Improved AA/AF, Video Decoders, Memory Interface Future GPUS

[ChrisRay]

There's been alot of talk about improving shader performance. Uh oh here comes the "Extreme Pipeline" jokes. But moving on. The other important aspects of a high end card.

1) Improved Anti Aliasing, quality And performance.

Given the rapid changes and focuses brought towards shader performance. I have noticed that over the past year or so. Since the r300/Nv3x. The actual compression and AA mechanism havent changed much.

Despite the Nv4x using 4x RGMS these days. Has the compression really changed from the Nv3x? Since 4x OGMS still works on the Nv4x you can see practically no difference in performance between the two so obviously anti aliasing performance hasnt been. Its obvious the r300/r420 is exactly the same.

Has the focus on anti aliasing. Given the current multisample limitations and and the addition of the fact that 4 sample AA seems to offer the best quality/performance stand off at this point. (with relatively minor improvements beyond 4x on both hardware from an IQ standpoint) Have we reached a dead end on AA improvements?


2) Anisotropic Filtering.

This seems to be. (IMO) taking a turn for the worse. There has been alot of arguing over whether adaptive nature of these new products is a good or a bad thing. So I'll try to make this as simple as possible. Given the current products ability to perform these functions at relatively fast speeds. And the current implementation of both vendors. Can we expect to see much in terms of actual quality improvement? We've already seen enough corners cut for performance. But quality is where we need to see improvements right now. With each IHV trying to one up each other in terms of performance. I am worried we've seen a dead end in regards to quality improvements with AF.

4) New Features

3dc, Ultra shadow, ect are all features added onto the new GPUS, which so far have seen relatively minor use. However that use could change given more time. Can we expect to see more and more unique gadgets? This brings us back to innovation verses technology. And the rewards of it as discussed in another thread. Some people have expressed that innovation comes with a lack of reward for the IHV.


5) So Whats next?

My guesses have left me a bit bewildered on what we should expect this coming next year. With DirectX 9.0C and SM 3.0 being the basic foundation of what we can expect technology wise the next year until the next implementation of DirectX next. It leaves me wondering if technology is going to provide anything new and worth while to current hardware owners of the latest generation. With the move to DirectX 9.0 there was a real reason to upgrade to modern GPUS. But given the current progress of DirectX. And a lack of new features on modern GPUS. Has the technology felt a bit less interesting?

Even if the next GPUS are 40% Faster than the ones we have. If they dont provide any compelling features/new technology. Why buy them? It's not like they are going to be slow. Hell its not like the r300 is slow.

 

[Apple740]

I am worried we've seen a dead end in regards to quality improvements with AF.

I think we can blame this (for a part) to all the reviewers, well, most of them. In none of the reviews from last gen cards i've read one single issue regarding texture shimmering (well, maybe one or two who did), while it's raining complaints in the forums all over the net. What we see in the reviews is a couple of screenshots and that's it, but shimmering can only be seen in motion. I think the reviewers don't have the balls to bring such issue's to the public, afraid of getting no card's to review anymore, or they simply will not bash the brand they love.

It's obvious that the vendors throw everything in speed this gen, and after that they go for quality. Hope one day this will turn around.

 

[Tim Murray]

Have we reached a dead end on AA improvements?

Absolutely not. There are just better things to worry about at the moment.

Give it a few years...

 

[ChrisRay]

Have we reached a dead end on AA improvements?

Absolutely not. There are just better things to worry about at the moment.

Give it a few years...

I have heard of several implementations of ways to improve AA. But the problem is. IHVS dont seem to be taking any interest in worrying about them. I really do hope we see some improvements to the current methods. IHVs have been focusing purely on shader revisions. And given the pressure of DirectX and microsoft to have these fully functioning GPUS. Theres not much room left in transistor budget to innovate.

 

[tEd]

On the AA site i think the next step should be to make it work with fp blending etc. but i'm not sure when to expect this probably not with the next gen.

