Testing Out The Nouveau Driver On Fedora 11 (Phoronix)

[Posted March 30, 2009 by jake]

Phoronix tries out the Nouveau driver in Fedora 11. Overall, they seemed fairly impressed with its capabilities. "The xf86-video-nv driver is officially maintained by NVIDIA, but it's their half-assed attempt at being open-source friendly. The X.Org driver's code is obfuscated, its 2D support is limited, there is no 3D acceleration at all, and it barely receives new features and support these days. Meanwhile, a group of open-source developers have been reverse-engineering NVIDIA's binary Linux driver to write the Nouveau driver that will offer 2D, 3D, and video acceleration and aims to be feature-complete. The Nouveau project has been around for a few years, but their code is starting to come to maturation with kernel mode-setting and a Gallium3D driver hopefully being stable by year's end."

to post comments

Testing Out The Nouveau Driver On Fedora 11 (Phoronix)

Posted Mar 31, 2009 0:06 UTC (Tue) by PaulWay (guest, #45600) [Link] (4 responses)

It should be pointed out here that by 'reverse-engineering NVIDIA's binary Linux driver', what is really meant is 'watching NVIDIA's binary Linux driver talk to the card and reverse engineering the protocol from that communication'. Nouveau is, as far as I'm aware, not actually disassembling or decompiling the binary driver - they're essentially doing with the PCI and memory busses what the SaMBa team does with the network.

Just so people don't jump in and start hollering about lawsuits and whatever.

Testing Out The Nouveau Driver On Fedora 11 (Phoronix)

Posted Mar 31, 2009 14:20 UTC (Tue) by leomilano (guest, #32220) [Link] (3 responses)

Good point

So, assuming you are building a system from scratch. What graphics card can you buy that is supported by a free software driver providing 3D acceleration? Is there such a thing? Some of the newest intel integrated graphics (like the one on the eeepc) seem decent, but really won't cut it for moderate 3D gaming, no?

Thanks in advance.

Testing Out The Nouveau Driver On Fedora 11 (Phoronix)

Posted Mar 31, 2009 17:11 UTC (Tue) by Thalience (subscriber, #4217) [Link] (2 responses)

Right now, you'd pretty much have to get an R500-class ATI card.

2D acceleration was recently added to the ati driver for R600 and R700 cards, but no 3D at this time.

Testing Out The Nouveau Driver On Fedora 11 (Phoronix)

Posted Mar 31, 2009 21:00 UTC (Tue) by leomilano (guest, #32220) [Link]

Thanks, and I just found this webpage, pretty useful!

http://www.free3d.org/

Testing Out The Nouveau Driver On Fedora 11 (Phoronix)

Posted Mar 31, 2009 23:07 UTC (Tue) by alecs1 (guest, #46699) [Link]

Hi, I want some honest numbers/answers on the 3D performance, because I don't have access to a R500 card.

How good is R500 3D acceleration radeonhd? For comparison, on Windows with a R200 I could run:
Half Life 2 and X2 The Threat, native display resolution, other settings around maximum.
Please, no benchmark result with those 3 spinning weels :-), I don't think one is going to compare 3 spinning wheels with going through clouds in X2.

On Linux, with Wine and fglrx I was able to start Half Life, but not play too long before crash, Tremulous and others ran fine. With the open source driver I never got further than jerky TORCS.

I'm afraid 3D acceleration on par with that:
1.is not going to be done with reverse engineering as it is too hard.
2.is hard enough that radeonhd is not showing anything spectacular until now.

I'm afraid Nouveau won't prove NVidia that it hides its secrets in vain. But again, if Ati doesn't have the resources to create decent Linux drivers, it doesn't mean it doesn't have the resources to reverse engineer competing technologies, as some previous commenter said.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 3:27 UTC (Tue) by Alterego (guest, #55989) [Link] (19 responses)

Why help NVIDIA by providing good open source drivers ?
If nothing was done Nvidia would lose market shares (the only thing they understand).

Instead helping to develop/improve driver for more open-source friendly manufacturers (ATI, Intel ...) would force nv to change its policy.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 3:46 UTC (Tue) by jwb (guest, #15467) [Link] (12 responses)

nvidia is doing themselves no favors by releasing a series of expensive yet awful products.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 4:21 UTC (Tue) by zlynx (guest, #2285) [Link] (11 responses)

If you want top performance, you *must* have an Nvidia card.

If you want a binary driver for Linux that does almost everything the Windows driver does, and is (in my opinion) better supported and more stable than ATI's Catalyst, you must have Nvidia.

That said, the latest computer I built has an ATI/AMD 4850. But every machine of mine before that (about 5 systems), between my Matrox G200 to my AMD 4850 has had an Nvidia card. Because they are the best.

I no longer choose to pay premium prices for the best video cards but that does not mean Nvidia's product is somehow "awful."

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 10:50 UTC (Tue) by sp.at (guest, #36249) [Link] (8 responses)

I am actively using an AMD HD4350 in my workstation under GNU/Linux too, and to be honest, it didn't even came to my mind to install ATI/AMD's catalyst driver.

