Linus Torvalds Calls NVIDIA The Worst Company Ever

[deanjo]

At the end of the day, if you buy Nvidia, you buy a closed platform. Like PS3 or iPhone. People who want general-purpose, programmable hardware but Intel or AMD.

Sadly, most people don't mind closed platforms in this day and age.

Or like many many many other devices. How many android phones for example are truly open from top to bottom? Most are using closed blobs as well.

[Calinou]

NVIDIA works and is cheaper (less power consumption and more quiet). AMD doesn't work. Even nouveau would beat Catalyst on some GPUs, I bet.

Th?orie du complot, quand tu nous tiens.

[crazycheese]

NVIDIA works and is cheaper (less power consumption and more quiet). AMD doesn't work. Even nouveau would beat Catalyst on some GPUs, I bet.

Radeon and nouveau work. Usable. 40-70%, but its good progress.
Nvidia has a good driver, but by far not perfect!

[Yaro]

I know this thread is somewhat old an dhasn't seen a post in a while, but I felt like this was the only one I could reasonably make this statement on.

I wish to stay objective here. I know it seems pretty fruitless on Phoronix, but it's there.

We have a number of objectives in getting to that "happy place" platforms like Windows and OS X frequently enjoy regarding video support, which is not worrying about whether X or Y video card will be feature complete on the platform. Both in hardware and in device drivers.

Objective 1 - Full performance, full featured support for Linux in its device drivers. Here I find that this objective has been almost met by the nvidia binary blob. The only thing it lacks I'd like to see is KMS and native fbcon support so we wouldn't have to fall back on uvesafb. Intel also has full-featured device drivers, but the only reason why it's not on top here is because the hardware itself isn't really designed to drive much. If Intel were to invest in a GPU they could put on-board as well as on dedicated cards for gamers, I think Linux would have their killer graphics hardware. Catalyst doesn't meet this goal, and radeon and Nouveau, while definitely sufficient for most desktop usage outside of gaming, are not going to bring about the Linux gaming market as they are now.

Objective 2 - Open source drivers. Intel meets his objective nicely with objective 1, barring the lacking hardware itself. Nouveau and Radeon are nice! I use Nouveau frequently, but I find myself frequently using the binary blob because it better meets Objective 1, which is, debates over "freedom" or "security" aside, all "normal" end users will care about: The driver being able to provide them all they want without hassle. There are technical reasons why open drivers are a good thing (Such as KMS/native fbcon, not to mention it's easier to assure that the driver is stable.).

Objective 3 - The "Just Works" concept. As I said, normal end users really won't much worry about Objective 2. Maybe they should, maybe they shouldn't, doesn't matter. What matters is what they *expect* not what we *think* they should expect. Good news and bad news for the open drivers on this. Open drivers are usually the first thing every distribution defaults to. So for them next to zero configuration is needed. This will keep MOST users on the open driver without complaint. Where this objective begins to miss the mark is when they install Steam, fire up, oh, Half-life 2 and find that the game renders too darkly because Mesa 9.x isn't installed, but Mesa 8.x. To us power users, the solution seems simple and obvious. To "idiot" users, all it means is "Linux doesn't run the game right" and except in certain cases they won't be made aware that, say, the binary blob provided by nVidia runs Half-life 2 VERY well. And we'd be --desktop_user for certain in those cases.

These objectives are *not* mutually exclusive. As I said, Intel is VERY close on hitting all three of these if they'd simply provide the right kind of GPU. nVidia meets objective 1 but falls short on 2 and 3. Catalyst meets none of these objecives. Nouveau and Radeon partially meet 1 and definitely hit 2 and 3.

