Using the SPU's to do AA

Today i was entertaining the thought of using SPU's to do AA instead of using the RSX. I'm guessing bandwidth and latency issues would be the main thing that would prevent this?. I can't remember how much bandwidth is required for 1080P60FPS, would the PS3 have enough bandwidth to get the data into the XDR ram and then get the RSX to read it back fast enough for 1080P60FPS.

I have no idea about graphic pipelines, would this even be possible on a forward renderer?.
 
AA is not an effect that you switch on and off, it changes the way the GPU rasterize and shade stuff. Saying that SPUs can take care of that doesn't make any sense, unless you don't want to use the GPU to rasterize your scene.
 
AA is not an effect that you switch on and off, it changes the way the GPU rasterize and shade stuff. Saying that SPUs can take care of that doesn't make any sense, unless you don't want to use the GPU to rasterize your scene.

Thank you for that, so I take it. It can only be used for backwards rendering? (Killzone 2).
 
AFAIK KZ2 doesn't use SPUs to shade the scene, even though it's definitely doable
 
They don't say exactly how it's done, but we have discussed many ways to do it on these forums, just use the search function.
 
AA is not an effect that you switch on and off, it changes the way the GPU rasterize and shade stuff. Saying that SPUs can take care of that doesn't make any sense, unless you don't want to use the GPU to rasterize your scene.

For traditional oversampling AA that's true, but there are several post process "fudge" techniques that gets pretty decent results. But even then, you'd probably still be better of doing it on the GPU.
 
Saying that SPUs can take care of that doesn't make any sense, unless you don't want to use the GPU to rasterize your scene.
You could do something similar to Snes9x's 2xSai or SuperEagle screenbuffer filters I suuppose..

I remember plauying Final Fantasy 6 on ym PC many years ago using these filters. My old K6-III was just about fast enough to run that game without dropping frames with a filter enabled.

An SPU ought to be screaming fast for this kind of owrk. :cool: Not sure what would happen to the image itself though.. 2D artwork could look strangely different depending on which filter was used.



Peace.
 
The possibility of using SPUs for post processing effects has been fairly widely discussed here quite some months ago. I am assuming GG are using such a technique for Killzone 2, considering how they have demoed the filters they put over their screen. I'm not 100% sure of course that they are actually using the SPUs for this and not GPU hardware, but someone else may be able to say something useful on this. Didn't one of us (was it Fran?) attend the Killzone dev meeting for this at GDC? Not sure if it was discussed, but I thought so. I'm not saying that you're going to do all your AA in post-processing, but the whole concept of post-processing I assume should have a big impact on the importance of AA.
 
The possibility of using SPUs for post processing effects has been fairly widely discussed here quite some months ago. I am assuming GG are using such a technique for Killzone 2, considering how they have demoed the filters they put over their screen. I'm not 100% sure of course that they are actually using the SPUs for this and not GPU hardware, but someone else may be able to say something useful on this. Didn't one of us (was it Fran?) attend the Killzone dev meeting for this at GDC? Not sure if it was discussed, but I thought so. I'm not saying that you're going to do all your AA in post-processing, but the whole concept of post-processing I assume should have a big impact on the importance of AA.

The presentation slides for the conference were released into the public domain as far as I am aware..

I've got them here on my desktop.. Can't remember for the life of me the url I got them from though..
 
To Squeak and Rainbow Man = Anti-Aliasing and selective edge blurring/smoothing are different things, the latter shouldn't be called AA techniques at all.
 
I disagree, applying a low pass filter after having sampled a signal is not what I'd call anti-aliasing, but something more along the lines of: "OMG!!! we didn't pre-filter this stuff, we're also undersampling it, we're screwed! let's do something!!" ;)
 
Perhaps Jaggie-reduction, or de-aliasing. De-aliasing is perhaps the most appropriate as it implies the aliasing exists first and then you have to deal with it in some dubious post process, in contrast to anti-aliasing suggests you employ techniques to remove aliasing. Although given the similarity between terms, they may get just as confused as flammable and inflammable.
 
I disagree, applying a low pass filter after having sampled a signal is not what I'd call anti-aliasing, but something more along the lines of: "OMG!!! we didn't pre-filter this stuff, we're also undersampling it, we're screwed! let's do something!!" ;)

I disagree.
Aliasing is a known issue and I'm sure we agree on that one (whenever a sampling degrade the original signal basically).
Anti-aliasing could be defined, especially in computer graphics, by "whatever can make it nicer".
Sub-sampling is just one branch of AA (and the most "clean" I agree with that).

I know I know, we are playing with words now ;)

I don't know the detail of the method used in Killzone but it wouldn't it make sense to use the SPUs to rasterized a high-definition depth buffer and use that for FSAA post-processing of a screen-resolution color-buffer? (sorry if it has been discussed already).
 
Anti-aliasing could be defined, especially in computer graphics, by "whatever can make it nicer".

Could we go for a better definition of removing aliasing artifacts at least.

The problem with postfilter alias removal is you can't really fix subpixel width/height polygons well which is the most annoying type of aliasing artifact imo.
 
Edge AA like on DS or N64 is actually just the last and first visible pixel in the scanline of the polygon being filled, alphablended to various degrees according to the angle of the edge.
It gets a result that is very close to "real" AA though.

Oversampling is of course the nicest way to do AA, but beggers can't be choosers and we are beggers when it comes to doing AA.
"Something" is better than nothing.
The discussion as to whether the term should be "AA" is only a semantic one, and isn't really important in this discussion.
 
Back
Top