[arghwhat]

Your pick. Nvidia are assholes, though, and should not be supported directly at any cost. They need to learn to grow up.

Also, what the hell are you using a couple thousand dollars in consumer nVidia GPUs for on a Linux system?

[Cacti]

Machine learning, where NVIDIA hardware and proprietary drivers is currently non-negotiable. It sucks but it is reality, and refusing to support their drivers in other software because NVIDIA doesn't support FOSS will accomplish absolutely nothing. Except people not using your software.

[arghwhat]

Why are you rendering your desktop through a compute GPU? Connect screens to the Intel HD outputs.

Refusing to support nvidia accomplishes that their shitty approach will die. They will then have to pick between either dropping linux support (which they don't want exactly due to ML and stuff), or having to do shit right.

IIRC, gnome is also dropping eglstream support (nvidia crap), leaving nothing supporting it.

[Filligree]

Machine learning, probably. Nvidia is still the best option for that.

[franga2000]

In that case, you don't really need graphics at all. It's perfectly possible to use a card with good drivers for your displays (even the Intel IGPUs are good enough for that) and do your machine learning with the nvidia GPU and whatever blob they provide.

[Cacti]

I don't have integrated graphics on these machines, and there is no room for additional cards, and if there was, I would either be putting in another NVIDIA card, or RAID, or 10Gb or the many other things that will actually improve our use of these machines. Also, these are company/org machines, I can't just go putting random ass hardware in them because someone wants to use a window manager.

What you're suggesting may be fine for casual desktop users at home, but in many cases it's simply not possible.

[franga2000]

Fair enough - there are edge cases. If you're running Xeons or the new AMDs and don't have an iGPU, you're out of luck. But this is Nvidia's doing, not the developers'. It's Nvidia who is stubbornly refusing to implement standards. I can't blame the devs who don't want to duplicate their effort. This is the price you (and I) pay for using Nvidia GPUs. Hopefully OpenCL (or similar) support catches up soon and Nvidia gets some competition, or we're headed for a dark future.

[arghwhat]

That's a unique case: a non-server xeon workstation.

However, it's up to nvidia to fix. No one intends to support their lazy eglstream suggestion for countless reasons, and only gnome is left, which IIRC intends to drop it too.

ML is a good example of why not putting up with them will work. They need that customer base. If all support is dropped, they'll need to act.

[josefx]

> No one intends to support their lazy eglstream suggestion for countless reasons,

So first Wayland came for the networked UIs because "X forwarding is shit" then they wayland users came for NVIDIA because "we don't like how they do it". I can see myself booting into a Windows 10 desktop in five to ten years because pretty much everything I relied on on Linux has been killed off by then, oh whait that is GNOME 10, it just looks and acts just as user hostile as Windows.

Excuse me while I seat myself into a corner with my trusty X window based software that at least does what I need it to do half the time.

[arghwhat]

NVIDIA being assholes is nothing new. Wayland is just their newest way to try to screw open source development in order to stay as proprietary as possible.

> Excuse me while I seat myself into a corner with my trusty X window based software that at least does what I need it to do half the time.

Enjoy your root escalation exploits and complete lack of any usable security model, then. Oh, and soon, complete obsolescence. I'll enjoy my out-of-the box wayland setup.

Oh, and yes, X forwarding is total utter crap. Wayland will have remote capabilities, but X forwarding was never a good idea. It was okay 20 years ago, but it doesn't make sense at any level today.