> May I suggest you buy an AMD GPU next time around

This doesn't help anyone. I bought an expensive Nvidia GPU 6 years ago, when AMD support on Linux was really bad. What do I do, throw out a perfectly good powerful graphics card and spend another $500 to satisfy the needs of some newfangled software which doesn't solve any of my problems?

I've had 6 years to experience the horrors of Nvidia, on multiple machines. I'm never going to buy one of their cards again unless there's a massive change in their leadership. But telling people 'tough luck; basic functionality that your computers have been capable of doing for decades will no longer work because you bought the wrong brand' is simply unacceptable.

Your 6 year old GPU probably works fine with the nouveau driver, which is supported by wlroots.

> What do I do, throw out a perfectly good powerful graphics card and spend another $500 to satisfy the needs of some newfangled software which doesn't solve any of my problems?

If you want to replace a card which was expensive six years ago, you can spend a hell of a lot less than five hundred dollars to do so. Even a brand-new RX 570, which will outperform a GTX 770 (the highest performance GPU for $500USD or less in 2013), should only cost you about $130USD today. If you're okay with pre-owned cards, you can get better than that for the same money.

> This doesn't help anyone.

It helps people who are in the market for a new GPU, who want it to work well with Linux, especially the cutting-edge stuff.

Even in 2013 though, I think I was able to run Wayland on a discrete AMD GPU (Terascale 2 or 3?) with at least OpenGL 3.3 available; and it worked OK. I don't think AMD got that working more recently than 2013.

I can understand not wanting AMD's cards on the top end as ca. 2013 (since they were not competitive, IIRC), but now they do have competitive offerings, and their support for desktop Linux is incomparably better, which is why I suggest people go with them.

I don't get why I'm getting flak here.

Not OP, but you are missing this points:

1 the expensive(this is relative) card works fine, do I throw it or donate it without any good reason(switching to wayland could wait IMO until this card is too obsolete)

2 Buying a new GPU is a big investment for some people,those money could be used to replace an old low quality display or other stuff.

> Not OP, but you are missing this points:

No, I'm not missing these points. It's just that nobody except NVIDIA can do anything about that, so I'm suggesting things you could do. If you can't get an AMD or Intel or Qualcomm (unofficial) or Broadcom (or soon ARM) GPU, and you won't (or can't, again because of NVIDIA being antisocial) run Nouveau instead of the NVIDIA proprietary drivers, and you're not willing to port wlroots to EGLStreams, then you simply will not be running wlroots-based compositors on your machine. That's just how it is.

If you bought a 2019 MBP hoping to run Linux with full support for all the hardware, you would be making a mistake; just because I say that it's a mistake, doesn't mean I think you can afford to replace all your computers again once you've made that mistake.

I don't get these replies. I go to specific effort to couch everything I say, I say "next time around", I say "if possible"; and everyone reads it as though I'm somehow out of touch with the fact that not everyone can spend hundreds of dollars on GPUs every year. I know, I'm just saying that steering clear of NVIDIA in the future is an option that will give you a good experience with standard Wayland compositors.

OK, there are a group of people where at the time we bought NVIDIA it was not a mistake, it was the better card for Linux Desktop, at that time AMD open source GPU was bad and the Catalyst proprietary one was terrible in performance, Steam games were only supporting NVIDIA.

Sure when I will need to buy a new card I will again have to check what is the best one for Linux and then I will probably get an AMD card again( I had AMD before NVIDIA on desktop and Laptop and back then it was a big disappointment, including the fact the AMD Catalyst drop support for my 1 year old cheap Laptop)

> AMD Catalyst drop support for my 1 year old cheap Laptop

My advice is: Unless you're running a qualified application on a workstation GPU, do not use Catalyst (or whatever they're calling it this year). It is almost always more finnicky. Mesa has the best performance, and the most meaningful features for OpenGL, Vulkan, and D3D9 for any any GCN 1 or later (and probably r600 or later, but I don't have personal experience with that) GPU. AMD staffs people to maintain Mesa, including for GPUs they do not support their proprietary driver on; and if you're running a fresh/rolling-release distro, you'll tend to have a great out-of-the-box experience with Mesa.

