The end of Nvidia proprietary Drivers on Linux !?!

A lot of supercomputer too (4 of the top 10). But i don’t know how far they need proprietary driver to use cuda.

I’ve been surprised to see that the top 1 supercomputer is based on ARM cpu :wink:

1 Like

I agree but there isn’t a alternative to proprietary drivers for nvidia graphics cards. Pushing to nvidia from supercomputers it could be still enough strong.

Yes i surprised too, Fugaku is the first supercomputer based on ARM.

video on soft from nvidia
https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Open-Source-ASF-2020

getting rid of the old concept of “proprietary driver” would benefit to their annual sales figure.
In short, the “drivers war” has to stop…

6 Likes

Is there any update on this front?

Still waiting to see what the corporate overlords do and what the kernel jedi do about it.
So, no, not really.

3 Likes

There’s apparently a kernel patch that fixes this, it’s just whether or not you guys want to use it.

The problem with this assumption is that it will take years for most of those Nvidia customers running servers to get to a kernel version that has this issue. They’ll be on 5.4 (or 4.19, or earlier) for years.

It could be 2-3 years before Nvidia even notices enough to be forced to take action. Meanwhile, we’re all in limbo for that time period.

1 Like

What limbo? Dont support bad actors - dont buy nvidia :wink:

(ok ok … jokes is jokes … of course manjaro strives to be compatible. But its inarguable nvidia makes that difficult)

3 Likes

That’s easy to say for people that don’t prioritize the best performance in areas like gaming.

But for people that do care about those things, there is no other option (at least not right now).

And there’s not likely to be another option any time soon. Even though AMD may compete on pure performance, their drivers are a complete disaster for months (or years) after a new architecture launches. Hell, we’re already seeing fixes for RDNA 2 that missed 5.9 and won’t come until 5.10, and that’s before the hardware even launches. Given that I own two Navi GPUs, I can speak on this with (very painful) first hand experience.

Trust me, I wish Nvidia didn’t make things this difficult, and I am a very, very big fan of AMD and root for them at every opportunity (like I said, I own two Navi GPUs lol). But AMD can’t run the software I want at the resolution and refresh rates I play at, and honestly it can’t even come close, so I went with a 3000 series GPU, and aside from this nvidia_uvm issue, the difference has been astoundingly positive. Which makes it all the more frustrating that we have this ideological war going on. Ethically, I’m fully on the side of the kernel devs, but practically, I think they went about it in the most petulant way possible. They absolutely could have spoken with Nvidia and worked something out, instead of just fast-tracking this into the kernel. I know even Gentoo devs have Nvidia contacts, so we know damn well the kernel itself does.

I get that you were kidding, but that “just don’t support Nvidia and buy inferior hardware” attitude is on display constantly, and with all sincerity, from other people in the community.

4 Likes

Dont see that as valid at all. AMDGPU works fine.
And besides … thats kind of a rich kid argument.
AMD pound for pound is more performance for the dollar.
Only people willing to spend hundreds of USD on a GPU alone will be able to break that barrier.
(and its still an inefficient cost/benefit ratio)

I heartily disagree. Nvidia is the inferior hardware here.
We might have different metrics for that … but I feel comfortable saying it.

7 Likes

Well that unfortunately shows your ignorance and nothing else…

And besides … thats kind of a rich kid argument.

Lol I live in a one-bedroom $500/month apartment and drive an 18 year old car I bought for $1800. I saved up for literally 9 months (no eating out, zero spent on any entertainment, nothing) to buy my RTX GPU. I’m not a “rich kid.”

AMD pound for pound is more performance for the dollar.

Not right now, they’re not.

https://www.phoronix.com/scan.php?page=article&item=rtx3080-linux-gaming&num=7

RTX 3080 is 2X the 5700 XT at way less than 2X the price.

AMDGPU works fine.