I don't think quality will improve much if at all to next gen. I personally would like to have something like 16x super-scene MSAA

On AF agree it is going more to worse quality than to better lately but it actually may turn around again as alu to tex ration in games will steadily increase it will be possible to have higher quality Filtering without too much of a performance impact. Hopefully IHV will also think so

 

[Tim Murray]

On the AA site i think the next step should be to make it work with fp blending etc. but i'm not sure when to expect this probably not with the next gen.

Bingo. When they're putting so much emphasis on shaders, it's hard to justify any improvements to antialiasing when sometimes it won't even work.

Now, where's Uttar? We had lots of silly talks about this topic.

 

[AlNom]

(what happened to 3) )

 

[Acert93]

(what happened to 3) )

#3 was not rendered properly because AA was enabled

 

[hovz]

Re: Improved AA/AF, Video Decoders, Memory Interface Future

There's been alot of talk about improving shader performance. Uh oh here comes the "Extreme Pipeline" jokes. But moving on. The other important aspects of a high end card.

1) Improved Anti Aliasing, quality And performance.

Given the rapid changes and focuses brought towards shader performance. I have noticed that over the past year or so. Since the r300/Nv3x. The actual compression and AA mechanism havent changed much.

Despite the Nv4x using 4x RGMS these days. Has the compression really changed from the Nv3x? Since 4x OGMS still works on the Nv4x you can see practically no difference in performance between the two so obviously anti aliasing performance hasnt been. Its obvious the r300/r420 is exactly the same.

Has the focus on anti aliasing. Given the current multisample limitations and and the addition of the fact that 4 sample AA seems to offer the best quality/performance stand off at this point. (with relatively minor improvements beyond 4x on both hardware from an IQ standpoint) Have we reached a dead end on AA improvements?


2) Anisotropic Filtering.

This seems to be. (IMO) taking a turn for the worse. There has been alot of arguing over whether adaptive nature of these new products is a good or a bad thing. So I'll try to make this as simple as possible. Given the current products ability to perform these functions at relatively fast speeds. And the current implementation of both vendors. Can we expect to see much in terms of actual quality improvement? We've already seen enough corners cut for performance. But quality is where we need to see improvements right now. With each IHV trying to one up each other in terms of performance. I am worried we've seen a dead end in regards to quality improvements with AF.

4) New Features

3dc, Ultra shadow, ect are all features added onto the new GPUS, which so far have seen relatively minor use. However that use could change given more time. Can we expect to see more and more unique gadgets? This brings us back to innovation verses technology. And the rewards of it as discussed in another thread. Some people have expressed that innovation comes with a lack of reward for the IHV.


5) So Whats next?

My guesses have left me a bit bewildered on what we should expect this coming next year. With DirectX 9.0C and SM 3.0 being the basic foundation of what we can expect technology wise the next year until the next implementation of DirectX next. It leaves me wondering if technology is going to provide anything new and worth while to current hardware owners of the latest generation. With the move to DirectX 9.0 there was a real reason to upgrade to modern GPUS. But given the current progress of DirectX. And a lack of new features on modern GPUS. Has the technology felt a bit less interesting?

Even if the next GPUS are 40% Faster than the ones we have. If they dont provide any compelling features/new technology. Why buy them? It's not like they are going to be slow. Hell its not like the r300 is slow.

why do u think im sticking with my r3xx

edit - to address your points, i think theres MUCH more important things than aa or af in the near future as far as advancing graphics. aa and af only matter today because developers code games for low end hardware, so the power of a good video card is going to waste.

 

[ChrisRay]

#3 was actually there. I just deleted it it because I felt one of my other numbers properly addressed it

Anyway, I dont think AA ((quality for instance)) needs improved at this point. I do feel that other than these things however. Modern hardware really has nothing to look forward in regards to features. Since we are limited by the DirectX featureset. AA/AF and little features are really all the high end cards currently have to work with that really differate themselves from others. Heck Had the r300 not been so great at AA/AF performance. Alot of people probably wouldnt have upgraded so quickly then.

