NVIDIA needs Linux in 2023 much more than Linux needs them.

10 years ago, it was different. Mostly Linux geeks who wanted to play games, who were minority of their sales, cared about it.

Now, basically all of NVIDIA valuation is built on AI, and almost all major players who buy their HW, and a lot of it, use it on Linux.

Cutting Linux off is not an option for NVIDIA.

Nvidia already has Linux (up to this change).

I'm not sure Nvidia needs Linux 6.6+. If this change is indeed a dealbreaker, they can very well fork it at the last working version and offer it in combination with their own distro.

This kind of bullshit pettiness will end up making the situation worse for desktop Linux users and consumers, while enterprises reliant on Nvidia equipment in Linux-based server systems will just jump to Nvidia's fork and distro.

No, I don't think they can really fork it. How are they going to get support for new hardware in their fork?

Hardware can be abstracted away by a hypervisor, as it's already done in most use-cases since a significant amount of machines are just cloud VMs nowadays. The only API the "hardware" has to implement is VirtIO (which real hardware could also start implementing if needed).

This kind of fork would of course mean death to the Linux desktop, since the only reason the Linux desktop is relevant and (somewhat) modern is because it's enjoying the improvements that are made to Linux by companies who primarily use it on servers.

If this fork happens, this VM-based distro is what enterprises will switch to and build upon, with conventional Linux being relegated to niche use-cases with much less activity and manpower.

Lol, what do you think runs those hypervisors, macOS and Windows...?

There are already VM/container-focused distributions (Alpine etc). There is no real ecosystem fork and likely never will be.

Nvidia is the default option because it's convenient at all levels of the stack. Close to two decades of investment into CUDA made sure that GPGPU with Nvidia is extremely smooth, and AMD support is a lot more spotty and less maintained. But add arbitrary restrictions like needing some outdated kernel or a special distro, and suddenly getting an AMD card to work with your favorite framework doesn't seem like so much more work. And once that reaches critical mass and patches are upstreamed, the convenience advantage of Nvidia is gone and competition will get a lot more fierce.

> making the situation worse for desktop Linux users and consumers

Most desktop Linux users don't need NVIDIA even a little, because AMD's GPUs will serve them just fine. They're not doing AI.

And a decade ago most users didn't need a GPU at all because they weren't doing 3D. I still have a laptop with a Vista compatibility card in it that was outdated on release because the push for using GPU accelerated graphic back ends didn't stop at the OS and everyone's grandparents also wanted to have a look at things like google earth and street view.

If and when everyone is running AI workloads, then all of the competition will need to have proper support for it.

Or at least we hope Nvidia wouldn't be left as the only vendor of modern GPUs in that case. That would be very unfortunate.

There’s this bubble where people who work with AI seem to assume that everyone else also wants to work with AI on their PC. It’s strange. Let NVIDIA have their CUDA and their datacenter GPUs, most of us just want to put pixels on a screen.

You greatly underestimate amount of work that would be needed to hard fork and maintain Linux kernel (even powerhouse like Google doesn’t do it, and they’ve build multiple operating systems), in a state that would be attractive to AI, where you care a lot about performance, latest drivers (for thing like networking cards and other bleeding edge stuff that comes with AI gold rush).

And willingness of others to follow. NVIDIA needs Linux to sell their HW. For Linux NVIDIA isn’t even 1% of their market.

But can NVIDIA maintain a patchset that just removes these new protections from the kernel instead of a complete fork?

Possibly. But I think from the legal point of view it could be much more direct “I’m doing this to break the license” approach, where it can viewed as them try to freeload off up to date work. For older version they had some possible explanation that they just used an API.

INAL.

Losing their dominant position in Linux would be catastrophic for Nvidia; they’ll do anything to avoid that happening.

Almost anything; they won’t GPL their driver like they ought to, because they made unwise decisions surrounding the use of material covered by software patents and non-Free licenses, and thus are unable to do so.

Aren't they re-architecting & creating a new, free software driver that's only responsible for memory management & conveying requests from the user through to to the _real_ computer running on the GPU?

Don't clutter up this thread with actual facts and information.

Unlikely. AI projects are tied to Nvidia not the Linux. They just port their driver to FreeBSD or Solaris or whatever. The AI crowd will follow. But they can also just release patches that as part of the driver install patches the end user's kernel to remove this janky stuff.

i'm out of the loop on this one. i did a good deal of HPC -- specifically electromagnetics simulation -- about a year ago and Spir-V worked OOTB for GPGPU on AMD processors. hell, i wrote my kernels in _Rust_ and they compiled and ran just fine (Embark Studio's `rust-gpu` target), which is about as edge-casey as i can imagine.

how is AI stuff tied to nvidia? what would make it difficult to port (or to be portable)?

