Nouveau Driver Still Tries To Compete With NVIDIA

[phoronix]

Phoronix: Nouveau Driver Still Tries To Compete With NVIDIA

After sharing the results last week of an optimized open-source Radeon driver trying to compete with AMD's Catalyst driver, it is time to turn the tables. In this article is a look at the latest open-source Nouveau driver code compared to NVIDIA's official closed-source Linux driver across a few generations of GPUs.

http://www.phoronix.com/vr.php?view=17480

[Veerappan]

On the cards where re-clocking worked, I am very happy with the results. It gives me hope that the GF9400M in my laptop (MBP 13") will continue to work acceptably in the future even if Nvidia stops releasing binary driver updates. Once the OpenCL support is to an acceptable state, the only thing holding me back from switching might be battery life, which I haven't done a comparison of recently. If re-clocking works correctly on that card (which it might already), then switching should be fairly painless.

[Hamish Wilson]

Not trying to be overly partisan, but where is the awesome performance compared to radeon that I have heard about?

The low-end card performance is quite impressive when compared to the blob, but in general you can still get higher performance on many AMD cards with the FOSS driver.

This is by no means meant to disparage the Nouveau developers who are doing an awesome job with almost no resources. But this paints a quite different story when compared to some of the stories told by AMD's critics on this forum.

[strcat]

Not trying to be overly partisan, but where is the awesome performance compared to radeon that I have heard about?

The low-end card performance is quite impressive when compared to the blob, but in general you can still get higher performance on many AMD cards with the FOSS driver.

This is by no means meant to disparage the Nouveau developers who are doing an awesome job with almost no resources. But this paints a quite different story when compared to some of the stories told by AMD's critics on this forum.

The performance definitely isn't awesome, but from what I've read the radeon architecture is more reliant on shader optimizations and the architecture puts more of the complexity into the driver programming. However, the radeon drivers are a lot further along with performance-related features like HiZ due to having support and documentation (even if it's quite limited).

On the high end cards, nouveau isn't reclocking the cards to the highest speeds so there's nothing to really compare unless you set them manually.

[pipe13]

Not being partisan, but did 3 tests really fail to run on the 8600GTS, or is there a typo in the charts and did this clean-room reverse-engineered-with-no-help-from-nobody FOSS driver actually complete all of Michael's benchmark tests? No slouch hacking either way, major kudos to the Nouveau team!

[curaga]

However, the radeon drivers are a lot further along with performance-related features like HiZ due to having support and documentation (even if it's quite limited).

I recall it was exactly the opposite, nouveau having had hiz etc enabled for a long time, but in radeon they are still buggy and not enabled by default.

Something to do with it being a "turn it on" on Nvidia vs "program these, these and these registers, and then the format you submit foo in changes to this, or depending, to this..." on Radeon.