NVIDIA Responds To Linus Torvalds

jones_supa writes "NVIDIA's PR department has issued a statement following the harsh comments by Linus Torvalds last week where he referred to the graphics company as the single worst company he's ever dealt with, called them out on not supporting Optimus, and other issues. Basically the company replied they're committed to Linux using their proprietary driver that is largely common across platforms, and this allows for same-day Linux support with full OpenGL implementation. They also say that they're active in ARM Linux for Tegra and support a wide range of hardware under Linux. Despite having not made any commitment to better support Optimus under Linux nor providing technical assistance to the Nouveau community, NVIDIA assures us that 'at the end of the day, providing a consistent GPU experience across multiple platforms for all of our customers continues to be one of our key goals.'"

Re:I'd agree with them on that..

[Penguinisto]

As sibling said - I don't think anyone particularly cares if they write closed-source software - just open the effing API and specifications, so the community can write its own drivers for it.

Also, Nvidia is still not providing any Linux support for the one chipset that seems to be the most commonly used in laptops... go figure.

Re:I'd agree with them on that..

[Yvanhoe]

No support of Optimus means that on laptops, nvidia cards are either unsupported or power hungry. NVidia made a statement saying they will never support such a feature in their linux drivers. Nouveau has repeatedly asked for the specification information of this. Note that this information is not critical at all from a strategical point of view. No answer. NVidia's message is clearly "linux users are second zone citizens and we will not help them the slightest".

Even when not thinking that everything linux should be open source, NVidia does not provide a working linux driver for its optimus cards (that is, 90% of cards sold in laptops today). With no open source solution and no closed source solution, we can simply stare as a fact that their support simply sucks.

Nvidia has said this all along..

[Severus Snape]

90% of the code used in the Linux driver is shared with the Windows driver, that was a claim made by one of their developers on their forums I read a year or so ago. Open sourcing the code is out of the question as all of that code isn't just from internal employees, as getting everyone who has written lines of code to agree to their code being available under a open source licence would be a huge task. Documentation would be great, there's the issue of IP though there. To be fair to Nvidia, they actively support Linux, I've used their cards for years and have never had much of an issue, in the old days, it was just a matter of shutting X11 and running their installer, it built the kernel module and you were good to go. Nowadays every distro I've used has the packages ready out of the box. I think Linus pain comes simply from running pre release kernels and expecting them to be supported before their even released! Nvidia normally provide patches in these situations anyway so I don't understand what Linus really wants them to do.

Re:Summary

[Matje]

what!? Did you even read their statement?

3) We are a very active participant in the ARM Linux kernel. For the latest 3.4 ARM kernel – the next-gen kernel to be used on future Linux, Android, and Chrome distributions – NVIDIA ranks second in terms of total lines changed and fourth in terms of number of changesets for all employers or organizations.

(emphasis mine)

Unless you yourself are even more active in Linux then they are, it would be more appropriate for them to say to you ... Fuck You.

Re:Cannot open drivers source

[bruce_the_loon]

I think he means that there is no real difference between a Quadro GPU and the consumer GeForce GPU, only a PCI ID and some limits in the firmware.

Re:Diplomatic response

[higuita]

Apps have a stable API, so non-FOSS software can work fine with linux...

now DRIVERS have to comply with the kernel API, that might not be stable over time and can change... hardware builders should integrate their drivers in the kernel tree or suffer the pain of outside development. Its their choice, having to work together with the community and have the pain for legal process and code cleanup (not all trash is accepted in the kernel) is harder in the beginning, but will pay off for everyone (users, developers and company) on the long run... or play dumb and keep the closed driver and keep updating it when things change.

Releasing the hardware papers will allow the community to develop their drivers without the company have to work much, so between open source drivers, papers or close source drivers, the company have a lot to choose.

Most companies choose the first or at every least, release some papers or demo driver. They are seen as heros.
Nvidia is one of the few that choose closed sources drivers and so earns the hate of many users and the kernel developers.

Again, its their choice. Also, its the user choice to buy their cards or not.
i personally prefer open drivers and stability over better performance and locked in over on my own machine. other might have other opinions.

finally Linus dont have a hidden agenda, he cares only about the kernel and closed source drivers make very hard to almost impossible to debug problems. He choose GPL as a license as it protect his work from being abused by others. Linus didnt even wanted to migrate to GPL V3, so is clearly dont have a hidden agenda.
Again, if NVIDIA dont like the kernel license, they can choose to work only with *BSD kernels.

Re:Nvidia has said this all along..

[Joehonkie]

He wants working Optimus on laptops. He was kinda clear about that.

Small number of bins of numbers of defects

[tepples]

it's common for perfect chips to be marketed lower than what they can actually do.

You do realize that the whatsit where the defect is doesn't actually work, right?

For one thing, perfect chips get marked as defective if there aren't enough defective chips to meet the demand for low-end hardware. For another, there are probably only a small number of bins of numbers of defects. If there are models with 48, 64, and 96 working whatsits, and 63 of them work, it'll be sold as a 48, and drivers won't be able to use 15 of the working whatsits.

Re:At least open the specs.

[sqldr]

I've read the specs for AMD. It's mostly just a list of registers and what numbers to dump into them to control it. It's hardly giving away how it works.

As an offtopic, there's over 500 of the bloody things. I sort of glazed over when I saw it. The people writing drivers with no support are doing a grand (but probably quite fun) job.

Re:Summary

[Andy Dodd]

There's what they say... And then there's the code.

If they're really contributing as much as they claim, then why is the mainline cpuidle support for Tegra in 3.4 so piss-poor compared to that of their own forked 2.6.36 branch? Where's the documentation on their CPU's idle/power management capabilities? Why is the Tegra code so badly branched that devices running Android 4.0 on Tegra are running 2.6.39 instead of the officially recommended 3.0.8?

Re:At least open the specs.

[DCFusor]

You must not be much of a reverse engineer. That tells you just what can be controlled in hardware - which jobs are hardware and which software, and that tradeoff, as well as the division of labour between the GPU and system CPU is the big deal.

Plus, with all this closed stuff, who knows, or can know, what software/hardware patents might be getting cheated? What stuff that's just trade secrets (but good stuff) needs to be kept secret?

Yes, it would sure be nice if NVidia could give us more support - I'm all for that one, as there's a lotta cool things you could do with their cards. Like CUDA? Last I checked they were pretty open with that one. All my linux boxen run NVidia cards as they're the best for the money, and "just work" for me all the time.

Re:At least open the specs.

[chmod a+x mojo]

1:Put module source in the DKMS tree
2: reboot into the new kernel so the module automatically gets rebuilt
3:????
4: profit

Is it really that hard nowadays?