NVIDIA Researching Subpixel Reconstruction AntiAliasing - Answer to AMD's MLAA?

[caveman-jim]

AMD have received lots of praise for their new implementation of the Morphological Anti-Aliasing technique, introduced in October with the AMD Radeon HD 6800 series and now due to to be available on the AMD ATI Radeon HD 5800, 5900 as well as the AMD Radeon HD 6800/6900 series in Catalyst 11.2.

NVIDIA, not known for letting the grass grow under their feet, appear to working on a similar technique that they will discuss at the i3D conference in San Francisco, held from February 18th-20th.

Subpixel Reconstruction Antialiasing (SRAA) combines single-pixel (1x) shading with subpixel visibility to create antialiased images without increasing the shading cost. SRAA targets deferred-shading renderers, which cannot use multisample antialiasing.

SRAA operates as a post-process on a rendered image with superresolution depth and normal buffers, so it can be incorporated into an existing renderer without modifying the shaders.

In this way SRAA resembles Morphological Antialiasing (MLAA), but the new algorithm can better respect geometric boundaries and has fixed runtime independent of scene and image complexity.

SRAA benefits shading-bound applications. For example, our implementation evaluates SRAA in 1.8 ms (1280x720) to yield antialiasing quality comparable to 4-16x shading. Thus SRAA would produce a net speedup over supersampling for applications that spend 1 ms or more on shading; for comparison, most modern games spend 5-10 ms shading.

We also describe simplifications that increase performance by reducing quality.

The techniques looks set to offer a compromise between MSAA and SSAA quality at a fraction of the performance cost of SSAA techniques, while applying to all render engines - especially those using deferred render, unsuitable for traditional MSAA.

Source - NVIDIA

[pISSANt]

Sweet! That is the only thing I wish my GTX 570 could do vs. an ATI/AMD card.

[acroig]

Excellent news for nV owners.

[Ristogod]

Not too shabby. Can't wait!

[Sasquach]

i thoguht they were against shader based AA techniques

[SIrPauly]

SRAA operates as a post-process on a rendered image with superresolution depth and normal buffers, so it can be incorporated into an existing renderer without modifying the shaders.

I would like to learn more about this.

[fellix_bg]

Basically, SRAA links the MLAA's post-process method with real sub-pixel coverage information for much more precise image aliasing detection, with the same or even lower performance hit, and broad title compatibility.

[noko]

Originally Posted by fellix_bg

Basically, SRAA links the MLAA's post-process method with real sub-pixel coverage information for much more precise image aliasing detection, with the same or even lower performance hit, and broad title compatibility.

Post process usually means after rendering, in frame buffer. So in the frame buffer, I am assuming using special floating point format for the additional information, sub-pixel information will be present as well for a MLAA type resolve using sub-pixel information for like polygon edges refining where the process will occur. Now if the sub-pixel resolution is result of super sampling, as in frame buffer is larger then view I can't see how performance wouldn't be similar to supersampling, maybe 2x SSAA will MLAA. Still more info is needed plus actual results to see how well Nvidia new AA works out. MLAA is rather compatible (with varying results) with most everything which makes it kinda kick ass. Hopefully SRAA will be similar plus better in quality. AMD MLAA combined with other types of Multisampling AA, in a way does just this, except you have to be able to do Multisampling first so hopefully SRAA can be combined like MLAA with other types of AA.

[Humus]

Originally Posted by acroig

Excellent news for nV owners.

More like excellent news for game developers. This doesn't sound like something that could be easily made into a driver override, but I'm sure there are many game developers that would be interested in seeing what this is all about and whether it will work for them. I for one am looking forward to hearing more about it.

[acroig]

Originally Posted by Humus

More like excellent news for game developers. This doesn't sound like something that could be easily made into a driver override....

Are you saying nV will have problems forcing it on at the driver level? Seems strange for them to develop this if they can't force it on but I'm no coder so my understanding of it could be flawed.

