NVIDIA Developer Opens Feature Pull Request For Open-Source NVK Driver

[Sevard]

Their open driver was never intended to be upstreamed. It exists as a reference implementation for NVIDIA user space components.

Maybe in your universe. In ours the truth is a bit different. In this blog post https://developer.nvidia.com/blog/nv...ernel-modules/ NVIDIA alone claims that

Upstream approach


NVIDIA GPU drivers have been designed over the years to share code across operating systems, GPUs and Jetson SOCs so that we can provide a consistent experience across all our supported platforms. The current codebase does not conform to the Linux kernel design conventions and is not a candidate for Linux upstream.

There are plans to work on an upstream approach with the Linux kernel community and partners such as Canonical, Red Hat, and SUSE.

In the meantime, published source code serves as a reference to help improve the Nouveau driver. Nouveau can leverage the same firmware used by the NVIDIA driver, exposing many GPU functionalities, such as clock management and thermal management, bringing new features to the in-tree Nouveau driver.

Stay tuned for more developments in future driver releases and collaboration on GitHub.

So they definitely want to upstream it someday, but they couldn't do this back then.

So if you still want to lie then at least try to find a lie that is not easy to verify.

[avis]

Maybe in your universe. In ours the truth is a bit different. In this blog post https://developer.nvidia.com/blog/nv...ernel-modules/ NVIDIA alone claims that

So they definitely want to upstream it someday, but they couldn't do this back then.

So if you still want to lie then at least try to find a lie that is not easy to verify.

"The current codebase does not conform to the Linux kernel design conventions and is not a candidate for Linux upstream."

Read that again. Slowly. And stop with "you still want to lie". NVIDIA never intended to upstream the open driver they'd released.

This is the second time in this thread that I'm being accused of lying despite not doing or implying that in any shape or form. FML! That's disgusting and despicable. I mean it must be a sad, boring and depressing day that if you didn't accuse me of lying, ain't it?

NVK is a new project written from scratch outside NVIDIA, something we're now discussing.

oiaohm

1. The device driver is talking to hardware directly
2. It's running in kernel space
3. It's OS dependent
4. For systems with Secure Boot (the vast majority of x86 and desktop ARM devices released in the past 15 years) it must be digitally signed and recognized by firmware.

The CUDA library is not a driver no matter how you want to paint it.

I don't need your possible insinuations about FUSE (not related to hardware) and WDDM (not related to Linux), thank you very much.

[LockedPotato]

Maybe nvk is valve requirement for a new.generation of a gaming console (not portable)?

If that were happening I could see that being the case. Gamescope does NOT work well with the proprietary Nvidia driver. Lmao.

[LockedPotato]

According to Wikipedia, it's not too far off:

That's just FreeBSD's network stack and Android's SurfaceFlinger display manager. The rest of the OS is wholly original. The Switch actually runs on a microkernel.

[tuxifan]

It's been proven that nouveau works on the Switch, under Nintendos OS.
In fact a lot of homebrews use it for OpenGL rendering: https://github.com/devkitPro/mesa

[Sevard]

Read that again. Slowly. And stop with "you still want to lie". NVIDIA never intended to upstream the open driver they'd released.

And literally few lines later:

There are plans to work on an upstream approach with the Linux kernel community and partners such as Canonical, Red Hat, and SUSE.

So NV alone wrote that they're working on that. And this blog post was written almost two years ago. So yeah, you are liar.

[avis]

And literally few lines later:

So NV alone wrote that they're working on that. And this blog post was written almost two years ago. So yeah, you are liar.

The original open driver was never intended to be unstreamed and the later statement does not negate that. You really have issues with reading comprehension, logic or even being remotely polite. Goodbye Mr Savant.

[Sevard]

The original open driver was never intended to be unstreamed and the later statement does not negate that. You really have issues with reading comprehension, logic or even being remotely polite. Goodbye Mr Savant.

I've never wrote that it's been intended to be upstreamed. I wrote that NV cannot upstream their "open source" kernel driver (that they are working on) without proper user space driver that makes use of their kernel space driver.

[avis]

I've never wrote that it's been intended to be upstreamed. I wrote that NV cannot upstream their "open source" kernel driver (that they are working on) without proper user space driver that makes use of their kernel space driver.

Amazing, you have now admitted I was right all along. Really, some people here can't do anything but hate and insult.

[oiaohm]

The CUDA library is not a driver no matter how you want to paint it.

Might be a library but does that library work with any Nvidia driver the answer is no it does not.

https://docs.nvidia.com/deploy/cuda-compatibility/index.html#use-the-right-compat-package

Like you have a 545.23.06+ driver loaded in your kernel you can only use CUDA 12.3 library nothing else works. Same with 530.30.02+ and CUDA 12.1 that a 1 to 1 lock so nothing else is going to work..
You have 550.54.14+ driver loaded at the stage only CUDA 12.4 library works.

Branches R525, R515, R510, R465, R460, R455, R450, R440, R418, R410, R396, R390 are end of life and are not supported targets for compatibility.

Avis is right to say CUDA is a library but that only part of the picture. Its a driver locked library. If you don't have one of the drivers that Nvidia had decided to long term support you only have 1 version of CUDA library you can use.

The CUDA library has been better than Nvidia Vulkan and Opengl Libraries on the version locking. But its still a driver locked library. You cannot install a CUDA library and presume it going to work you have to check the driver version if it compatible.

avis also CUDA documentation from 2007 calls the user-space library a driver so a person calling it a driver should just be pointed out they are using a old incorrect terminology that Nvidia used that Nvidia only corrected in 2009. The cuda libraries 1 to 1 bindings between the library and the driver and that was true until about 8 years ago.

Being able to use multi CUDA versions with a single Nvidia kernel driver is a fairly new thing. Lot of textbooks used in universities for CUDA still uses the term driver for the userspace library. Avis some errors you have to accept as normal.