When I bought and set up this system about two months ago I went straight ahead and installed the r6xx-r7xx-support branch of the radeonhd driver from git. As the development effort moved to the master branch I have switched over and right now am using the bleeding-edge version too, updating it once or twice a week.

Now you might ask how many problems or crashes I experienced and the truth is: not one.
To be honest I have not been doing anything 3D-heavy on this system, but 2D performance and stability is great, and the most important thing, that driver is not proprietary and if problems show up I could investigate them myself and try to come up with a solution.

So, with an AMD card I get pretty good performance from a free driver that seems to behave well, even when using the bleeding edge code from the freedesktop.org git repository, what would I want else?

Technically the Nvidia cards might be better compared to other vendor's cards, but what does that help if I have to rely on a binary driver?

I went for "worse is better" there and now it seems to me that it isn't worse after all.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 14:49 UTC (Tue) by Cyberax (✭ supporter ✭, #52523) [Link] (7 responses)

For example, you might want to use CUDA or OpenCL. Or may be OpenGL 3.

NVidia's driver is proprietary, but it's actually quite good.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 17:11 UTC (Tue) by drag (guest, #31333) [Link] (3 responses)

> NVidia's driver is proprietary, but it's actually quite good.

It's quite _fast_. What is "good" is much more subjective. It's crash happy and makes end user's lives more difficult when it comes to keeping their machines up to date.

Nvidia is the only game in town when it comes to having high performance GPU drivers, right now. If you need your machine for other purposes it may be best to stay away from it.

-----------------------

The nice thing about reverse engineered drivers is that it puts a nice shot through the heart about the theory that keeping your drivers closed source is going to prevent people from learning how your hardware operates.

That is if a few open source folks operating on the internet with a shoestring budget can learn enough about the hardware for working 3D drivers (which is going to be exponentially more complex then pretty much any other hardware device on your computer) then trying to hide your hardware interfaces behind a wall of machine code is a huge waste of time. Imagine what a couple guys getting paid full time from ATI can learn about that hardware using high-end equipment and such.

Dispelling myths and illusions about 'IP' like that is more then worthy enough for a reverse engineered Nvidia driver.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 23:13 UTC (Tue) by PaulWay (guest, #45600) [Link] (2 responses)

I think statements like "crash happy" and "expensive" and "fast" are pretty subjective without supporting evidence. The evidence I've seen is that NVidia's cards are slightly cheaper on the bang-for-the-buck scale, and the NVidia proprietary driver has been a lot more stable than the ATI cards until relatively recently.

I see it as basically a problem of inertia. NVidia's managers probably see themselves as doing their bit for open source, but they grew up in the 3Dfx/Matrox/Trident world where everything was fiercely competitive - to open source their drivers is probably seen as giving their competitors an advantage. So they do their part by at least giving us something, but still want to keep their secrets.

This 'competitive advantage' is really a fiction, as the architectures of the chips even in the NVidia range are different enough to require separate handling per chip - the difference between the processing architectures of ATI and NVidia and Intel graphics chips must be a gigantic chasm compared with the gap between individual chips. It's also pretty much irrelevant given all the standard arguments for open sourcing - lower development costs, faster bug spotting and fixing, better standards compliance and better community involvement to name the prominent ones.

I think we'll see a change at NVidia eventually. What it'll take is for someone internally to drive the process - compare Dirk Hohndel at Intel and Harald Welte now at Via.

Those efforts should be put on open source friendly cards.

Posted Apr 1, 2009 19:41 UTC (Wed) by rahvin (guest, #16953) [Link] (1 responses)

http://arstechnica.com/hardware/news/2008/03/vista-capable-lawsuit-paints-picture-of-buggy-nvidia-drivers.ars

Microsoft's data strongly indicates that the problems were real. Damon Poeter at CRN dug through the documentation to find that on page 47 of the PDF, NVIDIA drivers were identified as the cause of over 479,000 crashes, or just under 29 percent of all the crashes Microsoft logged. Microsoft's own drivers follow, at 17.9 percent, and the "Unknown" category takes third place at 17 percent. ATI is in fourth place (9.3 percent) and Intel in fifth place (8.83 percent).

If they crash windows that much and the Linux drivers are the same driver shoehorned into Linux what would you think of their stability without other data?

Those efforts should be put on open source friendly cards.

Posted Apr 3, 2009 4:20 UTC (Fri) by Kit (guest, #55925) [Link]

IIRC, those statistics were from shortly after Vista's release, when NVIDIA's Vista drivers were still incredibly immature. I don't realy think that source is very useful for determining the quality of a mature driver on another platform.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 17:45 UTC (Tue) by Alterego (guest, #55989) [Link] (1 responses)

Does the nouveau driver provide openCL, cuda, or openGL 3 ?

ATI provide "Stream Computing" SDK and should soon be OpenCL 1.0 and openGL 3.

I still think it would have been a much better effort to develop ATI opensource drivers rather than nvidia.

Those efforts should be put on open source friendly cards.

Posted Apr 1, 2009 9:03 UTC (Wed) by luya (subscriber, #50741) [Link]

Lets not forget nouveau development is done on Gallium3D framework. It will be a matter of time to support those features.

