New selective supersampling demo

[Reverend]

Realistically, though, how feasible selective algorithms such as the ones mentioned above? How many "man hours" do they take to code and debug for in addition?

Humus' demo showed the benefits of what he did on the ground, in a single scene, with nothing else in consideration.

I ask because it seems both Carmack and Sweeney are weary of spending their time on what seems to me an important IQ issue. It seems it only took Humus a little while to come up with his selective solution based on branching via the gradient parameter.

I don't think it's about the time Humus spent coming up with this.

Perhaps Carmack and Sweeney are weary of spending time on techniques only serve to improve the image quality to performance ratio.

Every developer have their own priorities when it comes to improving IQ in their games. Heck, even the IHVs may have a say in this and all this have nothing to do with AA!

AA have never been a priority for devs, other than slapping on AA support by querying resident card's DX CAPS.

Lastly, perhaps the word you chosed ("weary") can be substituted by "leery".

Like I said, Humus shows us what happens. Not what we should expect to see in games. Sad and frustrating as it is

 

[Ailuros]

2xMSAA and 2xSSAA do have the same memory consumption. I'm not sure where fellix said otherwise though.

He mentioned the overflowing of Xenos' eDRAM capacity.

 

[Luminescent]

Lastly, perhaps the word you chosed ("weary") can be substituted by "leery".

You're right, the word I meant was "leery"; I appreciate the feedback. I still stand in agreement with Humus' post a few posts back.

 

[Humus]

Well, to tell you the truth, Humus, I really like this method better. I mean, it's less accurate, but it should be much faster:
http://developer.nvidia.com/object/mipmapping_normal_maps.html

The idea is to basically take the length of the filtered bump map, and use that as a measurement of how quickly the surface normal is varying over the pixel.

Ideas like that are of course an option, and if they work out nicely it's often a better solution, even though they only apply to a particular class of aliasing. Personally I never had any luck with that particular method though. I could have made a mistake, but I spent a fair amount of time on it and never got results that would justify the use of it.

 

[3dcgi]

He mentioned the overflowing of Xenos' eDRAM capacity.

The same overflow will happen with MSAA and SSAA.

 

[flf]

You're right, the word I meant was "leery"; I appreciate the feedback. I still stand in agreement with Humus' post a few posts back.

Perhaps you meant "wary" rather than "weary."

 

[KimB]

Ideas like that are of course an option, and if they work out nicely it's often a better solution, even though they only apply to a particular class of aliasing. Personally I never had any luck with that particular method though. I could have made a mistake, but I spent a fair amount of time on it and never got results that would justify the use of it.

Interesting. And I do have to mention that even if one could get something like the method I posted to work well, the one you posted has the distinct advantage of being applicable to any form of surface aliasing.

 

[fellix]

The same overflow will happen with MSAA and SSAA.

Well, you'r right about memory footprint - it's almost 1:1 match between sample-count identical modes (2x2-vs-4xMS, 4x4-vs-16xS) but the performance hit degree is "impressive" for SS modes.

 

[KimB]

It is a 1:1 match. MSAA is more compressible, but due to its nature, you still need to reserve the full buffer amount.

 

[Xmas]

Of course you can do better in most cases if you store additional samples separately, but then it gets much more complicated. But it can be done and has been done.

 

[Ailuros]

The same overflow will happen with MSAA and SSAA.

Hence my initial question.

 

[Ailuros]

Well, you'r right about memory footprint - it's almost 1:1 match between sample-count identical modes (2x2-vs-4xMS, 4x4-vs-16xS) but the performance hit degree is "impressive" for SS modes.

If you're CPU bound 2x samples aren't that many, to make a huge difference. Of course is adaptive/selective Supersampling in combination with MSAA and AF way more attractive.