Nouveau Reclocking: Buggy, But Can Boost Performance

[Kano]

Well it seems that this still needs a lot of work. But i tested the speed differences last april using openarena already. i think it worked for 9400 gt and ion at that time while a 7600 gt always ran at full speed. most results are too complicated to read just for the ion ones i did a nicer description. the effect was very extreme there

http://openbenchmarking.org/result/1...KANO-110419300

[RSpliet]

Can someone explain to me why its usually the lowend cards that have the best relative performance with mesa? This is not only nouveau, but even lowend radeons have usually much higher relative performance then their highend counterparts.

Well, it is not unlikely that the graphics card is not the big bottleneck here. Two things I can come up with:
- The code in Mesa is not nearly as optimised as the one in the official drivers. This means that you possibly need more CPU cycles to do the same. Once you hit the CPU's speed barrier your framerate will not go up, regardless of the graphics card inside.
- A graphics card is basicly a chip with many many cores. Think of hundreds for mid- and high-end cards. This means that scheduling can be quite a big influence on performance for parallel-executable code. The more of these cores you utilise in parallel, the more performance you will get. Perhaps here's something to win too.

[log0]

This is just a guess, but I think that nouveau performs that well because it is reverse engineered from the binary, while the ati driver is being written based on the docs. If the nouveau is mimicking the binary driver to a certain degree, the performance shouldn't be that far off.

Now, what would happen if one compared the command/data steams from fglrx with the r600g output?

[AlbertP]

Can someone explain to me why its usually the lowend cards that have the best relative performance with mesa? This is not only nouveau, but even lowend radeons have usually much higher relative performance then their highend counterparts.

Lower end cards have a smaller difference between the performance levels, because they don't need as much power savings as higher end cards. It also happens sometimes on lower end cards that the default power state is not the lowest one.

For Fermi it's a different story as their lowest power state is barely useable, often one of the clocks is at 50MHz in that state. But still, for lower end cards the difference in clockspeed with the blob is smaller than for higher end cards, when both run at 50MHz with nouveau.

[madbiologist]

It can't just boost performance, it can also save power for certain GPU's. My laptop boots at power level 1, but I can now set Nouveau to level 0 to save power.

Which GPU is in your laptop?

[AlbertP]

It's a 8400M GS aka NV86.

[darkbasic]

Why does radeon suck so much compared to nouveau? Considering the open specs it should be much faster...

[AlbertP]

Radeon may indeed be slower, but often it's more reliable than Nouveau. And it had some reclocking support way before Nouveau did. Radeon is also the second open driver to have OpenGL 3 support. This is comparing apples and oranges. The features and stability should also be considered.

[darkbasic]

I know I know, nouveau sucks when comes to stability. I am curious about the motivations, maybe nvidia gpus are easier to program?

[AlbertP]

That's true. It's already said before on the forum, NVIDIA GPU's are easier to program