Those efforts should be put on open source friendly cards.

Posted Apr 2, 2009 3:00 UTC (Thu) by Arker (guest, #14205) [Link]

NVidia's driver is proprietary, but it's actually quite good.

Good?

I dont think that word means what you think it does.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 17:28 UTC (Tue) by Alterego (guest, #55989) [Link] (1 responses)

> If you want top performance, you *must* have an Nvidia card.

The benchmarks i have read show that NVidia and ATI are equal. Do you have url, comparing both and prices too?

Those efforts should be put on open source friendly cards.

Posted Apr 1, 2009 2:09 UTC (Wed) by zlynx (guest, #2285) [Link]

http://www.anandtech.com/guides/showdoc.aspx?i=3538
http://www.anandtech.com/video/showdoc.aspx?i=3498&p=9
http://www.maximumpc.com/article/reviews/nvidia_geforce_g...

Or look at the reviews on http://www.hardocp.com/ Go there and click Main menu: GPU / Video
Cards and check out their results.

The Nvidia stuff isn't always faster, it can depend on what game it is running.

For price, you can spend up to $2400 (four cards, $600 each). Really that would be ridiculous.
But you could.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 7:42 UTC (Tue) by mjthayer (guest, #39183) [Link] (5 responses)

Do enough people think that way for NVidia to care?

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 17:15 UTC (Tue) by drag (guest, #31333) [Link] (4 responses)

Probably.

It is said that Nvidia cares about hiding how it's hardware operates behind a shield of compiled binaries. If open source folks reverse engineer how the hardware works and begins posting their 'important secret ip' all over the internet then it is probably going to get their attention quite quickly.

Especially if they discover 'secrets' that need to be kept secret for their Windows-drivers-related DRM requirements.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 21:21 UTC (Tue) by job (guest, #670) [Link] (3 responses)

Nvidia has a lot of licensed third party stuff in there which makes the situation slightly more complicated than what you describe.

Those efforts should be put on open source friendly cards.

Posted Mar 31, 2009 22:22 UTC (Tue) by drag (guest, #31333) [Link] (2 responses)

I don't see what you said changes anything I said. I know that they do do cross licensing with other companies. Including Microsoft, and I know that they have contractual obligations to maintain certain things secret for the sake of DRM.

So, hence, they would actually give a shit if people started plastering their 'precious IP secrets' all over the internet.

----------

Anyways; both ATI and Intel have managed to disclose open drivers for their hardware. And ATI, after I am sure was great effort and expense, managed to release full documentation for their 3D stuff for almost all there hardware, including the stuff they are shipping.

Hell just this week they released the register information for their R700 hardware to go along with the 2D/3D documentation they released a few months ago. Including the bits for their "Stream Computing".

As GPU moves from '3d accelerator' to a core part of the modern computerNvidia is going to be compelled to open up also. After all, do you really want to install a proprietary binary driver just for the ability to execute code on your computer, to be able to use the CPU at all (seeing how GPU will be a coprocessor)? It's a joke.

Anybody interested in 'GPU HPC' should be very very interested and aware of ATI/AMD's recent documentation. ;) Even if it is not immediately useful.

Lets hope that LLVM/OpenCL/Gallium/GCC or whatever the hell else gets support for that stuff pretty quick.

http://developer.amd.com/gpu_assets/R700-Family_Instructi...

Those efforts should be put on open source friendly cards.

Posted Apr 1, 2009 20:14 UTC (Wed) by ikm (subscriber, #493) [Link]

If I understand correctly, neither ATI nor Intel opened up any gfx drivers that were previously closed. ATI provided docs instead, Intel have (I think) created their Linux drivers as open-source from the start.

I doubt Nvidia would be any different in that regard. The only thing they might realistically do in the future is to provide some docs.

Those efforts should be put on open source friendly cards.

Posted Apr 3, 2009 6:57 UTC (Fri) by job (guest, #670) [Link]

It is a complex work and there are many different types of secrets involved, labeling everything "precious IP" is not the whole picture. The information most usable to aspiring graphics driver writers such as register settings, initializations etc. is probably not the ones licensed from third parties.

On the other hand, other information is. That contributes to Nvidias reluctance towards open source. It is not as simple as they wanting to hide how their hardware works at any price.

Testing Out The Nouveau Driver On Fedora 11 (Phoronix)

Posted Mar 31, 2009 6:51 UTC (Tue) by PO8 (guest, #41661) [Link] (1 responses)

Phoronix's description of Nvidia's official, obfuscated open source driver as "the X.Org driver" is a bit misleading. X.Org does distribute and support this Nvidia-supplied driver. However, the Nouveau team, who distributes their driver through freedesktop.org, also participates in the X.Org project, and the X.Org project and X.Org Foundation have supported their work in a number of ways. When the codebase is mature enough, the Nouveau driver will almost certainly become part of official X.Org releases.

obfuscated open source oxymoron

Posted Apr 8, 2009 4:12 UTC (Wed) by xoddam (subscriber, #2322) [Link]

"obfuscated open source"?

Oops.