Dolphin, the GameCube and Wii emulator - Forums

(10-27-2010, 01:37 PM)MFZ Wrote: [ -> ]

(10-27-2010, 05:08 AM)[SS] Starscream Wrote: [ -> ]This is no AA and 3x efb.



4xAA and 1x EFB



Not much difference if any.

What is your AA setting set to in your graphic card's control panel ?

Catalyst Control Center is set to default, that's letting applications handle AA.

Quote:I beg to differ, there's barely any jaggies at 2x or 3x in the screenshots I posted.

The 3x efb scale screenshot clearly still had loads of aliasing. You could tell it was not anti-aliased. It's also important to note that the type of game will make a difference since efb scale is only going to improve areas affected/covered by the efb.

Take WW for example. The efb in WW is used to apply a DOF effect to make distant areas look blurry. Of course blurry is different than pixelated. My screen resolution is 1366 x 768 right now since I'm using an hdtv as my monitor for this system, at that resolution 2x efb would be the appropriate amount (1280 x 960 efb resolution). With efb scale set to 1x distant areas look very pixelated. With 2x they look blurry, like they should. With 3x they look clear and sharp because the the efb is being created at a high resolution (1920 x 1440) blurred, then scaled down to my resolution which gets rid of the blur. THis is technically inaccurate but I like the look of it. However, regardless of what I set the efb scale to near by areas and my character don't look any different since they aren't part of the efb in this particular area (outset island). This would change if I entered a "hot" area like dragon roost where efb effects would also be applied to the character (the whole screen actually). To clean up the aliasing on link and nearby areas in outset island I need to use SSAA no matter how high I set the efb scale because the efb won't affect them. You see what i mean? So basically at outset island:

1x efb + No SSAA = Link is clear but has aliasing. Distant areas are pixelated and have aliasing.

2x efb + No SSAA = Link is clear but has aliasing. Distant areas are blurry with slight aliasing.

3x efb + No SSAA = Link is clear but has aliasing. Distant areas are clear, no aliasing.

1x efb + 4xSSAA = Link is clear, no aliasing. Distant areas are pixelated with slight aliasing.

2x efb + 4xSSAA = Link is clear, no aliasing. Distant areas are blurry, no aliasing.

3x efb + 4xSSAA = Link is clear, no aliasing. Distant areas are clear, no aliasing.

2x efb scale does not have the same effect as 2.25xSSAA. They do serve different purposes all together. You cannot use that as a justification because it's simply not true. Now I realize that this may be difficult to believe when playing a game like super mario galaxy since the efb in that game covers the entire screen, but trust me, they have two different purposes.

(10-27-2010, 09:20 PM)[SS] Starscream Wrote: [ -> ]

(10-27-2010, 01:37 PM)MFZ Wrote: [ -> ]

(10-27-2010, 05:08 AM)[SS] Starscream Wrote: [ -> ]This is no AA and 3x efb.



4xAA and 1x EFB



Not much difference if any.

What is your AA setting set to in your graphic card's control panel ?

Catalyst Control Center is set to default, that's letting applications handle AA.

Gotcha, I was just curious, man.. cause those pics do look very similar.

(10-28-2010, 03:32 AM)NaturalViolence Wrote: [ -> ]

Quote:I beg to differ, there's barely any jaggies at 2x or 3x in the screenshots I posted.

The 3x efb scale screenshot clearly still had loads of aliasing.

Well just so we can end this, here is a screenshot with SSAA 4x and EFB scale disabled, there is about the same or slightly more aliasing as when the EFB scale is set to 2x and 3x. I would imagine that 2.25x SSAA would look a bit worse.

I don't think EFB scale has much to do with the efb, it's basically changing the internal resolution of the textures which also smooths them out much like AA does. Just talk to rodolfoosvaldobogado like I suggested, he should be able clear it up since he knows better than any of us.

By the way, compare them side by side and it should be apparent. There is slightly more jaggies with 4x SSAA than 2x Internal Resolution, if not the same amount.

Quote:I don't think EFB scale has much to do with the efb, it's basically changing the internal resolution of the textures which also smooths them out much like AA does.

You don't think the efb scale has much to do with the efb!!!?!?! Xtreme I am going to pretend I did not see that since you are normally one of the more intelligent members of this forum. So I propose a simple test that will prove this statement false. Disable the efb and change the efb scale, it will have no effect because the efb scale literally scales the efb. Super Mario galaxy is a bad game to use for an example of this since the efb is used everywhere in that game. For a game like that, yes, efb scale is more important. But don't say that we don't need SSAA because raising the efb scale accomplishes the same thing. That's just flat out wrong.