AFAIK the vast majority of ML training/inference on the GPU runs on CUDA, and has done so for many years, while other vendor-specific GPU backends are slowly gaining support.

It doesn't help that AMD's equivalent open toolchain, ROCm, is barely even supported by their own GPU lineup.

FWIW, ROCm GPU support is getting better, particularly in the distro-provided packages. e.g. https://salsa.debian.org/rocm-team/community/team-project/-/...

I think Linux should mirror what OpenBSD does, remove all support for nVidia until it is opened up.

But these days, commercial companies have more say over what and how hardware is used in Linux than we do.

And OpenBSD has how much share in workstation/graphics market?

> less games

AMD cards work very well in linux.

yeah the open source driver is pretty solid

the propritary driver for nvidia was generally decent, but i got an AMD card for my last build because the FOSS driver was good

>[OLD] AMD cards work very well in linux.

The new ones crash constantly and have new crash bugs every release and still 2 years later don't properly support 120hz @ 4k or anything HDMI 2.1 despite being advertised that they a) worked on linux and b) supported HDMI 2.1

HDMI 2.1 is not currently compatible with an open driver (although there were rumors that AMD was working on the licensing issues). In the same vein as it's not really productive to complain about NVIDIA wanting to link proprietary code, it's also not really productive to complain about AMD wanting to link proprietary code.

Buy a DP to HDMI active converter and stay open, or else accept the proprietary software into your stack. If that is not possible, look elsewhere for your software, GPL is not for you.

>HDMI 2.1 is not currently compatible with an open driver

This is a false narrative, it's literally just not implemented.

>Buy a DP to HDMI active converter and stay open, or else accept the proprietary software into your stack. If that is not possible, look elsewhere for your software, GPL is not for you.

This is just so... user hostile. The paid amd developers can't get around to finishing their drivers so use a proprietary dongle with closed source everything in between your screen and your computer, and if I don't accept this "GPL" isn't for me. Lol.. just.. no.

> This is a false narrative, it's literally just not implemented.

It’s literally not, it’s what AMD’s maintainers have said themselves?

> HDMI 2.1 is not available on Linux due to the HDMI Forum. See: https://www.phoronix.com/scan.php?page=news_item&px=HDMI-Clo...

https://gitlab.freedesktop.org/drm/amd/-/issues/1417#note_83...

Like it literally does come down to the same thing as people say about nvidia: this is closed source and it’s not compatible with copyleft licensing. GPL can’t touch proprietary symbols any more than proprietary code can touch GPL symbols.

Nvidia can support it because they have a closed firmware and the symbols are buried in blobs. And that has its advantages too, like this situation!

If you can chase HDMI forum to open a minimal set of symbols, fine, and it sounds like AMD has been working on that. But there is a lot of closed IP in the graphics world that will simply never be opened and it’s unlikely that the closed driver will ever be opened up as a result (the open driver will be a rewrite without IP taint). Graphics IP is just not a world compatible with copyleft and they’re ok with that, if copyleft suffers consequences oh well, Linux is probably <1% of their revenue base.

Copyleft is fundamentally about inventing a new, joyfully communist codebase, not gluing proprietary code onto community code. That’s why it’s GPL and not BSD - Nvidia’s license or hdmi forum’s license is not a problem on FreeBSD! And in this case you simply must invent your own 48gbps display standard, but with hookers and blow, and then drive sufficient adoption in the market, just as Linux had to do for software. Or someone already did that and it’s called DisplayPort.

But hdmi is proprietary license and you can’t use it unless and until they relicense to something GPL compatible. That’s the price of the Linux kernel, is it delivering some value here that BSD or minix wouldn’t for you?

Again though it is simply the reality of how things are in the audiovisual field, you will never get Dolby or Fraunhofer or or motion picture expert group or creative to open up any of their proprietary IP so you will simply have to accept a feature deficit OR use a license other than GPL that tolerates linking with proprietary code.

It is a testament to how thoroughly and utterly open-source (and specifically copyleft) has won that people are completely unable to contemplate the possibility that some code simply cannot and should not be released under viral copyleft licensing. And people can’t even consider other open licenses like BSD etc.

I have a newer card and it works very well