Early Radeon vs. Nouveau OpenGL Tests On Linux 4.4

[phoronix]

Phoronix: Early Radeon vs. Nouveau OpenGL Tests On Linux 4.4

With Nouveau Kepler re-clocking beginning to work better on the in-development Linux 4.4 kernel, here are some fresh benchmarks comparing the open-source NVIDIA driver on the Git kernel compared to some AMD Radeon graphics cards on its open-source driver.

http://www.phoronix.com/vr.php?view=22434

[shmerl]

Is there a plan for automatic dynamic reclocking depending on the load? That's what the closed Nvidia driver does.

[karolherbst]

Michael if you want, you could try out the two patches from this branch: https://github.com/karolherbst/nouve...kepler_voltage

it's only a guess, but it might solve the reclocking issues with the 780 ti, if not the vbios would really help here

[karolherbst]

Is there a plan for automatic dynamic reclocking depending on the load? That's what the closed Nvidia driver does.

yes, I have a prototpy working on my local system. Sadly reclocking itself seems to be not stable enough to even think of adding something like that currently. First most of the reclocking issues have to be sorted out.

the Prototpy works great though on my mobile chip (770m), so I would say that next year there will be something like that.

[kernelOfTruth]

karolherbst from which upstream repo are you pulling ?

Reclocking meanwhile seems to work quite well, more reliably, on this GTX 760 Phantom even if only on the 2nd-highest stage (0e),

between 0e and 0f there's still no difference in performance - there's still errors in dmesg - but it appears to still work (like someone mentioned in the IRC logs)

[Guest]

Phoronix: Early Radeon vs. Nouveau OpenGL Tests On Linux 4.4

With Nouveau Kepler re-clocking beginning to work better on the in-development Linux 4.4 kernel, here are some fresh benchmarks comparing the open-source NVIDIA driver on the Git kernel compared to some AMD Radeon graphics cards on its open-source driver.

http://www.phoronix.com/vr.php?view=22434

Is there an explanation why HD6870 (97 FPS) is faster than HD6950 (64 FPS) in OpenArena 0.8.8?

[kernelOfTruth]

What is it with those "unapproved" posts ? New policy ?


WTF ! Where did my post go ? Can one of the mods actually "approve" and unlock it ?

[shmerl]

yes, I have a prototpy working on my local system. Sadly reclocking itself seems to be not stable enough to even think of adding something like that currently. First most of the reclocking issues have to be sorted out.

the Prototpy works great though on my mobile chip (770m), so I would say that next year there will be something like that.

Sounds promising! It's a major feature that holds Nouveau back for now.

[You-]

Is there an explanation why HD6870 (97 FPS) is faster than HD6950 (64 FPS) in OpenArena 0.8.8?

I don't know of the reason, but if I remember correctly, the 6950 was n the middle of major changes from AMD - the earlier cards (68xx and below) were VLIW5, it was VLIW4 and the later cards (7xxx) were GCN.

I just wonder due to it having a different architecture that only lasted one generation, maybe it hasnt had the same sort of focus as the earlier and later cards.

[dungeon]

Is there an explanation why HD6870 (97 FPS) is faster than HD6950 (64 FPS) in OpenArena 0.8.8?

Yeah that one needs explination, also why HD 6570 and R7 370 results were missing from some of the tests.

Not sure why those are missed, but about HD6950 slowness phoronix openarena mostly benchmark bloom reflection that isn't scalable really somewhere is slow somewhere fast... just guessing but it might be that, usually one effect holding things down on some gen but not on another - that would be first thing for me to check out if i have both cards... particulary because other results scale fine.