An alternative to super sampling, MLAA(Morphological AA)

[StripTheSoul]

that's true.

[skid]

(10-21-2010, 02:57 AM)Xtreme2damax Wrote: However 3x EFB scale causes a 5 FPS drop or more in some scenes, while no slowdowns in OpenGL with the option.

I did not study this area of the code closely, but both dx9 and ogl seemed to be doing the same thing (which was my goal in aiding the merge of the video plugins). At a guess, the speed difference might be in the dx/ogl driver or the dolphin shaders.

[Xtreme2damax]

I'll keep testing, maybe it was a fluke or something.

I think implementing support for AA that doesn't require a graphics card from the future might be something to consider. SSAA is very demanding and even my 460 GTX and possibly other modern graphics cards aren't able to handle it. SSAA almost works fine by itself, but when using a higher EFB scale it really starts taking a hit on whatever system.

[StripTheSoul]

+ there's not even a 2x MSAA option anymore which would for most be a good compromise between image quality and performance...

[Squall Leonhart]

(10-20-2010, 08:42 PM)ector Wrote: MLAA can indeed be easily added to Dolphin assuming there's a source code implementation out there with a license compatible with the GPL, which seems to be the case?

once the dev has worked out the last few issues he will put the source up.

[fagoatse]

It's already working in dolphin with series 5 cards(dx9 only, drivers' inf needs to be modded).
It apparently works in pcsx2 too.

[Link_to_the_past]

(10-21-2010, 01:59 AM)Xtreme2damax Wrote: Any idea why high EFB scale is causing slowdowns with DX9 and not OpenGL? Switching EFB scale to 3x, then back and forth a bit will have it running at full speed at even 3x EFB scale, EFB scale even appears to be working so I'm guessing that at least might be a bug.

If you are using rivatuner try overclocking just the shaders clock of your card. It is the only thing that matters with efb scale, a 10% increase in shader's clock can give you a 10% increase in framerate if possible. It will probably make up for those 5 fps that are dropped. Btw with my 9600gt i am getting similar slowdowns with both opengl and directx9 plugin when upping the efb scale, there is no "bug" between plugins.

[zurginator]

I'll pass.

Orange is where I noticed 1 is better than the other, red is where it's blatantly obviously worse.

Granted, this is a jpg, but it's still pretty obvious.

[Squall Leonhart]

AMD's implementation of MLAA is fail apparently

Quote:jK says:
October 24, 2010 at 4:22 am

This method isn’t new. It was/is the only way to do AA in Deferred Shading Renderers. But this implementation is really one of the worst I have seen because you usually take the pixeldepth (and normal) into account and just blurrs on geometry edges. AMD instead blurrs on colorspaces edges, causing blurred GUIs and missing detail on textures.
Somehow this reminds me a bit of the N64

[zurginator]

(10-21-2010, 09:14 AM)Xtreme2damax Wrote: I'll keep testing, maybe it was a fluke or something.

I think implementing support for AA that doesn't require a graphics card from the future might be something to consider. SSAA is very demanding and even my 460 GTX and possibly other modern graphics cards aren't able to handle it. SSAA almost works fine by itself, but when using a higher EFB scale it really starts taking a hit on whatever system.

Filters have a performance hit? I was not aware.

SSAO = best filter ever.