It's remarkably straight-talking for a press release, actually. Money quote is this one: "While we understand that some people would prefer us to provide detailed documentation on all of our GPU internals, or be more active in Linux kernel community development discussions, we have made a decision to support Linux on our GPUs by leveraging NVIDIA common code, rather than the Linux common infrastructure.".
Basically, they recognize that they don't support DRI2/KMS, or libva, or hybrid graphics. And they don't care. They think their stuff works better for them and the customers they care about. And they don't expect this to change.
I guess I can respect that, even if I disagree.
(Then they go into PR mode in the final point, arguing that NVIDIA has contributed lots and lots of Linux patches for the Tegra SoCs, which is obviously true but irrelevant to the controversy.)
Basically: "We only care about keeping a checkbox next to 'linux support' in our feature grid. We don't care about how shitty the experience is for linux users"
There could be a number of reasons for what Nvidia is choosing to do here. I've worked for a company that provided a kernel module separately (though, it was all GPL). There were a number of reasons for this:
* The architecture of our driver did not fit with the subsystem we were playing in (this was a SCSI module). We worked with the package maintainers, and without a near complete re-write of our code, it would not have been accepted.
* Part of the above design was cause by cross-platform code we had (parts of the driver was shared between Windows, Linux, AIX, others).
* When supporting customers that run older version of a distro, there would be no-way to give them the latest features if the code was part of the kernel. So our latest product offerings would be incompatible with older systems. We would have been at the mercy of RedHat and SUSE to backport our drivers into their older releases. (or would have had to pay them a lot of money to do it).
We had a driver that basically started development in 1997 or so, that there was no-way we were going to rewrite it just to get it into the linux kernel.
For us, it was just way cheaper to supply the drivers outside of the kernel.
> For us, it was just way cheaper to supply the drivers outside of the kernel.
While that's certainly fair enough, and rational in economic terms, when a user base reads between the lines that "you're not worth spending the money to do things properly", I can also see it eliciting a more visceral and less rational response such as Linus'. I think that in any case though, money isn't the issue so much as Nvidia wanting to not share their toys.
I agree, they get points for not trying to hide the fact that Linus is right.
Unfortunately, "Yeah, so what?" is not the kind of PR you want here. This was an opportunity to go "Yes, that's how it is now, but in the future, we're planning on supporting all this wonderful stuff on Linux!"
Instead, it's an advertisement for ATI/AMD's Linux support.
Although I think it's better they say this upfront than take an opportunity to lie to the community if there is no plan to support Linux properly in the future.
If you mean the ATI proprietary drivers, I can believe that nVidia's suck slightly less. However, the majority of Radeon cards have full 2D/3D support in Open Source drivers (and those drivers actually have developers from ATI working on them), whereas the reverse-engineered nouveau driver supports a smaller subset of nVidia cards with much lower 3D support (and no support from nVidia whatsoever).
So, I'd still take ATI over nVidia due to ATI's much better Open Source drivers with some development done by ATI, though I'd take Intel over ATI for having entirely Open Source drivers with much more direct development from Intel.
I guess different people had different experiences. My nVidia cards have worked just fine on Linux but my HD7750 ATI card does not work properly (random screen flashes and poor rendering) on Ubuntu no matter which driver I try.
Which is ridiculous, considering they cherry-picked RandR of all the technologies in the "Linux common infrastructure" to support with their latest driver release. I suppose I should be grateful that my projector can work without opening up NVIDIA Settings and tweaking about, but it certainly flies in the face of this statement and belies their true motivations: they'll implement the bare-minimum Linux-only feature support that will keep their business/edu customers from fleeing to AMD, and that's about it.
RandR is a Xorg extension, it's not a Linux kernel thing. To continue to provide graphics options for their customers that use OS's with recent Xorg installs (they make a FreeBSD driver available too) they had to support RandR.
This is right in line with 'we support our customers/users' not whatever the Linux kernel guys feel like doing today.
That's spun and unfair. Booting with a working framebuffer isn't something that tickles only the "kernel guys" interest. Running the GPU on your hybrid laptop likewise. You realize that, with just a handful of exceptions, the NVIDIA part on most Optimus laptops simply doesn't work under linux, right? And that NVIDIA hasn't so much as lifted a finger to fix this?