Now, to the topic at hand. Is nVidia in the wrong for not cooperating with the kernel devs? Yes. I think so. Because despite possible third-party IP there are *some* parts of their interface they can definitely at least shed a little light on, or they never would have provided the (Now-obsolete.) nv driver. Let's also not forget their objective in Linux support is not actually providing a usable desktop to end users but a powerful graphics infrastructure for graphical workstations and CUDA-based high-performance computing. One reason why I (And I bet a lot of others.) got excited about Valve porting to Linux is because they might *just* give nVidia the motivation to have a desktop objective as well. While it may not help with "openness" it CERTAINLY will help their driver improve for desktop users, including MAYBE features we'd sorely like their blob to have, short of being open source.

Was Linus Torvalds right in his approach to nVidia? No. I honestly think the "fuck you, nVidia" attitude helps nothing. In fact, while I do believe maybe calling them out on not being all that helpful was fine, the attitude probably would have sooner served to both drive nVidia into less cooperation (Would YOU cooperate with anyone telling you to "fuck off" and gives you the bird? I wouldn't.) but also shows to end users of other platforms that we're both overly political and openly HOSTILE when we don't get our way ("Why should I use Linux when they're clearly so closed against anyone doing things how they want?" Whether this is true is irrelevant, it's a perception problem, plain and simple.). One admitted problem is that many people look at Linux, and often see an overzealous "free or nothing" attitude that makes them question whether they'll get what they want from Linux.

I want open source drivers. I *want* Nouveau to succeed, but I need to keep an objective eye if I want to make the "most" out of my desktop using Linux, and unfortunately it often means binary blobs, unpleasant though that may be.

[hoohoo]

I know this thread is somewhat old an dhasn't seen a post in a while, but I felt like this was the only one I could reasonably make this statement on.

<snip>

I don't agree with the open source driver fetish here. AMD and NV drivers do what I want, and no, I don't care about the edge cases the fanatics will bring up.

AMD & NV are not going to disclose everything needed to open source the stuff that is currently available as blobs because that would expose too much IP.

Get used to it.

[uid313]

AMD & NV are not going to disclose everything needed to open source the stuff that is currently available as blobs because that would expose too much IP.

Get used to it.

You have no idea what you're talking about.

AMD have an open source driver.
Intel have an open source driver.

Sure they can have open source drivers without expose too much IP.

Also, the graphics card can have a firmware that abstracts away the IP, such as AtomBIOS used on AMD cards.
Or they can selectively pick what parts to include in drivers, example some vendors do 2D acceleration, but may not chose to include video acceleration.

[Yaro]

I don't agree with the open source driver fetish here. AMD and NV drivers do what I want, and no, I don't care about the edge cases the fanatics will bring up.

I don't really have a fetish about the open drivers, just a preference for it because in most cases it shows quality (And security assurance.) we can count on. The nvidia binary blob is what I usually use because while Nouveau is nice, it falls short in a few suble but eventually deal-breaking ways, such as missing the mark on gaming and, in recent kernels it happily freezes up my system. Mos anyone I talk to reveals that the Nouveau people are not always the best at being motivated to hunt down bugs, as apparently this has been an issue since kernel 3.6.

I also pointed out that drivers being open source usually makes it easier for programmers OUTSIDE of the development of the driver to be able to identify problems. The problem with blobs on the kernel level is that the only ones who can "guarantee" that the blob won't eat your kernelspace are the people who develop it. You can't submit bug fixes, nor can anyone downstream fix it to work properly if it's broken. In this case it's not just a fetish nor is it about "edge case."

I actually go by what Linus says is "use what works. It just so happens most the time it's open source." (Paraphrasing.) Unfortunately for now at least on nVidia cards the best option seems to be the binary blob if you want a high quality driver.

As for "edge cases" I'm not going to agree that KMS *or* a framebuffer console are edge cases whatsoever. To even use the native Linux graphics stack now the driver has to use KMS in some way to be fully compatible (Catalyst and the nVidia blob, since they are proprietary, cannot use kernel DRM and provide their own.) *Everyone* who has an Intel chipset uses KMS since Intel only ever provided open source drivers.

Worse still for binary blobs: When Wayland replaces Xorg, at least on the desktop, KMS will be the absolutely mandatory to get a desktop at all on Linux. Catalyst and the nVidia blob won't be able to get us there, it has to be open source drivers unless they finally allow non-GPL symbols to use the kernel DRM stack.