These are all issues non related to pixel pipeline/vertex setup. Which havent moved much in the last 3 years IMO. Which is why I question their decadent state.

 

[AndrewM]

I think in next gen we'll see FSAA for FP targets. People will definately complain if this isnt there, because all next gen titles will be using FP render targets.

We will definately see more shader ops too, which should allow for decent in surface AA - the same stuff that JC and TS have been talking about.

Other features, who knows.

We still need more memory bandwidth.

 

[jvd]

Well with speed increases we should get fsaa increases too .

Its very hard for me to see jaggy edges with 6x fsaa on . The problem is i can't use it in many games . Even with the x800xt pe of the 3 big tittles out this year (doom3 , eq2 and half life 2 ) that push the graphical boundrys i can only use it on one .

Also many don't talk about temporal aa . But it is a feature that greatly enhances image quality.

Last point is its the same problem with aniso.

The reviewers are at fault .

They wont budge from 4x fsaa . If 4x fsaa is playable why not move up to the next bracket . 6x and 8x fsaa should be benchmarked and reported on . When the reviewers start addressing higher modes of fsaa the ihvs will deliver even faster and higher modes .

If nvidia sees thats reviewers are giving ati cards the win because 6x fsaa is playable in many games they will find a way to get a mode with the same speed and image quality . If ati sees that 8x from nvidia is getting them the wins on benchmarks they will come out with an 8x .

 

[Ailuros]

I am worried we've seen a dead end in regards to quality improvements with AF.

S3 DeltaChrome has a totally different approach and to that a quite interesting one. Trouble is that S3 is amongst the small vendors (with very small market penetration) and DC must have some let's call it bandwidth constraints overall.

Apart from that it does offer 16x sample non-angle-dependent AF.

As far as antialiasing goes or any other major improvements in feature-sets and what not concerns, I personally do not expect major changes prior to WGF2.0 and possibly much wider memory interfaces sometime in the future.

Performance-wise I'd love to know when and under which presuppositions single-cycle 4xAA could be possible for instance. It would set a landmark as to when faster and higher quality AA would set in (think 4xAA with equivalent to current 2xAA performance penalty and 8x sparced AA via a sample loop).

 

[Megadrive1988]

in order for games to look more like CG, i believe we'll need 32x FSAA if not at least 16x at extremely smooth framerates. i know, probably wont happen until 2007 or later.

 

[rwolf]

Actually the NEXT big thing was getting the big iron hardware into the mainstream like the 6800GT and the X800XL. The more high end hardware out there the more likely developers will raise the bar. If developers still produce games that have to run Geforce 3 and Radeon 8500 hardware they are less likely to raise the bar.

 

[Arun]

Now, where's Uttar? We had lots of silly talks about this topic.

I'm never where you want me to be, eh?

Actually the NEXT big thing was getting the big iron hardware into the mainstream like the 6800GT and the X800XL.