So sure, if "customers/users" includes the only handful of 3D content shops out there running linux toolchains, and not the millions of people buying NVIDIA-enabled laptops, I guess it makes sense. But for everyone else it kinda sucks.
Hardly. NVIDIA says they "made a decision to support Linux on our GPUs by leveraging NVIDIA common code, rather than the Linux common infrastructure." HN responder says that's bullshit since they finally support RandR. I point out RandR is not 'Linux common infrastructure' but a Xorg technology and supporting that is very literally "supporting Linux ... by leveraging NVIDIA common code" because the NVIDIA common code is targeting Xorg technologies and not Linux specific ones.
Then you move the point to something completely different.
NVIDIA is clearly not interested in providing any Linux only support. Don't expect them to support anything under Linux that isn't also common to other platforms they support.
You're just making things up, stop it. DRI2 and KMS are no more "linux kernel" things than "X" things; all you have to do is check the domain hosting the specs. These are well established, well-supported APIs that have been very well received in the Linux world. And yet NVIDIA wants nothing to do with either, for silly internal reasons with poor justification.
> You realize that, with just a handful of exceptions, the NVIDIA part on most Optimus laptops simply doesn't work under linux, right? And that NVIDIA hasn't so much as lifted a finger to fix this?
I have an Optimus laptop (Thinkpad W520) and that's not really true. The problem is that you can't run both the Intel and NVidia chips at the same time without it being a PITA. If I want to use my internal display while I'm on the road, I move my NVidia Xorg.conf out of the way and it's fine. I'm not saying it's optimal, but it definitely works.
There's not just one Optimus implementation, there are many ways to wire Optimus. My colleague bought an Optimus laptop (can't remember which one now) that absolutely doesn't have the nVidia part of Optimus working. See here for details: http://nouveau.freedesktop.org/wiki/Optimus
> Booting with a working framebuffer isn't something that tickles only the "kernel guys" interest
Is it? If my computer boots up directly to X anyway, why do I care about a framebuffer beyond the standard vesa stuff?
PS: If Linus would get off his high horse about closed source drivers, and commit to have at least a somewhat stable ABI, that would do more for hardware support on linux than 1000 years of ranting and moaning, which if anything, will only alienate the hardware makers even more.
"If Linus would get off his high horse about closed source drivers"
... then he would be Microsoft and Apple, and not an Open Source operating system developer, maintainer, producer, and evangelist. All his years of working openly and freely will be squandered and his life's work beholden to closed-source, closed spec device manufacturers.
But his stance actually makes us much MORE beholden to the 3rd parties. If every minor kernel version didn't require the vendors to recompiled drivers, and quite possible change code because Linus changed the ABI again, that would result in more hardware working better for more people.
Live the open source dream if it makes you feel good. I live in a world where if it doesn't work, I couldn't care less about it.
I don't think reality bears this out. Now that Broadcom has come to their senses, NVIDIA is pretty much the only major consumer device vendor in the PC space with binary blob drivers. Everyone else has, ultimately, folded in the face of pressure from the dogmatists. I think Linus is pretty much winning, and the fact that NVIDIA is feeling pressure here is a good thing.
Do you have any examples of situations where this is making us "MORE" beholden to third parties? I honestly don't see many (though in the SoC space, the PVR drivers are a very similar situation).
Didn't AMD release a ton of documentation for their GPUs specifically so that the open source radeon drivers could become a realistic option when using their GPUs?
Nobody I know with an AMD GPU uses the propriety fglrx drivers.
Open source 3D drivers will have a hard time EVER being useable for anything beyond glgears... everything in that space is going to be really patent encumbered.
Having an unstable, internal kernel interface has nothing to do with being opposed to closed source drivers. Linus' problem is when people bitch and moan when things change and it breaks closed source drivers. He's repeatedly said that anything internal is fair game to change. If your driver is in the kernel it will be fixed if the interface changes; if it's closed then you better keep up on your own.
Quite clearly he's talking about Nvidia as a company, not about the properties of one particular subsystem. The statement stands well enough on its own.
Basically, they recognize that they don't support DRI2/KMS, or libva, or hybrid graphics. And they don't care. They think their stuff works better for them and the customers they care about. And they don't expect this to change.
I guess I can respect that, even if I disagree.
(Then they go into PR mode in the final point, arguing that NVIDIA has contributed lots and lots of Linux patches for the Tegra SoCs, which is obviously true but irrelevant to the controversy.)