NVIDIA vs. Nouveau Linux Driver Performance With Extra Re-Clocking Patches

[phoronix]

Phoronix: NVIDIA vs. Nouveau Linux Driver Performance With Extra Re-Clocking Patches

Last week I posted benchmarks of the AMD proprietary vs. open-source Radeon R600/RadeonSI Gallium3D drivers of various graphics cards on the newest open-source code. Today I'm doing a similar treatment on the NVIDIA GeForce side with seeing how their proprietary driver compares to the latest open-source Nouveau code.

http://www.phoronix.com/vr.php?view=22596

[karolherbst]

Michael
it looks like if you didn't install the module the right way. I know that a lot had issues with that on IRC while trying out my branch. It is a bit messy, because the install target doesn't really do the right thing :/

If you want to check, in dmesg there should be something like "nouveau 0000:01:00.0: clk: base: 705 MHz, boost: 797 MHz" sorry for that though :/

[Michael]

Michael
it looks like if you didn't install the module the right way. I know that a lot had issues with that on IRC while trying out my branch. It is a bit messy, because the install target doesn't really do the right thing :/

If you want to check, in dmesg there should be something like "nouveau 0000:01:00.0: clk: base: 705 MHz, boost: 797 MHz" sorry for that though :/

I think it took fine, or could there have been some other interaction issue? At first my 4.3 build didn't seem to work as I got screen corruption and lock-up when reclocking the first card to 0f, but after I reinstalled your module, it hadn't happened again and appeared to be reclocking. Or could there have been some other issue?

[karolherbst]

I think it took fine, or could there have been some other interaction issue? At first my 4.3 build didn't seem to work as I got screen corruption and lock-up when reclocking the first card to 0f, but after I reinstalled your module, it hadn't happened again and appeared to be reclocking. Or could there have been some other issue?

well dmesg would tell us everything. But the screen corruption is still expected. Actually you could fix that by reclocking back to 07 and then to 0f, the machine shouldn't crash anymore, just the screen looking bad.

[Guest]

Phoronix: NVIDIA vs. Nouveau Linux Driver Performance With Extra Re-Clocking Patches

Last week I posted benchmarks of the AMD proprietary vs. open-source Radeon R600/RadeonSI Gallium3D drivers of various graphics cards on the newest open-source code. Today I'm doing a similar treatment on the NVIDIA GeForce side with seeing how their proprietary driver compares to the latest open-source Nouveau code.

http://www.phoronix.com/vr.php?view=22596

The article should also be showing Nouveau performance without the extra Kepler re-clocking patches.

[karolherbst]

The article should also be showing Nouveau performance without the extra Kepler re-clocking patches.

doesn't matter. When the card reclocks, you get the good perf, if not, then you are back to normal.

[karolherbst]

Michael
anyway, you could also note that with my branch the gpus will only clock to their "base clock" and don't reach the higher clocks by default (due to risk to overheat or draining too much power), all in all it was just some testing stuff and most likely won't ever land inside nouveau, because it was more of a hack, which could be tried out by others without the need to hack further in nouveau.

[Guest]

doesn't matter. When the card reclocks, you get the good perf, if not, then you are back to normal.

How will that allow me to get an insight into by how much the driver with x="extra Kepler re-clocking patches" is better than the driver without x?

[karolherbst]

@atomsymbol: there is no difference. With default it is actually worse, because it only allows nouveau to clock to the base clock, totally ignoring the boost clocks.

But there is a load parameter I added to actually allow higher clocks, but that wasn't the point of the patches. The point was just to stabilize and try things out, like are the REed parts of the vbios right? How does it behave compared to the blob? Are the clocks near to what the blob uses? Stuff like that. It's more an experiment than actual "spee improvements"

[rabcor]

Not quite as impressive as the fermi benchmarks, but well it's getting somewhere.