No, it doesn’t. Navi was a complete disaster for 6-8 months after launch, and is still a disaster for huge numbers of users on Linux. There are currently like 100+ open bug reports (some of which have been open since the cards launched) regarding random driver crashes and hangs with RDNA 1 GPUs on Linux. I’ve experienced them myself. And essentially nothing’s been done to fix it other than enabling GPU reset so you just lose your whole X session instead of having to do a hard reboot.

AMDGPU is fantastic for Polaris and Vega hardware (both of which I’ve also ran on Linux), but it is not in a good state for RDNA 1. It can’t even remotely be argued that it is, the absolute inundation of bug reports says it all.

Nvidia is the inferior hardware here.

It’s objectively not, by any measure. Price-performance != performance, and whether or not the drivers are open is software, not hardware.

Your issue seems to be a tendency for demagoguery. I’ve already said I love AMD (I also own two Ryzen CPUs, and have only ever ran AMD GPUs until now), but I can at least acknowledge their issues. Mostly from personal experience. People with ideological bones to pick seemingly can’t. That’s not to say Nvidia doesn’t also have those people, there are far more of them that Stan for Nvidia than AMD (though with Linux it’s mostly the AMD side). That doesn’t make either side more tolerable, though.

Also, whether or not it’s a “rich kid” argument (which it’s not) should be irrelevant. This is going to affect people with 1650 Supers just as much as it’s going to affect people with RTX 3090s. And when you get down to the 180-200 USD price point, AMD is not really better on price-performance than Nvidia. Where AMD shines in price-performance is from $280-400, with the 5600 XT, 5700, and 5700 XT. The 5500 XT is flat-out terrible price-performance. But that’s all #off-topic. Bottom line, Nvidia shares blame for this, but it could have easily been avoided as well had the kernel devs gone about it in any other way.

4 Likes

If nothing else, my time on these forums has taught me that if I ever need to buy a discrete GPU, not to buy NVIDIA… whatever the benchmarks.

2 Likes

I guess this explains things for me. My NUC has Polaris. Never run into any issues.

I know it’s slightly off-topic, but while reading this, the main thing I was thinking was “I don’t think anyone’s been able to spend much on dining out and entertainment since about March”. Nothing involving leaving home, anyway.

Alright … looks like you want to take it personally. Fine enough I guess … I am not going to explain relative financial scenarios acrossed the globe or anything. You think nvidia is fine. cool. Nothing about how either of us feels will have any impact on its support.

1 Like

The nvidia_uvm module no longer loads on 5.9.

1 Like

I am just seeing this thread now. @pobrn: thank you for pointing me here. :wink:

But what I am reading here is in many cases as religious as the kernel developers who want to close out nvidia. I honestly do not get it. This fight of the kernel developers against nvidia has no winners. Preventing anybody from using parts of the kernel API because of their heritage is not good. Reminds me of Microsoft fighting DR DOS by changing their code and making DR DOS crash. This was insane when Microsoft did it and it is insane when the Linux developers are doing it to NVIDIA.

And the GPL is no excuse. The GPL is not forcing the developers to do that. It is their free choice. They do it because they are fundamentalists. Religious zealots they are.

But the good news is that the Linux kernel is open source. And any distribution can patch anything in the kernel. So my expectation is that either nvidia finds a work around or that the various distros will simply patch this anti-nvidia code out of the kernel. But what we are experiencing now with kernel 5.9 is surely not the end of proprietary nvidia drivers on Linux.

1 Like

I don’t know about you, but one is usually reminded of esoteric religions and obscurantism, in the case of proprietary software.

1 Like

They don’t want to close out nvidia in general, but proprietary. Once Nvidia goes opensource, they are more then welcome in the kernel.

2 Likes

@BS86
I am sorry but you don’t get the point. This is not about open source. openzfs is open source as well. And still the kernel developers are fighting zfs because it is not GPL. Here is a list of all open source licences: Licenses by Name | Open Source Initiative

This is about GPL fanatics wanting to have everything under the GPL.