[razor1]

yep where would the sub-pixel coverage information come from?

[caveman-jim]

Originally Posted by acroig

Are you saying nV will have problems forcing it on at the driver level? Seems strange for them to develop this if they can't force it on but I'm no coder so my understanding of it could be flawed.

It would be an AA option that was availble for Developers to buy or be given through TWIMTBTP, so it would detect NVIDIA hardware and offer the SRAA option in the game menu.

Looks like NVIDIA learned from the Batman 'standard MSAA' debacle. Hopefully AMD are licensing or offering MLAA to developers, for use with their hardware.

[razor1]

Originally Posted by caveman-jim

It would be an AA option that was availble for Developers to buy or be given through TWIMTBTP, so it would detect NVIDIA hardware and offer the SRAA option in the game menu.

Looks like NVIDIA learned from the Batman 'standard MSAA' debacle. Hopefully AMD are licensing or offering MLAA to developers, for use with their hardware.

hmm no it sounds like some of the data that is needed for this solution has to made by the graphics engine itself.

[caveman-jim]

Originally Posted by razor1

hmm no it sounds like some of the data that is needed for this solution has to made by the graphics engine itself.

hmm no i think you didn't understand what I said. the solution would be included at development time as a value-add from nvidia, not as a driver option. so obviously it would get the data from the graphics engine as its part of the engine itself.

[razor1]

Originally Posted by caveman-jim

hmm no i think you didn't understand what I said. the solution would be included at development time as a value-add from nvidia, not as a driver option. so obviously it would get the data from the graphics engine as its part of the engine itself.

The thing is, its not something that has to be part of the TWIMTBTP or a developer has to buy it. Same thing goes with MLAA, the papers will be released free of use, just like any of their other coding techniques. And for MLAA here is a good paper on it.

http://visual-computing.intel-resear.../mlaa/mlaa.pdf

You really think it was AMD that created MLAA?

[caveman-jim]

You really think I don't know where MLAA is from? tard.

[razor1]

So how do you think AMD can license something if they don't even own it?

[Humus]

Originally Posted by acroig

Are you saying nV will have problems forcing it on at the driver level?

Yes.

Originally Posted by acroig

Seems strange for them to develop this if they can't force it on but I'm no coder so my understanding of it could be flawed.

If the technique runs better on Nvidia hardware than AMD, then it's in their interest to promote it to developers. IHVs come up with new rendering techniques all the time. If the technique isn't inherently faster on the inventor's hardware, they will at least make sure to come up with a particular implementation of it that will run faster than on the competition and hand that over to developers.

[caveman-jim]

Originally Posted by razor1

So how do you think AMD can license something if they don't even own it?

What makes you think they don't own their implementation of MLAA? What makes you think they can't develop MLAA for a game engine and not own it?
How were NVIDIA able to claim MSAA for Batman was thier proprietary IP and restricted to their hardware?

[acroig]

Originally Posted by Humus

If the technique runs better on Nvidia hardware than AMD, then it's in their interest to promote it to developers. IHVs come up with new rendering techniques all the time. If the technique isn't inherently faster on the inventor's hardware, they will at least make sure to come up with a particular implementation of it that will run faster than on the competition and hand that over to developers.

Thanks, I get it now.

[razor1]

Originally Posted by caveman-jim

What makes you think they don't own their implementation of MLAA? What makes you think they can't develop MLAA for a game engine and not own it?
How were NVIDIA able to claim MSAA for Batman was thier proprietary IP and restricted to their hardware?

See algorithm and actually code can be copyrighted. But it doesn't really matter, because the math can be done a thousand ways to the sun type deal, that's why when someone gets IP for code, its done in a black box. To make it easier there are papers out there and source code available for MLAA, so it doesn't make any sense to license something like that. It would be a waste of money and time.

Now nV didn't own the MSAA process its IP is on the code.

Now AMD can do this, but isn't not worth it for them, because developers can pick it up from the papers that are already available and that would work on all cards.