Are you saying NVIDIA and ATI should give billions of dollars to charities, put that in their "longterm business plan spending" and hope everyone in the world becomes so rich they can just keep buying their chips nonstop?
Sorry, I know that wasn't what you meant, just couldn't resist I do understand your point, but I don't agree with it: mainstreaming the performance and featureset of the past generation of products is perfectly natural in the industry (yes, there are exceptions like the GF4 MX, but they are still mainstreaming SOME features; just they slow down the process). I don't think it's very sensible to be calling that the "NEXT big thing".
As for what I think will matter next... Well the big question, once Longhorn/SM4.0./WGF 2.0. are out, is what NVIDIA and ATI are going to focus on. Feature differences will become less and less noticeable, and we won't be going in "other directions" for quite a bit longer. Whichever company controls the market at that point is likely to control it for at least one more decade ala Intel in my opinion, so we're in an interesting era for sure.
Regarding video processing, I think the real revolution is gonna happen when instead of just following industry standards and implementing them in the hardware, IHVs actually innovate and create their own, much more powerful standards. I suspect we can manage a better compression ratio if the different stages of the architecture are unified and decompression would potentially have several GFlops available every single *frame*. Traditional compression algorithms aren't particularly GPU-oriented, which is why NVIDIA decided to create a separate video processor instead of just using the PS architecture in my opinion.
As the standards stabilize, it seems likely to me there will be a few "big" engines licensed a lot more frequently than today, there'll be less and less reason to reinvent the wheel. Perhaps the IHVs will even bother making their own engine, with a standard (D3D maybe?) render path and a proprietary one. That'd be quite interesting really; things like Glide are unthinkable nowadays, they'd be doomed before even being announced, but fullblown engines optimized for one IHV, if they're cheap to license, might have a future. It'd be a big risk though, so I'd honestly be surprised if the happened, although I think it might be good for the industry overall.


Uttar

 

[Fodder]

Re: Improved AA/AF, Video Decoders, Memory Interface Future

aa and af only matter today because developers code games for low end hardware, so the power of a good video card is going to waste.

You can pump as many polygons as you like, but it'll still look like a jaggy blur without AA and AF.

 

[jb]

Chris,

I am rather puzzeled by this post. You publicly blaset me on NV forums for making this very suggestion. Then you come here and you ask more detials? Anyways:


1) I dis-agree that "x4 AA is enough" (paraphrased your idea not your exact words). I can still spot Alasing at 1600x1200 x4AA in the game with out needed to blow up the image to x400. I am sure that others here have better and worse of a time doing the same thing. But to say other forums of AA offer minor IQ improvement is plain wrong. Again different people can see things better/worse than you can. Also things like "Temporal AA" like what ATI had on the R3xx+ are other types of AA improvements is what we may see in the future. I do however agree that x4AA is the sweet spot for the moment.

2) I agree that we seem to have stalled this time around. However here's hoping that IQ will get better (or at least user options will to disable will be built into drives on day one and not x months later).

4) This is the old cart before the horse agurement. You need the IHVs to add the hardware so the developers can code for it. The probably is when marketing and fanboy's get invloved. Features that can remain vendor netual that add value (and ease of use) are the best. These features adds value to the end user the day the buy the card vrs them playing the waiting game in hopes some game uses feature X before they upgrade.

5)

It leaves me wondering if technology is going to provide anything new and worth while to current hardware owners of the latest generation.

This is the Question you always have to ask every year. Did the tech really jump from the R9700 to the R9800? Not really. What about NV30 to 35 (performane aside)? Again not really. This year ATI showed a minor bump, NV had a major break through). This is a question we have yet to answer buy ourselves. It will depend on which features are added, just HOW fast it is vrs last year model and what are the needs of the user. No matter what, fanboys will flock to the NVnews forum to talk about just how "733t" the new NV cards while ATI fans will do the same at Rage during next year....

 

[Tim Murray]

I DON'T think T-AA is the future, or at least not for a long time. it's nice to have for older games, but considering you have to have vsync on, it's not ideal by any means.

 

[ChrisRay]

Then you misunderstand the nature of my post. I dont think we need better AA quality right now. I think we need better compression and performance values. Even now 4x MSAA performance isnt what I'd call ideal on either hardware. (On performance factors)

If you read my post. I am interested in performance strides in this area. Not quality. While AA is most certainly usable on modern/high end hardware it currently still remains a niche in which people with only high end cards will use it. The focus on my prior point with AA, I came here for clarification on how the future of AA will look. And much to my expectations. It's not going to change. But we have actually made very little effort towards it. In its current form. AA will always remain a niche for PC graphics. Which is a shame because very few people to this day have learned to appreciate it.

Innovation Verses reward. ((Faster AA)

Ailuros actually has some good ideas. Perhaps single cycle 4xAA could be a possibility?