Open-Source ATI R600/700 Mesa 3D Performance

[Melcar]

Agree on the graphs not been that hot. Having all those thin colored lines close together is hard on the eyes.

[korpenkraxar]

Am I the only one who thinks the graphing of the results is absolutely abhorrent?

When there are many parallel lines like in this graph, I would very much appreciate to have the text associated with each color box in the legend above the graph to also appear as a tag somewhere along each line. Being half color blind does not make for quick and easy identification of which line represent which card...

[Melcar]

Here is an old radeon vs. fglrx test I did. All those games a rather playable. Would like to know what you did to get Tremulous running though; it's a slide show even under masa 7.8, and tends to crash.

Never mind. Got it to work by disabling ST3C compression on driconf.

[linuxjacques]

Wait, you don't like some of the results so you won't show them? What the heck!

It would be nice to know if things have failed miserably or are not sufficiently different to even see a difference, too, you know: you even already have all the data.

Agreed!

Not showing the "bad" results boggles my mind.

You give the impression that everything at least works which is simply is not true.

[bridgman]

I thought that my HD3650 would have been nearer to the HD3850... it's almost half the performance!
It's quite a low level rather than a mid-range! (at its times I mean... nowadays it ain't even worth considering). ok, I bought it as a low budget transition card, but still...

The 3650 should be about half the performance - it processes 8 pixels per clock vs 16 for the 3850, and does 120 shader ops per clock vs 320 for the 3850. Memory bandwidth is also about half.

The 3650 vs 3850 numbers seemed about right given the hardware differences.

[Melcar]

Here is an old radeon vs. fglrx test I did. All those games a rather playable. Would like to know what you did to get Tremulous running though; it's a slide show even under masa 7.8, and tends to crash.

[Apopas]

Am I the only one who thinks the graphing of the results is absolutely abhorrent? To get any valuable information out of the mess that is, for example, the GL and XV video performance graphs, they should have been at least two times their size. It'd also be nice to have access to the numeric data in tabular form as an option. More often than not, I'd like to compare numbers instead of trying hard to figure out which shade of $colour represents which card?

There are still numbers on the top of the pictures, which show Average, Peak and Low frame rates.

[mendieta]

Mesa is known to be slowest in race games (so factor 5 IS representive there), but in quake 3 engine based games it can gain 60% or more in the best case (here and there with some not yet accelerated/optimised features disabled) average performance compared to fglrx.

Ah, that makes sense! I have WoP installed and it ran very nicely with the Open Source stack. I haven't tried to benchmark it because it is ahuge slow download with the PTS. I may try it overnight. Thanks a lot for the insight! (and I certainly hope Mesa improves in that front )

[dungeon]

Michael,

Here is my own test (I wonder if the factor 5 in 3D performance is representative):
http://global.phoronix-test-suite.co...483-2293-23385

Mesa is known to be slowest in race games (so factor 5 IS representive there), but in quake 3 engine based games it can gain 60% or more in the best case (here and there with some not yet accelerated/optimised features disabled) average performance compared to fglrx.

[Heiko]

Why no comparison with fglrx? When 2d performance of the open source drivers are benched by Phoronix (an area in which the open source driver performs quite good compared to fglrx) the results are always compared with fglrx.

Now you are measuring 3d performance, which open source won't win from fglrx and you leave fglrx out. It would be nice to see how far off the open source driver stack is from the proprietary driver.