Radeon R300g Morphological Anti-Aliasing Performance

[phoronix]

Phoronix: Radeon R300g Morphological Anti-Aliasing Performance

Last week benchmarks were delivered of the AMD Radeon R300g with MSAA, after the much-used feature was finally implemented in a proper and working state for this open-source graphics driver. Coming out today are new Morphological Anti-Aliasing (MLAA) benchmarks from this Gallium3D driver to compare to the Multi-Sample Anti-Aliasing results.

http://www.phoronix.com/vr.php?view=18373

[Nobu]

I find it interesting that, with the exception of the last test, MLAA is consistently (albeit, only slightly) faster at 4x than at 2x--and in the last test, faster at 8x than 2x.

The numbers are very low, so it's not saying much, but I'd be interested in an explanation if there was one.

[marek]

The question is why the framerate sucks so much. It's very unlikely for 2x MSAA to be more than 2x slower, but the article shows 11x lower framerate in openarena. It doesn't add up. Has anybody else been seeing this on his/her machine?

[agd5f]

As noted in the last MSAA review:
http://phoronix.com/forums/showthrea...746#post305746
even the non-MSAA scores are really low compared to the previous phoronix r300 review.

[dungeon]

For me, there is zero point to even test these antialiasing technics when one don't have good performance without, something very beyond 60 fps and no lower . So Michael when you get 120 fps and up in some games then please do it, make antialiasing benchmarks only with something but on top of that "wasted" render performance, only with apps where you have that wasted performance otherwise please don't - there is no point .

Whole point in antialiasing is to make render more beautiful on eyes .

I think, when things are proper implemented and i have 120 fps already, i will have 60 fps with MSAA 2X? That would be coolish

[bridgman]

I think, when things are proper implemented and i have 120 fps already, i will have 60 fps with MSAA 2X? That would be coolish

The same card was running most of these tests at 100+ fps almost two years ago (see link below) and should be even faster now. That's why we're all surprised that this round of tests is showing much lower numbers even with MSAA turned off. It's possible that the higher levels of GL support are causing the app to take code paths which give higher quality but lower performance, however other users are also seeing higher numbers than we're seeing here.

http://www.phoronix.com/scan.php?pag..._compare&num=1

EDIT -- of course it's always possible that the tests from two years ago were showing artificially *high* numbers

[dungeon]

He, he, it is time for Michael to do some bisecting and not benchamarking . Then do benchmarks with or without beer .

[bridgman]

If any other folks are running an X1800 or X1900 card it would also be interesting to see what kinds of numbers they are seeing.

I had an X1950 at home but replaced it with an HD 5670 when the guys got Evergreen support working... now I can't seem to find the X1950 ;(

[marek]

My OpenArena 0.8.8 benchmark results!

I tested 2 resolutions. 1440x900 uses unoptimized MSAA because of a limitation in my hardware (see here for detailed information - the patch was committed today). 1280x720 is fully optimized, so the hardware runs at full speed.

  Resolution: 1440x900

No AA:     62 fps

2x MSAA:   49 fps
4x MSAA:   35 fps
6x MSAA:   27 fps

2x MLAA:   38 fps
4x MLAA:   38 fps
6x MLAA:   37 fps

It's in line with what I would expect. MLAA seems to have a fixed cost, while MSAA varies depending on the MSAA mode.

  Resolution: 1280x720

No AA:     77 fps

2x MSAA:   65 fps
4x MSAA:   57 fps
6x MSAA:   56 fps

2x MLAA:   49 fps
4x MLAA:   49 fps
8x MLAA:   47 fps

The cost of MSAA is much lower at this resolution thanks to the MSAA optimizations being enabled by the driver, and it even outperfomed MLAA. 6x MSAA is almost as fast as 4x MSAA, which makes sense if I consider how the optimizations work. Of course, it always depends on the app.

System info: ATI Mobility Radeon X1700, Ubuntu 12.10, Linux kernel 3.8.0-rc3, Mesa git, Xfce4, no compositing.

[dungeon]

Your numbers are good, now if Michael could run OpenArena 0.8.8 on those and other resolutions.

Don't know where is the gap between your and Michael setup, you both run Ubuntu... Maybe Unity do something wrong again.