Hacker News Reader

Comment by atemerev | Hacker News Reader

So, RedHat decided that they do not want nvidia users anymore? I’d be more than happy to switch to Wayland, but I can’t.

View on HN

I'm under the impression it's more like NVidia does not want Linux users, or at least has turned some WM developers against them[0].

[0] https://drewdevault.com/2017/10/26/Fuck-you-nvidia.html

Nvidia is mostly used on Linux, as most of their GPUs are installed in datacenters for deep learning acceleration purposes. More than that, Nvidia is the only major video hardware manufacturer that keeps up to date drivers for other free UNIXes, e.g. FreeBSD.

> "Nvidia is mostly used on Linux, as most of their GPUs are installed in datacenters for deep learning acceleration purposes."

Supposing that's true, it has little to do with the reality that Nvidia blatantly doesn't give a damn about the linux desktop uses of their cards.

These constant comments about the issue blaming everybody but the actual party responsible (nvidia) really are becoming tiring. If you can't be bothered to research whether hardware is well supported by linux before you buy it, then what the heck are you even doing in this industry? Merely searching "is nvidia total shit on linux" online should have been enough to inform your hardware purchase. You and nvidia made this bed, stop blaming other people when it's not comfortable. Not a single time in my life have I purchased an nvidia card. You have the choice too.

Datacenters are not desktops though, which I assume is what X11 is used most for. But, like I said, my comment was just my impression regarding the matter, based on the blog post I linked. Thanks for your reply! :)

Most of their GPUs are still used for gaming and graphics acceleration, but you are right that their cards and drivers generally work well on Linux and FreeBSD. I think the main issue is that they have no intention of making their driver open-source which understandably irks the Linux community.

GNOME, which is what powers the RedHat desktop, supports nvidia on Wayland just fine, and KDE got nvidia/EGLStreams support since KWin 5.16. So at least 2 major DEs do, though quite a few others don't, and have taken a principled stand not to (Sway for example won't).

When it comes to RedHat though, they're all-in on GNOME anyway so keeping X around for nvidia users isn't really a concern for them when their desktop experience supports nvidia on Wayland just fine.

No, just checked it with the latest Fedora 30, it defaults to X when it detects an Nvidia card.

Just b/c it defaults to X does not mean it doesn't work or isn't supported on Wayland.

It defaults to X, because it works very poorly with Wayland. Until very recently, it was impossible to run Gnome with Wayland at all (even though it theoretically supported Eglstreams), as there were many kernel errors generated by gnome-shell, etc. Now it sort of works, however, lags and freezes make it completely unusable in most cases. I am testing it every Fedora release. It _is_ possible to use Gnome with Wayland, I managed to get it working with Arch Linux once. However, it took me 4 days, manual patches (literally editing the WM code myself to fix some issues as suggested by the community) to get the configuration right, and the next rolling Arch update broke it beyond repair.

Again, I am not against Wayland. Wayland API is beautiful and I want it adapted more and sooner. It is just sad that me and many other users are not getting it. Developers blame Nvidia, because they do not open source their driver (which is a valid concern), however, there are many other binary blob drivers in mainstream Linux kernel, and everybody except hardcore open source apologists are OK with that. Why Nvidia is different, I don't understand.

Developers don't blame nvidia b/c they don't want to open source their driver. They blame nvidia for insisting on others shouldering the tech debt of EGLStreams (a closed, proprietary nvidia-only solution with no documentation whose implementation can't easily be tested and verified by compositor authors) instead of enabling the use of GBM with their drivers like every other Linux graphics driver. nvidia doesn't have to open up their drivers for that, AMD's closed source driver (so not the Mesa stack), supports GBM just fine.

Fedora 30 still uses 5.15. Archlinux had 5.16 for 3 weeks already.

Kwin 5.16 might technically support EGLStreams, but I have an Nvidia card and tried it out - after setting env variable to even make it try to work, Plasma crashed during startup, kicking me back to login. I tried again, and it managed to start up, but then plasma-shell crashed when I tried to search something. Performance varied dramatically, the mouse cursor switched from 60fps to 1fps randomly.

Kwin EGLStreams is not ready.

(Update: I upgraded to the freshest Fedora and tried to enable Wayland again. It finally worked! After 3 years of trying to make it happen! It is not perfect (the screen flickers occasionally, input latency is slightly increased), but it is at least usable. The window movement, animations, etc, everything is buttery smooth, as it should be. So, my rants are over, Wayland FTW.)

Actually, I wouldn't bother if Nvidia would go out of business, as I find it disgusting how they ignore consumer wishes since ages. I mean, in the past I have bought GPU hardware from Nvidia, AMD, and Intel but for a Linux desktop, the experience has been so much better when not using Nvidia, that I try to avoid their products nowadays.

This won't help current owners of Nvidia hardware (as myself), but you shouldn't blame RedHat for issues that obviously have been caused by Nvidia.

Unfortunately, there is only one CUDA, and OpenCL is not an alternative.

Yes, I know :-/

I am not sure what the status of AMD ROCm is, but I guess it will take a while until it becomes a viable alternative.

I'd be happy to say "hurrah!" and close the page, but it is also not an alternative :(

Here is the FFT example in ROCm: https://github.com/ROCm-Developer-Tools/HCC-Example-Applicat...

Compared to FFT in CUDA: https://github.com/drufat/cuda-examples/blob/master/cuda/fft...

1163 vs 216 lines of code! CUDA is really straightforward; ROCm is still low-level and incomprehensible (unless you are a C++ expert who happens to love low-level details).

They don’t ignore consumer wishes. They have the best GPGPU toolchain around. That’s why I only use NVIDIA. Nobody supports developers for High-Performance GPU computing like they do. They also support PC gamers very well.

> They have the best GPGPU toolchain around.

Probably depends on the use-case: I mean, weren't the cryptocurrency people using AMD before they switched to custom FPGAs? But for sure, for the TensorFlow users, there doesn't seem to be a viable alternative to NVIDIA and there must be a reason why CUDA is more popular than OpenCL.

What I mean with ignoring consumer wishes is that they neither offer an open-source driver like AMD and Intel do and they do not even support others in doing so, like the Nouveau people. Subsequently, as a Linux user with NVIDIA hardware, you are either stuck with the buggy binary driver or the feature-incomplete Nouveau driver while AMD and Intel users have relatively stable and feature complete open-source drivers.

Why do you think Linus Torvalds likes Nvidia so much?

https://duckduckgo.com/?q=Linus+Torvalds+Nvidia

Does that tool chain require a desktop environment?

For as long as X is considered to be a reasonable alternative to Wayland, NVidia is unlikely to suffer as a result of its drivers being unable to support Wayland. If Wayland becomes the only practicable option for graphics in Linux, then NVidia is going to be forced to support it to keep their Linux market share.

But that is the point of the post -- X maintainers will abandon X and it will not be considered "a reasonable alternative" anymore. Nvidia supports Wayland just fine with their Eglstreams API, but some WM/drivers developers make a principled stand of not using it.

Maintaining a compositor isn't an easy challenge. GNOME has the resources, plus nvidia help, in supporting EGLStreams so it can.

NVIDIAs drivers do support wayland, they just don't support a separate (not part of Wayland) protocol that the in-kernel mesa drivers support.

Whilst it's currently disabled by default, GNOME on Wayland with NVIDIA does run, just badly.

Reply on news.ycombinator.com