Maybe something will change on this soon. There was speculation about this: https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-O.... But I'm not holding my breath, and it would be nice if the solution wasn't "wait and hope until Nvidia releases the software necessary to control their GPUs".
Do you have more info on this ? There is a big difference between not supporting open source, and actively sabotaging it. What are they doing, exactly ?
TL;DR starting with the 9xx series, Nvidia started making it so their GPUs would only run firmwares signed by them (likely to prevent counterfeits, i.e. 2060s sold as 2080s). So it is impossible to control the fans and reclock the GPU. There's no workaround, so even as the person who owns the device, I can't run my own firmware. AMD has signed firmwares too, but they actually release sufficient blobs to fully run the device.
When I had nvidia hardware, I never tried nouveau, so I never ran into this.
https://github.com/ROCm-Developer-Tools/HIP https://github.com/ROCmSoftwarePlatform/hipDNN
CUDA just works.
The 5000 series has been out for an entire year without ROCm support at this point.
It is amazing to watch how much of a struggle AMD is having with getting PyTorch to work with ROCm. It makes me appreciate what a good job Nvidia must have done with CUDA.
(To be fair AMD has it planned, but it's behind several other priorities to them.)
LOL what a fucking joke. Not even close.
Also that one VFIO bug makes one of the great advantages of Radeon disappear and there is no fix in sight.
Fuck AMD. Bunch of marketing hype and they ship garbage half baked crap to the market.
I thought AMD had great Linux driver support, and open source drivers.
Turns out AMD only has the open source driver part of the things, and this driver does not support a large chunk of their products.
I'd take a binary driver over no driver any day (and in fact, I returned the amd card and am using intel now, which works great).
AMD: some open source drivers, but many things are inop
Nvidia: all closed source, but everything works
Source: I have a Vega 56 and I am gonna give it the sledgehammer when 3000 series arrives. Fuck that shit to high heavens my dude. ROCm is buggy as fuck. Only the bare basics work on Linux. Drivers are buggy, and crash all the time. Even the most basic bullshit is not implemented right. Like there is no fan control. That's how bad it is.
FP32 performance for the VII is comparable to GTX20280Ti e.g. on Resnet training. Tensor Cores are cool, but 90% of people don't use FP16 that much.
I'm not sure that is the best reply that stackoverflow has ever seen.
Most frameworks do support ROCm though, so the ball is on AMDs court.
But your mileage may vary, I don't want to make people swear.
If AMD is unwilling to properly support its consumer GPUs in GPGPU workloads and NVIDIA is, then well... sounds like a good reason not to buy AMD.
It's really not a major ask, people have been running GPGPU programs on consumer cards since forever, even though they're "consumer" and sold as graphics cards and not compute accelerators. NVIDIA basically does this on purpose, people use their personal hardware to get a foot in the door of the ecosystem and end up writing programs that get run on big compute GPUs which NVIDIA makes big profits on, it's very intentional.
If AMD chooses not to do that, well... can't really blame people for avoiding their stuff when NVIDIA is willing to let you do this stuff on their cards and AMD isn't.
Even there though, only Intel has a buttery smooth experience. Ryzen for laptops is half baked (terrible USB C docking performance, occasional crashing on Windows and Linux with the 2xxxU series CPU/GPU chips) and AMD GPUs still require manual intervention to load proprietary firmware.
AMD does make some performant mobile GPUs though, they work well in Debian!
I switched to Wayland a few years back as vsync is quite nice to have, but whenever I go back to Xorg for AnyDesk or TeamViewer on AMD or Intel there is a fair bit of tearing.
Nvidia could have a competitive open source driver tomorrow if they released redistributable firmware that allowed reclocking the GPU.
As a FreeBSD user I'm still "trapped" on Xorg anyway, and in any case I'd rather stick with what works than learn some whole new way of doing things for marginal benefits.
I never said it was. I said it's a whole new way of doing things for marginal benefits; is that not accurate?
OpenGL 3.3 when the proprietary drivers do OpenGL 4.1 for example.
I have an Intel GFX card. Had an AMD 5700. Linux support sucked, switched back to intel.
So AFAICT, only intel and nvidia have good Linux driver support, and only intel has good open source linux driver support.
Microsoft own previous gen Xbox emulator on the next gen xbox (i think it was original xbox emulated in the 360, but i might be wrong) was impacted by the team having to reverse-engineer the GPU because nvidia refused to let the emulator people to have access to the documentation provided to the original team.
I am not sure if I am missing some update, need to set some undocumented kernel flag and/or BIOS setting, if it's a software issue or Í just made a mistake somewhere. Debian 10/11.
Meanwhile, as much as I wanted to get away from Intel, their drivers have never posed any issue at all.
https://www.gpuzoo.com/GPU-AMD/Radeon_HD_6320_IGP.html
The open source driver is kind of ok if the only thing we expect from it is getting a working X session.
