at the pics: what does the TITLE of the page say? (Radeon 9700)

WHERE do you find this presentation? (Radeon 9700 Pro Specifications)

(Beyond that, and this I know is insignificant so don't get me wrong, the address to the presentation says R300.)

The presentation focuses on the 9700, yes.. however the topics in question do not. As you well know, Smoothvision 2.0 is not 9700 specific.

[K.I.L.E.R]

We all know that SSAA is easily implemented and it was mentioned in the latest Catalyst driver interview.

We all know that they are no longer looking into it, at that time state anyway.

We also know SSAA was never promised.

We also know that the R300 is incapable of handling MSAA properly as it doesn't use MSAA as stated. There is a secret registry key that told me that MSAA is really fake SSAA.
Last bit was to check if you guys are awake. :)

[Mintmaster]

Since the 9700 Pro has greatly increased bandwidth and fillrate as well as other optimizations I can only assume that 4x FSAA would work pretty great in "mid" resolutions.

Its texture fillrate, which is important for SSAA, it equal to 8500 per cycle, since it only has 8 bilinear texture units.

Are you sure about that? If you're supersampling a polygon without textures or one texture, it would still go twice as fast on a 9700 than an 8500.

Even the downsampling at the end can be done in one cycle per pipe for the 9700 (for 2x2 SSAA) by bilinear sampling.

[Dave Baumann]

Are you sure about that? If you're supersampling a polygon without textures or one texture, it would still go twice as fast on a 9700 than an 8500.

Without textures, yes. With 1 texture and bilinear r300 is faster. With one texture and trilinear same number of samples (8 in two clocks for R300, 4 in 1 clock for R200).

R300's texture samplers are bilinear, according to the developer documentation, and require two cycles for trilinear sampling. If each of R200's units are bilinear then they will require one cycle for trilinear.

[Mintmaster]

But that always held true even without SSAA, and we can see that the 9700 is quite a bit faster per clock than the 8500 (close to 2x faster) in nearly all applications, even multitexturing apps like Quake3.

Texture fill rate is no more important to pixel performance with SSAA than without.

[Ante P]

Yes, being too slow is one of the possible problems to not expose it.

Well now that we have 4x SSAA scores from the Ti4600 (4x SSAA: 5639 3D Marks) I think it's pretty safe to say that even 4x SSAA would be pretty fast on the 9700 Pro.

[Ante P]

We also know SSAA was never promised.

Personally I still think the promise is up on their webpage since they haven't removed that presentation.
What I don't get is why they keep the presentation there, remove it or change it IMHO, no need to fool their customers.

[WaltC]

Now I see this stupidity has hit the B3D forums. Sigh...get over it people. If ATI doesn't like it, they aren't going to put it in, no matter what they said earlier or how many complaints they get--and I don't blame them. Know why? If they put it in the number of complaints would double because of all the people who wanted it who, when they got it, weren't happy with it because it was just as limited as ATI said it would be when they decided not to use it in the first place. Personally, I'm more than happy to let them decide these things as I think they probably know their hardware a little better than the armchair critics.

[dksuiko]

Personally, I'm more than happy to let them decide these things as I think they probably know their hardware a little better than the armchair critics.

How ironic... What a convenient way of looking at it, eh, Walt?

[Ante P]

Personally, I'm more than happy to let them decide these things as I think they probably know their hardware a little better than the armchair critics.

How ironic... What a convenient way of looking at it, eh, Walt?

"we decided that you don't really need FSAA at all, I asked my wife yesterday if you could see any difference in the before and after pictures and she couldn't, thus we removed the feature"

arm chair critics my butt, I'm a consumer
I paid money for a product which doesn't live up to the specifications that the manufacturer has posted on their site
nothing more nothing less

[dksuiko]

That's a response to my post?

Maybe you missed the point on what it was saying.. In any case, I just found it funny that when a similar topic is brought up against nVidia, WaltC is one of the first to jump at it.. Yet, when the situation is reversed and about ATI - he turns around and says its stupid and let ATI do what they want with their hardware since they know better than we do.

[Ante P]

That's a response to my post?

Nah, it was a "follow up" to your post, aimed at Walt.

[Tahir2]

I sent SapphireTech and ATI an email regarding this issue and await ATI's response as SapphireTech emailed me back (very quickly as well!) stating they would forward my email to ATI.