And you'd be surprised how many "desktop" Linux users don't use Xorg at all but instead use things like tmux in conjunction with a high resolution framebuffer. KMS is the only reliable way to make sure the framebuffer sets itself to the native mode of your monitor on the framebuffer (uvesafb and vesafb both can only rely on your card's video BIOS, which for most cards doesn't support a very inspiring number of resolutions, rarely any in the neighborhood of 1920x1080.)

AMD & NV are not going to disclose everything needed to open source the stuff that is currently available as blobs because that would expose too much IP.

I seem to recall pointing that out in my post. AMD is doing their best, but have not exposed enough of their spec to make designing the open radeon driver a simple matter. It's better than nothing and it's probably the reason why that driver went from supporting next to nothing to having almost universally better support than Catalyst under Linux.

nVidia's released nothing, probably for more IP problems than AMD. It's actually remarkable Nouveau is as high quality (But not as high as the blob, in my experience.) as it is. From what I've seen they lack manpower, organization, and basically any sort of cooperation from nVidia or they'd be pretty successful.

I stand by my assessment of the Intel situation. If they simply made more powerful hardware we'd have fully functional, high performing, completely open video that works and delivers all we'd actually want in Linux graphics:

1. Because it's part of the kernel tree it'll be loaded by udev and thus "just work" without any setup. Catalyst and the nVidia blob can't do that.

2. KMS means no more ugly switch between the console and Xorg, but a smooth instant change. No more monitors going dark for a couple seconds. This is perhaps a small benefit, but it's nice.

3. Probably the best driver to run Wayland on between feature completeness and KMS support, two "features" you rarely see in Linux video drivers together at the same time (Either you have great OpenGL support on your card or you have KMS, at least that's the story with AMD and nVidia for now, Intel gets you both but lacks the hardware capability to really back it up.).

4. Driver both upstream and downstream can fix and adapt.

Get used to it.

I'm used to it, believe me. This doesn't mean we should abandon developing open source drivers, especially when it means avoiding kernel taint and supporting things like Wayland which require KMS and will therefor not be supported at all by Catalyst or the nVidia blob.

[scjet]

U're just plain wrong

"...Was Linus Torvalds right in his approach to nVidia? No. I honestly think the "fuck you, nVidia" attitude helps nothing...."

@Yaro: You are COMPLETELY wrong here.
You see, it's NOT about "nvidia", or "amd", or whoever sleazy proprietary graphics card maker's exist.

It's, in reality, ALL about ALL Linux/Unix User's,..., big and/or small.
Actually, what is really sad, is that not enough "manufacturers" have the "BALLS" to try to build hardware just for Linux.
Let's face it, Windooze is just another completely proprietary "Tablet", along with Apple, Google, ..., that you'll have to buy again, and again, and again,..., whenever "THEY" see fit.

Soooooo, you're absolutely wrong, and Torvalds and me, are absoltely RIGHT, in telling nvidia, ... ,: "fuck you". !

Wake up, man !

[Yaro]

"...Was Linus Torvalds right in his approach to nVidia? No. I honestly think the "fuck you, nVidia" attitude helps nothing...."

@Yaro: You are COMPLETELY wrong here.
You see, it's NOT about "nvidia", or "amd", or whoever sleazy proprietary graphics card maker's exist.

Why not? How we are able to use graphics on Linux is almost entirely up to them. nVidia provides no documentation, so Nouveau has to reverse engineer their drivers. AMD provides some documentation and cooperates with radeon developers, that open driver is comign along.

Intel provides official, completely open source drivers. Our "support" such as it is is entirely reliant on what the manufacturers of our hardware offer us. Acting like a bunch of idiots because they don't cooperate isn't going to convince them to increase support, it's only going to scare off potential users and hardware vendors from Linux.

It's, in reality, ALL about ALL Linux/Unix User's,..., big and/or small.

