Radeon Gallium3D Is Running Increasingly Well Against AMD's Catalyst Driver

[tomtomme]

I guess the driver is optimized for VLIW5 while the 6950 uses VLIW4 (see: http://en.wikipedia.org/wiki/Radeon_...#Chipset_table - http://en.wikipedia.org/wiki/TeraSca...rchitecture%29 ). I wished someone would spend some VLIW4 love to r600g.

The newest article (kernel 3.17 features) says that Cayman dpm was disabled a while ago.
Seealso the comments

[by.peroux]

The HD6870 seems to not be too far away from being just as good as the catalyst drivers, at least in some tests. I wonder if mesa can eventually surpass catalyst performance.

Myself 6870 owner I used mesa little bit. On one hand it's fairly well packaged with distro and perform quite well (I tried mostly blizzard games with wine). But on the other hand, the fan control is, in my case, so awfull that I have to use amd's blob or it will drive me nut. Until the fan speed doesn"t scale with he card heat, I would not use mesa, even if performances were better than catalyst driver.

[curaga]

Considering only two cards in existence use VLIW4, one can understand why the FOSS driver devs don't see it as a priority. Especially if it requires much work.

You forget the APUs and FirePros. 6930 + 6950 + 6970 + four Trinity + four Richland + FirePro V5900 + FirePro V7900 = 13 cards by my count.

[dungeon]

You forget the APUs and FirePros. 6930 + 6950 + 6970 + four Trinity + four Richland + FirePro V5900 + FirePro V7900 = 13 cards by my count.

Actually 15 pciids i counted, but out of nearly 700 pciids radeon driver supports, that is around 2% .

[dungeon]

But i am not sure if all those are affected by low performance, maybe APUs are not Does somebody report this also happens with Trinity/Richland APUs ?

[curaga]

http://openbenchmarking.org/result/1311301-SO-RADEONAMD72

Richland blob vs open results from Michael. 3.13 kernel + mesa 10.1 is a bit old by now, but it was only 75% of Catalyst in the aggregate sum.

[dungeon]

http://openbenchmarking.org/result/1...SO-RADEONAMD72

Richland blob vs open results from Michael. 3.13 kernel + mesa 10.1 is a bit old by now, but it was only 75% of Catalyst in the aggregate sum.

Ah, APUs are fine then... "raw" performance (OpenArena 0.8.5) was even better then fglrx



So, seems like only dedicated Caymans are affected .

[akincer]

I check the Radeon feature list on x.org periodically and the TODO boxes are getting fewer and fewer. Some have been in WIP or MOSTLY for quite a while. Are there any thoughts from AMD folks on how close some of those WIP and MOSTLY boxes not marked DONE are to getting there?

[curaga]

Ah, APUs are fine then... "raw" performance (OpenArena 0.8.5) was even better then fglrx

..snip img...

So, seems like only dedicated Caymans are affected .

Uhm, the vliw difference only really affects the shaders. Which OA 0.8.5 doesn't use. Look at Xonotic and Nexuiz.

[sireangelus]

Michael, with all my kindness, you added frame time, you changed the graph but you are still unable to present it in a way that makes sense.
Frame time measurement is taken to look for "spikes" that might make you lag during gaming. As in, the variance between the average frame time and a frame that took a very long time to render, and how much does this happen during the test.