Nouveau's OpenGL Performance Approaches The NVIDIA Driver

[phoronix]

Phoronix: Nouveau's OpenGL Performance Approaches The NVIDIA Driver

As I began to share over the weekend, the community-created Nouveau driver that's open-source and is written by clean-room reverse-engineering the NVIDIA binary display driver, has reached a serious milestone. For low-end NVIDIA GPUs, the Nouveau driver based upon the Mesa Gallium3D architecture is now as fast, or even faster, than NVIDIA's official proprietary driver.

http://www.phoronix.com/vr.php?view=15812

[runeks]

Impressive! But how can this be?! AMD releases all sorts of documentation to the community, hires developers to work on drivers - and Nvidia, remaining closed source, needs to be reverse engineered, and it outperforms the AMD driver compared to their respective binaries!

I don't understand this is possible...

[NSLW]

Impressive! But how can this be?! AMD releases all sorts of documentation to the community, hires developers to work on drivers - and Nvidia, remaining closed source, needs to be reverse engineered, and it outperforms the AMD driver compared to their respective binaries!

I don't understand this is possible...

There is an hypothesis that someone fiddled with benchmarks just to make nouveau performance looks better than it is in reality.

[MùPùF]

There is an hypothesis that someone fiddled with benchmarks just to make nouveau performance looks better than it is in reality.

Believe me, we have no interest in doing so

[NSLW]

Believe me, we have no interest in doing so

I haven't got slightest doubt. I meant that benchmark set was specially set to give nouveau better chance at achieving longer bars. See where the doubt come from:

http://phoronix.com/forums/showthrea...759#post184759

http://phoronix.com/forums/showthrea...768#post184768

http://phoronix.com/forums/showthrea...769#post184769

http://phoronix.com/forums/showthrea...792#post184792

http://phoronix.com/forums/showthrea...817#post184817

http://phoronix.com/forums/showthrea...826#post184826

[Drago]

Impressive! But how can this be?! AMD releases all sorts of documentation to the community, hires developers to work on drivers - and Nvidia, remaining closed source, needs to be reverse engineered, and it outperforms the AMD driver compared to their respective binaries!

I don't understand this is possible...

I was asking the very same question. It proved to be that, nvidia chips, are easier to program, so you get the results fast. Maybe that is the reason for very successful nvidia binary drivers.

[NSLW]

I was asking the very same question. It proved to be that, nvidia chips, are easier to program, so you get the results fast. Maybe that is the reason for very successful nvidia binary drivers.

Yes, That's very reasonable explanation. I noticed the same with my shoes When I try to wear green ones then my feet slips in very easy and lacing goes easy too so I get my shoes on my feet very fast. On the other hand the red ones are terrible I've even thinked that little boy who doesn't know how to lace shoes could beat me five times at lacing if he would have green shoes O_o

[FireBurn]

Can I just remind you again that there isn't a 2.6.39 kernel and 2.6.39-rc1 still hasn't been tagged yet

[oliw]

I haven't got slightest doubt. I meant that benchmark set was specially set to give nouveau better chance at achieving longer bars. See where the doubt come from...

The most telling part is one comment in that thread that points out that nouveau doesn't support things like AA/AF but the tests include them. So while the binary nvidia driver is rendering everything beautifully with all available features, the open source version is pushing out a haggard shell of a render. That would give it a substancial speed boost. Benchmarks need to be like-for-like.

(As highlighted in the other thread) The benchmarks here do seem to skate over things like CPU time and, more importantly given the incomplete state of so many drivers, image quality.

Hopefully future benchmarks will be a little bit more full so people can make useful conclusions based off them. As it is we have two camps: One which believes everything written on the site and the other that outright refuses to believe anything written here. Until you can unify the groups by improving your testing procedures, benchmarks like this hurt things more than they help.

[[Knuckles]]

Michael, a typo:

For these initial Nouveau tests using the latest Mesa 7.11-devel Gallium 0.4 Git code as of 2010-03-18...

I'm guessing g3d is from 2011.