Now if one wants to do some complex OpenGL stuff, then it might work, or not.
I used to have a HD 7950 and it always worked perfectly, same with my current Vega 56.
The usual linux answer to hardware problems, keeps being to buy new hardware.
But at some point you have to consider if it's really worth it keeping a 10 year old laptop around. It's painful to say them goodbye, I know, I have been there, but for me it's just not worth it.
No need to throw a perfectly working laptop to enjoy the DirectX 11 and OpenGL 4.1 capabilities that it was sold for.
Even on Windows AMD drivers are the most unstable, bugged software that’s even been shipped. It’s been a long standing joke that AMD “has no drivers”.
The open source driver replacement only does OpenGL 3.3.
I guess I should be happy it does any OpenGL at all.
Still no driver for compute 1 year later. I'm so happy i decided to return it and switch to intel instead of waiting for AMD or some random joe on their free time to add support for it to their open source driver.
So yeah. I'd take a working proprietary driver over no driver any day.
Is this an Ad Hominem ? Linus does not mention there a single thing that they are actually doing wrong.
> Drew DeVault, on Nvidia support in Sway [2]:
Nvidia has added wayland support to both KDE and GNOME. Drew just does not want to support the nvidia-wy in wl-roots, which is a super super niche WM toolkit whose "major" user is sway, another super super niche WM.
Drew is angry for two reasons. First, sway users complain to them that sway does not work with nvidia hardware, which as a user of a WM is a rightful thing to complain about. Second, Drew does not want to support the nvidia-way, and it is angry and nvidia because they do not support the way that wl-roots has chosen.
It is 100% ok for Drew to say that they don't want to maintain 2 code-paths, and wl-roots and sway do not support nvidia. It is also 100% ok for nvidia to consider wl-roots to niche to be worth the effort.
What's IMO not ok is for Drew to feel entitled about getting nvidia to support wl-roots. Nvidia does not owe wl-roots anything.
---
IMO when it comes to drivers and open-source, a lot of the anger and conflict seems to steem from a sentiment of entitlement.
I read online comments _every day_ of people that have bought some hardware that's advertised as "does not support Linux" (or Macos, or whatever) being angry at the hardware manufacturer (why doesn't your hardware support the platform that says it does not support? I'm entitled to support!!!), the dozens of volunteers that reverse engineer and develop open source drivers for free (why doesn't the open source driver that you develop in your free time work correctly? I'm entitled to you working for free for me so that I can watch netflix!), etc. etc. etc.
The truth of the matter is, that for people using nvidia hardware on linux for Machine Learning, CAD, rendering, visualization, games, etc. their hardware works just fine if you use the only driver that they support on the platforms they say they support.
The only complaints I hear is people buying nvidia to do something that they know is not supported and then lashing out at everybody else due to entitlement.
You're now somehow arguing with people that they should stop complaining about Nvidia's business practices. I would agree with that in the sense that Nvidia can do whatever they want: nobody is obliged to buy Nvidia, and Nvidia is not obliged to cater to everyone's needs. It's a free enough market. But even if you don't agree with some/most of the complaints surely you must agree that Nvidia's track record of pissing of both other companies (and people) is problematic for when they take control of a company with an ecosystem driven business model like ARM's?
I'd agree with you this is OP's argument, however it's main flaw is in explicitly omitting the fact that NVidia is not the only party that's "free" to do things.
We're not obliged to buy their cards and we aren't obliged to stay silent regarding its treatment of the open-source community and why we think it would be bad for them to acquire ARM.
I am always amazed at the amount of pro-corporate spin from (presumably) regular people who are little more than occasional customers.
We still are. I asked about "which specific business practices are these", and was only pointed out to ad hominems, entitlement, and one sided arguments.
Feel free to continue discussing that on the different parent thread. I'm interested on multiple views on this.
> You're now somehow arguing with people that they should stop complaining about Nvidia's business practices
No. I couldn't care less about nvidia, but when somebody acts like an entitled choosing beggar, I point that out. And there is a lot of entitlement in the arguments that people are making about why nvidia is bad at working with others.
Nvidia has some of the best drivers for Linux there are. This driver is not open source and distributed as a binary blob. Nvidia is very clear that this is the only driver that they support on Linux, and if you are not fine with that, they are fine with you not buying their products. This driver supports all of their products very well (as opposed to AMD's, for example), its development is made by people being paid full time to do it (as opposed to most of their competitors which also have people helping on their drivers on their free time - this is not necessarily bad, but it is what it is), and some of their developments are contributed back to open source, for free.
