Dolphin, the GameCube and Wii emulator - Forums

Well, I got this and the 4x integral patch applied. All I can say is...ouch. >.< My poor computer's graphics setup is taxed to its max. I'm frequently falling below fullspeed. I'll probably just have to suck it up and play without SSAA for now. The 4x scale looks quite nice as it is.

There's an idea, though: Multi-GPU support for Dolphin! ...No?

remove the 5870 maybe there are some problems...
but with the i7 @3.4ghz you should get stable speed in full hd (smg2 NOT!)

look at this chart and find the best compromise between internal resolution and ssaa....
2x and 4xsaa should be nice but hard with your 5970...
a 6970 should run 2xefb and 4xssaa without problems like you can see in the second chart there is one guy with the newer model and he is running 3xEFB without any fps drops

http://forums.dolphin-emu.org/showthread.php?tid=14201&highlight=benchmark

Dannzen, those are in windowed mode. I was playing in fullscreen, with 4x integral scaling, and 4xSSAA. That's quite a few million pixels to push. I was still maintaining a playable framerate, but not fullspeed. I shoot for fullspeed at all times.

Quote:Dannzen, those are in windowed mode. I was playing in fullscreen, with 4x integral scaling, and 4xSSAA.

It won't matter. The internal resolution will remain the same either way.

Native resolution: 640 x 528 or .338 MP
After 4x integral scale (4x scale in each direction): 2560 x 2112 or 5.4 MP
After 4xSSAA (2x scale in each direction): 5160 x 4224 or 21.6 MP

Even for a 5870 that's too many pixels to push, especially when you're doing multiple renders per frame. Not that you don't already know that, I just wanted an excuse to do the math . You should be able to handle 3x efb scale + 4xSSAA or 4x efb scale + no SSAA without any framerate issues in most, if not all, games.

(04-11-2011, 05:16 AM)NaturalViolence Wrote: [ -> ]It won't matter. The internal resolution will remain the same either way.

Native resolution: 640 x 528 or .33 MP
After 4x integral scale (4x scale in each direction): 2560 x 2112 or 5.3 MP
After 4xSSAA (2x scale in each direction): 5160 x 4224 or 21.2 MP

Even for a 5870 that's too many pixels to push, especially when you're doing multiple renders per frame. Not that you don't already know that, I just wanted an excuse to do the math .

Ah, right, good point there! I'm used to most applications scaling with window size, hehe. There's definitely a lot of pixels to push. The fact that some games are even playable when you're rendering over 21 million pixels per frame is quite a feat.

Believe it or not if you only needed to render the frame once you'd be fine. The only reason you get slowdowns is because some places in some games require 5+ renders to get one frame. This combined with the fact that dolphin has to emulate everything through shaders makes me baffled at why anyone would ever wonder why dolphin can requires so much gpu horsepower compared to the wii.

You should be able to handle 3x efb scale + 4xSSAA or 4x efb scale + no SSAA without any framerate issues in most, if not all, games.

forgive me if I'm wrong, but isn't it pointless to have ssaa? Doesn't efb scale serve the same function as ssaa? Wouldn't ssaa 4x efb scale 1x be the same as efb scale 2x no ssaa? If that's the case, I'd say being allowed to have whatever efb scale you want would be nice.

It's actually an implementation detail, SSAA uses a custom filter while the EFB scale just uses whatever is exposed by the GPU.
We're rendering everything to the internal EFB texture (which's size is specified by the EFB scale) and then draw a textured fullscreen quad with the EFB texture on it to display that on the screen. Ideally, the GPU would just downscale everything fine there, but as it happens if your display resolution is 640x528 and your EFB resolution is 1280x1048, the GPU effectively only picks each second pixel from the source EFB texture. Thus, we effectively still render at 640x528 which is why this "fix" was necessary.
All we do really is apply a custom filter here which really averages all four source samples rather than only using the first one.

^He finally admits it. I thought I would never see the day *sniff*

Again, I did NOT break it ;P