Hacker News Reader

Comment by izacus | Hacker News Reader

The thing is - nVidia uses shared source for all of their drivers. This is why you always have the newest GPU feature support and performance on their Linux driver - it's the same codebase as it is for Windows and was for macOS.

Linux community on other hand wants nVidia to restructure their driver just to fit Linux model and essentially lose the source parity or publish the source completely. It's not an invalid request, because nVidias APIs don't really fit into Linux kernel design and userspace APIs. It's also not very likely to happen, especially due to poisonous behaviour of kernel maintainers towards nVidia, unwilligness to compromise and nVidias refusal to add more resources to refactor the driver.

View on HN

> It's also not very likely to happen, especially due to poisonous behaviour of kernel maintainers towards nVidia, unwilligness to compromise

Please, there is nothing to compromise about.

Nvidia doesn't want to work with the kernel community. They don't want to document how their driver works. They don't want people to look at how their driver works. They don't even want people to develop alternative drivers for their cards.

Nvidia earned 100% of the bad blood existing between them and the kernel community. They are even worse to work with than Qualcomm. Should give you an idea of how bad they are.

Do you think Windows or MacOS make compromise on the way they code their kernel to accomodate nVidia ?

No, it's just a matter of economics.

Yes, Windows has a stable driver API you can build upon.

Linux does not, because they want all drivers to be inlined (included in the kernel) and open source. nVidia does not want to release their trade secrets, so they don't want to open source their drivers.

(No idea about macOS)

Of course, AMD managed, so there’s no technical reason that nVidia cannot.

I didn't say there's a technical reason, I said they want to keep their source code secret, which to me is a reasonable argument.

Well then sway maintainer's reason to explicitly deny support for proprietate nvidia driver is a reasonable argument too.

Definitely, it's his software and he can do whatever he wants, and I'm not criticising him, I'm criticising Linux for not offering a stable driver API. As far as I know they don't even want to allow shims like the one Android Treble introduced. It is not a technical decision, it's a political one.

Refusing to open source their drivers is not a technical decision, it’s a political one.

There is a technical rationale that should be read to understand kernel dev's PoV : https://github.com/torvalds/linux/blob/master/Documentation/...

Except that AMD doesn't have feature parity with Windows driver on open source version and has all the Nvidia driver headaches on closed source version.

I'm not sure what AMD really managed here.

Because it is NOT the same driver. AMD, just like nVidia, want to keep their own secret sauce secret, because that's where the money is.

Is so much of their profit bound up in their software?

I thought these were hardware companies.

It doesn’t have “all” the headaches on the closed source version, afaik - it still implements the standard APIs and thus works with systems that were designed to work with every other driver from AMD to Mali, unlike the nVidia proprietary driver.

AMD's open source driver is fully usable for normal usage (including gaming), and in general more performant than the closed source version.

https://www.phoronix.com/scan.php?page=article&item=rx590-li...

Still behind the proprietary Windows driver or Nvidia's proprietary Linux driver.

Can you share some source of your statement?

They are not behind their windows driver on OpenGL. And neither for Vulkan, I think. AMDVLK scores a few wins here and there, but RADV is generally more stable, performant, and compatible. Plus being open-source, it is easily extended, like valve is doing with their AC0 experiment [1].

Last I heard, people were installing Linux to use the AMD OpenGL drivers to play emulators (it might have been Cemu, actually ran trough wine).

[1]: XDC2019 slide deck https://xdc2019.x.org/event/5/contributions/334/attachments/...

MacOS user: what's a driver ?

Seriously though, one of the pleasurable things about using macOS is that you don't have to worry about hardware compatibility. Because there is none, outside the small list of parts officially supported by Apple itself. There are some NVidia cards on that list, yes.

Linux has a fairly stable driver API, it just doesn't have an ABI.

The GPU API doesn't seem particularly stable to me.

> The thing is - nVidia uses shared source for all of their drivers. This is why you always have the newest GPU feature support and performance on their Linux driver - it's the same codebase as it is for Windows and was for macOS.

Sure this is almost true, but not quite - there is obviously Linux specific code to support X11 and VDPAU that isnt in the Windows driver, for example. Of course, NVIDIA can decide to not support open drivers, or not support Wayland in their closed drivers, and as a consumer I can choose not to buy their products.

You can have an open source driver without mainlining it. This was what the whole discussion between the Linux devs and AMD was over the AMD HAL. If you want to maintain it yourself, you can do it however you want.

Linux has modules, and had very good support for out of tree modules.

It's easier to switch to Radeon, and ROCm instead of CUDA.

The same is true of AMD, they share most of their display configuration code (in the kernel) and userspace code between a set of drivers on Windows, Linux, and macOS; they just also publish documentation and redistributable firmware, and maintain a second very-high-quality driver in Mesa, and due to the availability of documentation and that Mesa GL driver, there is a second Vulkan driver (RADV) which is also excellent.

Reply on news.ycombinator.com