AMD Radeon HD 6000 Gallium3D Attempts To Compete With Catalyst

[RealNC]

would be interesting to see what perf would be like at high power levels, it might make the numbers a lot closer.

The article suggests that the benches were run at the high power levels already:

"Particularly on the higher-end models, the lackluster power management causes increased power consumption for the system and the fan to operate at above-normal speeds. The Radeon HD 6870, for example, was screaming. The Sapphire Radeon HD 6870 is extremely quiet with its Vapor-X cooler, but constantly when using the open-source driver the fan noise was unbearable."

[pingufunkybeat]

60% is really remarkable.

And while it's true that more modern stuff (like Unigine) will tax the drivers more, these naked frame-pushing benchmarks are still important because they show the CPU bottlenecks which are one of the main causes of lower performance.

Obviously, things are improving here. About half a year ago, r600g was slower than 30% of Catalyst, nowadays it's around 75% on some cards/workloads.

With MLAA and VDPAU around the corner, and OpenGL 3.0 almost here, it's a good time for open source.

[darkbasic]

Uhm... HD6870 is slower than HD5870, but it's much faster than mine HD5870 in nexuiz, that's very strange...

[Panix]

The article suggests that the benches were run at the high power levels already:

"Particularly on the higher-end models, the lackluster power management causes increased power consumption for the system and the fan to operate at above-normal speeds. The Radeon HD 6870, for example, was screaming. The Sapphire Radeon HD 6870 is extremely quiet with its Vapor-X cooler, but constantly when using the open-source driver the fan noise was unbearable."

Imho, it's useless if the power/fans aren't working properly.

[jrch2k8]

Imho, it's useless if the power/fans aren't working properly.

plz remember that the graphic stack team is very small compared to catalyst so they can't fix all magically in 6 months

beside OSS driver is a matter of taste if you want UVD and full PM support go to catalyst and wait until OSS drivers get there but if you want a beautiful stable and fast desktop tear free and electricity is cheap in your country(like mine) + you want to remove blob from your life + contribute something to improve the stack OSS is for you

in my case my 4850x2 sound like a jet plane turbine even on windows so i don't care about it too much and mid power profile is good enough to keep these beast in 60 C wich is very acceptable for me (54C catalyst) beside my phenom II X4 965 BE handle 1080p h264 video using Xv like a champ without killing too much cpu (ffmpeg.mt + vlc)
so i give a damn about UVD and believe or not wine games run very well for me using OSS drivers (stuff like civ5 or lineage2 hi5 or wow ) im not saying it runs at 999fps but as long as i don't get graphic lag is damn good for me (i want to play the darn game not make a race of who get the greater fps in a 85hz monitor lol) and kde4 runs like an work of art in r600g im just addicted to it (no lag, no tear, 100% stable, ridicously fast and responsive, kwin full effects maxed out, etc).

so for my taste and needs r600g is already a piece of heaven for me maybe for you is not there yet but that doesn't mean is useless, so plz show some respect for the many ppl working very hard even when they are so few in the linux graphic stack

[89c51]

And while it's true that more modern stuff (like Unigine) will tax the drivers more, these naked frame-pushing benchmarks are still important because they show the CPU bottlenecks which are one of the main causes of lower performance.

what stops the devs from fixing the CPU bottlenecks ????

[darkbasic]

The fact that they are only few people and open drivers still lack major functionalities?

[pingufunkybeat]

what stops the devs from fixing the CPU bottlenecks ????

Lack of time and manpower, mostly.

It seems like there is no single main bottleneck which needs to be optimised away cleverly, and then you're done. It's more like a set of small slowdowns which all add up, and need to be found individually and optimised. 1% here, 0.5% there, and so on.

People like Jerome Glisse have done quite a bit of work at locating such bottlenecks, but it's really hard work, and solutions are not always easy (they might require a redesign, or a different approach).

So it's something that will continue to improve. Luckily, r300g is already so highly optimised, and many of the same tricks can hopefully be applied to r600g with some effort, when somebody finds the time.

[crazycheese]

WOW!
im impressed.
didnt think the drivers would be around 50% in pretty much every benchmark (even faster in urban terror).

UrbanTerror has not removable, permanent 125fps cap.
This is because of quake3/ioquake, that binds fps to physics and network sync.
Overall, urbanterror has terrible bullet physics by the way and it seems to be very cpu bound.
The benchmark must be flawed, unless Michael managed to remove it.

[darkbasic]

If the game allows you to make benchmarks maybe they remove the fps cap in "benchmark mode"