Hacker News Reader

Comment by stingraycharles | Hacker News Reader

And why would this not be acceptable? Shouldn’t we be happy that nvidia is adapting this driver?

No if it goes against Linux standards. That's just accepting the lowest common denominator quality driver. If AMD an do a better job with AMDGPU, so can their much larger competitor.

That's the sort of talk companies with market share of Microsoft or ecosystem of Apple can afford.

I am a Linux desktop user myself but:

a) The AMD drivers were unusable crap for a very long time, things have improved only thanks to the open source driver (that ATI/AMD did not write, they only contributed specs). Basically 3D and OpenGL in Linux was for years synonymous with "buy Nvidia, it works".

b) I don't understand this "holier than thou" attitudes when Linux desktop isn't even a blip on the radar and pretty much contributes zero to the bottom line of these companies. However, it tends to have the most vocal, noisy and abusive users ...

Instead of being glad that Nvidia actually supports Linux (and FreeBSD and few other free OSes) for ~20 years now (before ATI even had a Linux driver at all!), we heap abuse on them because their drivers aren't to our taste? Would it be better if they cut their losses and stop providing this support? Especially for stuff that doesn't have any alternatives, such as CUDA? And no, Intel integrated graphics really isn't an option for high performance graphics and ATI's open source drivers are quite a bit slower. Of course, if all you need is requiring accelerated desktop so that you can have nice transparency/compositing and accelerated video playing, then you won't see this.

I don't get where did you get the idea about the "lowest common denominator quality driver" from Nvidia. Sure there are bugs and issues (e.g. with suspend) but in my experience the Nvidia graphic drivers were the least troublesome when it came to application support, OpenGL spec + extension support and various bugs. And I have been using 3D drivers in Linux essentially from the very beginning, when the top end hardware meant a 3DFX card or later a Riva TNT. My diploma thesis ran on a 3DFX back in the day ...

That said, it doesn't mean we shouldn't complain when something doesn't work but let's not forget the size of the market. A dose of reality helps - jumping up and down and yelling really really loud to look important doesn't make one so.

> that ATI/AMD did not write, they only contributed specs

This is a lie.

All of the main contributors for the amdgpu drivers are AMD employees. The files in drivers/gpu/drm/amd/amdgpu all have AMD copyrights in the headers.

I'll grant you, 18 years ago, Nvidia's drivers were superior by virtue of existing at all. That's a long time ago. The competition has improved, and the Nvidia experience has steadily gotten worse. I have a desktop machine at work that locks up hard with the proprietary driver, and another one where X randomly crashes if you use nouveau. Any goodwill they had with me for providing proprietary drivers on Linux over a decade ago has long since evaporated. Their hardware may benchmark well, but in my experience it's unstable and I suspect they're papering over a poor design with opaque drivers.

>Would it be better if they cut their losses and stop providing this support? Especially for stuff that doesn't have any alternatives, such as CUDA?

They'd lose more if they did that. Do you think the whole world of HPC/ML/scientific computing will switch to windows to use nvidia cards ? The answer to that is no. They'll just move over to AMD. ROCM is coming along fairly well. Or something else will pop up. Don't get too attached to blobs.

They could drop graphics support on Linux with no significant losses to their hpc/ml market.

Yes, that is somewhat true I think. They could just expose an accelerator device without any graphics capabilities. I was replying to the parent's point that CUDA is not hostage. They lose more if they drop CUDA on linux.

> And why would this not be acceptable?

Quoting the article:

> Actually, Nvidia doesn’t support us. There are three standard APIs which are implemented by all graphics drivers in the Linux kernel: DRM (display resource management), KMS (kernel mode setting), and GBM (generic buffer management). All three are necessary for most Wayland compositors. Only the first two are implemented by the Nvidia proprietary driver. In order to support Nvidia, Wayland compositors need to add code resembling this:

   if (nvidia proprietary driver) {
        /* several thousand lines of code */
   } else {
        /* several thousand lines of code */
   }

And it goes on, at length, describing the problems this causes. It's not like it's a mystery.

Reply on news.ycombinator.com