Not to mention SSAA won't work properly unless you have efb scale set to fractional.

There's a reason the name was changed by the way.

Internal resolution is more fitting and proper since it has more to do with the internal resolution of the textures. Just like the internal resolution feature in Gsdx for PCSX2, same thing basically. It can achieve a similar result as AA which is all I was saying. I know you don't know me that well yet, but once you do you will realize that I have difficulties wording some things properly which is related to my disability so I am easily misunderstood or misconstrued. I also don't function that well when I am tired which is a contributing factor to errors in my typing.

Quote:Not to mention SSAA won't work properly unless you have efb scale set to fractional.

I was just proving that similar quality to 2.25x SSAA is achievable with efb scale. The whole discussion of this thread revolves around re-implementing 2.25x SSAA since some people can't use 4x SSAA. I don't see why setting the internal resolution to fractional would be needed for SSAA to work properly. I thought this whole discussion was about SSAA's superiority over efb scale, so why enable it if SSAA is just fine and can do super awesome jaggie killing on it's own?

Rodolfoosvaldobogado was the one that stated that 2x internal resolution produces similar quality as 2.25x SSAA, ask him, that is why he didn't bother implementing it again despite requests and that was the explanation he gave. Maybe I am a bit off on the specifics, it was either 2x, fractional, or integral resolution that was equal to the quality of 2.25x SSAA. I'm quite sure he knows what he is doing and talking about better than any of us in this thread. I'll have to see if I can find where he explained this and posts the link, I believe it was in either an issue or one of the commits where someone commented on re-implementing 2.25x SSAA because 4x and 9x was too slow for them.

Quote:I thought this whole discussion was about SSAA's superiority over efb scale

Once again. This has nothing to do with superiority. Neither is superior, they are different, they do different things, one is not an equivalent for the other and neither is a good substitute for the other.

Quote:that is why he didn't bother implementing it again despite requests and that was the explanation he gave.

That may have been an added excuse but the main reason was because it didn't work properly and caused problems with some games since you are using a sample ratio that is not 1:1 when downscaled. I'm pretty sure he even talked about it in the committ where he removed it. And multiple people have talked about it and confirmed various problems with 2.25x SSAA that don't exist with 4x or 9x on various threads. 2.25xSSAA is a stupid idea simply because it's not integral, 2x2 is integral, 3x3 is integral, 1.5x1.5 is not.

Quote:I don't see why setting the internal resolution to fractional would be needed for SSAA to work properly.

Since linear and numerical skew the quality of super sampling. It has to be a 1:1 ratio otherwise you are going to take a quality hit with SSAA. And the only way to achieve a 1:1 ratio at this point in time is to use fractional scaling.

Quote:Internal resolution is more fitting and proper since it has more to do with the internal resolution of the textures

Well if you're talking about pc games SSAA is internal resolution. That's literally all you are doing when you use SSAA is increasing the internal resolution. Problem is thanks to the efb we have two internal resolutions with dolphin. Before and after the efb is created. This makes the words "internal resolution" an even more stupid term to describe it. EFB scale was a better term. If he wanted to name it the smart way he should call it "efb resolution" and make the options listed as "640 x 480", "1280 x 960", "1920 x 1440", "integral (auto)", and "fractional (auto)".

Now, I'm sure this is a stupid question, but: if 2.25x SSAA makes problems, wouldn't just 2x SSAA be possible?

That can be implemented two ways. 2x1 or 1x2. Both should work fine. I'm also guessing that dolphin uses an ordered grid sampling pattern instead of rotated or jitter, I really should take a look at the code for this one of these days when I have time.

1x2 or 2x1 = 2xSSAA, usually one axis is super sampled, but afaik, its usually horizontals that suffer the most aliasing.....
2x2 is 4xSSAA
3x3 = 6xSSAA
4x4 = 8xSSAA

then you have hybrid samples, which super sample then multisample

as EFB scaling is a scaler, however you multiple the EFB x SSAA similar to what you do on a hybrid msaa mode.

so EFB 2x + SSAA 4x = 8x the size of the original texture internally, then scaled down to the monitor resolution.
EFB3x + SSAA 4x = 12x the size of the original

however in every game i've seen, using anything other than Fractional with a SSAA mode introduces more jaggies without any effect on blur. Natural Violences LOD slider suggestion would counter that better.