People are angry about this. Why? The only thing that comes to mind is entitlement. Somebody wants to use an nvidia card on Linux without using their proprietary driver. They know this is not supported. Yet they buy the card anyways, and then they complain. They do not only complain about nvidia. They also complain about, e.g., nouveau being bad, the Linux kernel being bad, and many other things. As if nvidia, or as if the people working on nouveau or the Linux kernel for free on their free time owes them anything.
I respect people not wanting to use closed source software. Don't use windows, don't use macosx, use alternatives. Want to use linux? don't use nvidia if you don't want to.
If you ask your impression of Nvidia's business practices, and they give you their opinion, you can't somehow invalidate that opinion by retorting with debate fallacies. That's the "fallacy fallacy" if you're sensitive to that. This is not a debate competition about who's right, this is people giving their opinions based on Nvidia's past and current actions. You asked a question, and they answered. This is not a competition. Please give them the basic respect of acknowledging their opinion.
We are discussing a topic, and people throwed multiple arguments that do not make sense.
You are claiming that I should just shut up and respect their feelings, but that is worthless.
Two examples:
---
Somebody's argument was: "Linus doesn't like them, therefore I don't like them".
The reason these are called logical fallacies is because these arguments are illogical. I told them that this was a logical fallacy (argument of authority - just because someone with authority makes an argument does not mean they are right), and ask them _why_, what is it that linus and you do not like.
I am happy I did that, because many of them have raised multiple actually-valuable arguments in response. For example, because nvidia's hardware throttles down if the driver firmware is not signed, and this makes the open source drivers slower for no reason.
That's a valid and valuable argument. Linus doesn't like them is worthless.
The person who raised this argument learned something from somebody else which knew what Linus did not like, and so did I.
---
The same happened when I called out the entitled choosing beggars. "Why are you angry at nvidia for not providing an open source driver ? You knew before buying their product that only the binary driver was supported."
Read the responses. The reason they are angry, is because they don't have a choice but to use nvidia, because the competition products (AMD in those cases) are much worse. AMD does have open source drivers, but they are crap, and they don't support many of AMD's products, at least for compute, which is something that many (including myself) use for work.
These people have picked a platform that values open source code, but due to their job requiring them to actually get some work done, they must use nvidia for that, and they don't like having to compromise on a proprietary driver.
Honestly, I think this is still entitlement, but I definitely sympathize with the frustration of having to make compromises one does not like.
---
From the point of view of whether nvidia buying ARM is good or bad. I still have no idea. ARM does _a lot_ of open source work, its major market are Android and Linux communities.
I understand that people are afraid that Nvidia will turn ARM into a bad open source player. It can happen. But without Android, iOS, and Linux, ARM is worthless. So a different outcome of this could be that NVIDIA buying ARM ends up making NVIDIA more open source friendly, since at least the Linux market is important for nvidia as well (~50% of their revenue).
It definitely makes sense for regulatory authorities to only allow somebody to buy ARM that will preserve ARM deals with current vendors (apple, google, samsung, etc.), and that also will preserve ARM open platform efforts.
If nvidia does not agree to that, they should not be allowed to buy arm.
Except that they stop supporting older HW at some point. That, together with occasional crashes learned me not to buy nVidia HW again.
NVIDIA insisted on pushing its own EGL streams even as the wider community was moving in a different direction.
They suffer from a major NIH syndrome and do not know how to work with others at all.
Hey! Wait a second, there. Nvidia isn't bad because it has a properietary Linux driver. Nvidia is bad because it actively undermines open-source.
Quoting Linus Torvalds (2012) [0]:
> I'm also happy to very publicly point out that Nvidia has been one of the worst trouble spots we've had with hardware manufacturers, and that is really sad because then Nvidia tries to sell chips - a lot of chips - into the Android Market. Nvidia has been the single worst company we've ever dealt with.
> [Lifts middle finger] So Nvidia, fuck you.
Nvidia managed to push some PR blurbs about how it was improving the open-source driver in 2014, but six years later, Nouveau is still crap compared to their proprietary driver [1].
Drew DeVault, on Nvidia support in Sway [2]:
> Nvidia, on the other hand, have been fucking assholes and have treated Linux like utter shit for our entire relationship. About a year ago they announced “Wayland support” for their proprietary driver. This included KMS and DRM support (years late, I might add), but not GBM support. They shipped something called EGLStreams instead, a concept that had been discussed and shot down by the Linux graphics development community before. They did this because it makes it easier for them to keep their driver proprietary without having work with Linux developers on it. Without GBM, Nvidia does not support Wayland, and they were real pricks for making some announcement like they actually did.
[0]: https://www.youtube.com/watch?v=iYWzMvlj2RQ
[1]: https://www.phoronix.com/scan.php?page=article&item=nvidia-n...
[2]: https://drewdevault.com/2017/10/26/Fuck-you-nvidia.html