I'm sorry, was there grammar in that sentence?

Actually, what is really sad, is that not enough "manufacturers" have the "BALLS" to try to build hardware just for Linux.

"Balls" has nothing to do with it. It's entirely about whether such a thing is even remotely cost effective. You expect designing hardware and drivers as well as mass production, marketting, and even licensing is cheap? On the desktop, Linux is not anywhere near prevalent enough for such an investment for hardware companies to even consider it. Building hardware "just for Linux" on the desktop is a waste of money, especially given that operating systems, by design, should not require hardware to be custom fit for them. Why do we need hardware "just for Linux" when the hardware we have is perfectly usable and just needs support?

Let's face it, Windooze is just another completely proprietary "Tablet", along with Apple, Google, ..., that you'll have to buy again, and again, and again,..., whenever "THEY" see fit.

First off, using "quotes" for "emphasis" is just plain incorrect usage. Learn how to use them or don't use them "at all."

Next, Windows is an operating system, Apple and Google are companies. None of them are tablets. Your entire sentence makes absolutely no sense.

Further, I don't see Microsoft, Google, *or* Apple forcing *anybody* to do anything. You're perfectly fine using the hardware or devices you want. I don't have to buy anything again and again whenever they see fit. That entire argument is incorrect.

Finally, Microsoft, Google, and Apple are entirely IRRELEVANT to what hardware we use under Linux, that's on hardware manufacturers, most of whom, including nVidia, are supporting Linux. Make a logical argument please or go back to middle school.

Soooooo, you're absolutely wrong, and Torvalds and me, are absoltely RIGHT, in telling nvidia, ... ,: "fuck you". !

Wake up, man !

I'm wide awake. That's why I don't blindly agree with Linus Torvalds on everything or assume just because a company isn't fully embracing open source they're suddenly evil.

Nor am I making entirely irrelevant rants or throwing out utter nonsense to try and justify something a PROFESSIONAL never would have done (I should note Linus IS a professional software engineer and if he had worked for any other organization or been any other software engineer he'd have been fired for the "fuck you" alone.).

You made no logical, condusive argument why you or Linus should tell nVidia "fuck you." And my point still stands, it doesn't motivate ANYONE to actually fix any existing problem and it makes the Linux community look hostile to any hardware manufacturers who may consider supporting Linux in the future if they choose to exercize their LEGAL RIGHT not to share.

I'd have rather actually opened up channels of communication with nVidia. Negotiated with them. Talked to them. Do something vastly more productive and open than a "fuck you." Instead I would have at least tried to form a coherent argument to convince nVidia to cooperate. If anything Linus just gave them one more reason not to help them out.

[artivision]

Things are very simple. Wine has two ways to work with D3D games. 1)An OpenGL based D3D-10 Compiler (WineD3D) that also works on Windowz. This is CPU light but Graphical not efficient (between 50-80%), and that's because D3D and OGL are different. 2)A to-GLSL Shader Re-constructor. This has 90% Graphical efficiency but uses 2-3 times more CPU, and that's because you represent a Shader to GLSL-source and then you compile it again to GLSL-bytecode. Today there is not merge of those two techs, so best of both is not able. Also those two programs there aren't Multi-threded, so opening two Compiler (rendering) threads is impossible. You can only open one, two threads are possible if the Game can do separate Game+Rendering. So Games like Guild_Wars_2 are nearly not playable (less than 30 FPS) with any system setup and GLSL=enabled (Wine default). CPU bottlenecks and GPU does never fill. There is only a way to survive in this difficult situation and that is "winetrics glsl=disabled" in order to use WineD3D, that doubles your FPS. The bad thing is that only run with Nvidia Cuda GPUs and Nvidia closed_driver. So its up to Wine to correct the CPU problem in order to play everywhere good, or vendor drivers must care about WineD3D to play good with their driver, so far only Nvidia works. Intel is in a good situation because we can do things on their open_driver, because its Open!!!