Thanks for the response, my story about Catalyst was very old. I was trying to explain to some new Linux people that AMD was terrible a few year back. So I bought a new laptop and in 1 year Catalyst dropped support for the GPU in that laptop forcing me to only run distributions with old enough Xorg because the open source drivers were bad at that time.

I know that now AMD open source is better and my new computer will probably run an AMD CPU and an AMD GPU but until my current PC gets outdated I will have to continue use my NVIDIA 970 which is still a decent card for cashual gaming

I've never experienced those supposed horrors across multiple operating systems (FreeBSD, Linux, macOS and Windows 7/8/10), the only time I owned an AMD graphics card it was a total disaster, among other things I remember are really bad Linux drivers (proprietary or otherwise). I never had any issues with my current GTX 1080, 4xGTX 2080 workstations or the 2xV100 nodes I have access to.

> I've never experienced those supposed horrors across multiple operating systems (FreeBSD, Linux, macOS and Windows 7/8/10)

You've completely missed the point. The horrors are in compatibility. NVIDIA chooses to make it difficult for their customers to use their hardware with new applications, and they hold back the whole ecosystem. In addition to the topic at hand (Wayland), their release model holds back Linux kernel releases when their driver becomes incompatible with the upstream kernel because it is not maintained there.

If you just want to run whatever boring applications NVIDIA chooses to allow you to run, congratulations! Go enjoy your "experience", and play it "the way it's meant to be played"! :- )

> among other things I remember are really bad Linux drivers (proprietary or otherwise)

You're living about seven years in the past.

> By all means, if you want to do NVIDIA's job, be my guest.

If I wrote an X compositor that supported NVIDIA's proprietary interface, it would work with pretty much every window manager that's ever been written.

If I patch wlroots to support NVIDIA, I'll still only be able to use window managers based on wlroots; I won't be able to use ones that spun the Wheel of Incompatible Wayland Compositor Libraries and landed on Weston, swc, Orbment, or whatever GNOME uses instead.

This is why it was so infuriating when NVIDIA refused to implement the stanadrd interface: they made the ecosystem "fragmented" for their users, and the PR cost of that bears on the compositor writers, rather than NVIDIA.

Having different compositors at the level they're at allows a lot more interesting innovations to be made, and aside from NVIDIA's tomfoolery, it has not caused significant inconsistency. Everyone uses libinput, almost everyone uses colord, and everyone implements GBM.

Wayland is substantially different from X, and the ways in which it is different are good. NVIDIA did the one thing we needed people not to do.

Have you considered that maybe it is the other way around: The standardised solution is inferior and NVIDIAs superior? Given the fact NVIDIA engineers probably know a lot more about modern graphics stacks than some Redhat employee this seems likely.

Some NVIDIA engineers probably know a lot about graphics stacks; charitably I'd say that those ones aren't the ones tasked with insisting on EGLStreams, which lacks a considerable number of crucial features.

To be honest though, NVIDIA engineers do not spend any time engineering graphics stacks at all; they write graphics drivers for stacks that are already largely specified. They're really good at shipping hundreds of megabytes of dirty hack code to patch popular shaders from ISVs to run slightly faster on their hardware, and probably great at writing shader compilers and threaded GL drivers.

Instead of repeating arguments made elsewhere maybe this post sheds some light on why EGLStreams are a superior solution: https://devtalk.nvidia.com/default/topic/925605/linux/nvidia...

Note the first comment there shows that their EGLStreams implementation shipped with vsync broken. EGLStreams seems attractive to NVIDIA because it was easy to implement (incorrectly). The features they talk about enabling on Tegra are not impossible with GBM, it just may involve adding a feature.

So the argument it is a bad implementation because it doesn’t work with the Wayland model... maybe that model is broken then.