Open-Source NVIDIA Driver Moving To NVK + Zink For OpenGL On Newer GPUs

[user1]

AMDVLK.

But AMDVLK is not a Mesa driver. And I'm not sure how much is AMDVLK helpful as a source of early new hardware documentation. Plus, RADV reuses RadeonSi's winsys component and there is also some common code shared between RADV and RadeonSi, so there's also that.

And have you checked out the (AMD) driver quality for older D3D versions on Windows? It is a real shit show.

What does this have to do with the Linux drivers? My impression (from personal experience) is that the driver quality of both RadeonSi and RADV is better than AMDVLK and all the other non-Mesa AMD drivers.

And nobody said that we should remove radeonsi from Mesa right now. It is just a matter of time until maintainers will stop caring. No matter if they work at Valve, RedHat, AMD or are indepdendant.

It is not that somebody asks to remove it. It will just bitrot over time.

I understand nobody said that, but you all don't seem to understand that if this will theoretically happen (no matter when), it will hurt RADV, because it's not an official, AMD supported driver. Of course, it won't be an issue if at some point AMD will start supporting RADV as an official AMD driver and give them access to early documentation.

[ehansin]

I read about this on the Collabora blog in the somewhat recent past, that they were thinking of doing this. I don't know enough to have a strong opinion, but if Zink is "good enough", then as mentioned you can can focus development efforts on the Vulkan driver. This could create an effect where more and more vendors create Vulkan drivers (thinking ARM related graphics cores, but not limited to that) where you get OpenGL for "free". We do know getting open-source graphics drivers for those cores is not always easy to come by.

And this I do not know for sure, but I thought I read once that different OpenGL implementations where not always able to run all software due to quirks, etc. I believe I read there was flight simulator software that bundled their own OpenGL with it to alleviate this. If running Zink on top of Vulkan kind of became "the thing", then there becomes some consistency there. But may be a little off on some of this.

[-MacNuke-]

I understand nobody said that, but you all don't seem to understand that if this will theoretically happen

In the end you can argue about theoretical things all day long. It will happen eventually. Nobody will keep outdated functionality alive for all eternity. It is just a matter of time until near to nobody uses it anymore because the majority of applications use Vulkan. People will find bugs and unimplemented things on newer hardware for OpenGL. And then people will have to wait until someone implements/fixes that or ... just use Zink.

It happend on Windows, it happend on macOS and it will happen on Linux too.

[user1]

In the end you can argue about theoretical things all day long.

And you can continue ignoring reality all day long and praise a future in which the very driver on top of which Zink runs, will lose a crucial documentation source from the hardware manufacturer itself.

[schmidtbag]

This is excellent - I agree there is no good reason for such limited resources to spend the time and effort making an Nvidia-specific OpenGL driver for this newer hardware. I think Nouveau can still exist as it serves a purpose for older generations, but modern Nvidia GPUs are plenty powerful enough to not be held back by Zink's overhead.

As for AMD, I have to agree with user1. RadeonSI is well optimized, it's largely maintained by AMD themselves, and it works for hardware old enough where Zink's overhead might actually be detrimental. While I'm sure a lot of the device-specific optimizations for Zink come down to the Vulkan layer, I don't see why AMD has to reinvent the wheel in what may result in a worse overall experience. Nvidia isn't gonna do anything worthwhile for the benefit of Mesa in any regard, so when you're left to 3rd parties and volunteers, Zink is the only sensible way forward for their hardware.
I think if AMD does a complete overhaul of their architecture where there may be logistical or compatibility issues, then it would make sense for them to swap to Zink.

[user1]

And just to be clear, when there is no mature and performant OpenGL driver, like in the case of Noveau and Asahi, of course I think it makes much more sense to use Zink instead.

[-MacNuke-]

And you can continue ignoring reality all day long and praise a future in which the very driver on top of which Zink runs, will lose a crucial documentation source from the hardware manufacturer itself.

It has nothing to do with "ignoring". It is just a matter of time until AMD will reduce development resources on the OpenGL driver. And if RADV is doomed because of that (which I think is not the case) than we have to use AMDVLK...

[schmidtbag]

It has nothing to do with "ignoring". It is just a matter of time until AMD will reduce development resources on the OpenGL driver. And if RADV is doomed because of that (which I think is not the case) than we have to use AMDVLK...

OpenGL is pretty much as mature as it's ever going to get. RadeonSI is already 100% compliant (at least if you ignore some of these) so AMD doesn't really have to do anything else other than optimizations (which at this point is more down to firmware/hardware than Mesa) or compensate for minor architectural tweaks. As I said before, if there's a big enough change to the architecture, then they ought to switch to Zink.
So - AMD is already in a position to devote minimal resources to OpenGL development.

[FireBurn]

And just to be clear, when there is no mature and performant OpenGL driver, like in the case of Noveau and Asahi, of course I think it makes much more sense to use Zink instead.

Asahi has an OpenGL 4.6, OpenGL ES 3.2 fully compliant driver...

[user1]

Asahi has an OpenGL 4.6, OpenGL ES 3.2 fully compliant driver...

But is it mature and performant? Since Asahi is a relatively young project, it's probably not. Supporting the latest OpenGL versions doesn't also mean being mature and performant..

And I'm actually surprised they haven't just went directly with Zink, because this is exactly the case where using Zink makes the most sense (new projects like Asahi). It takes many years until a native OpenGL driver becomes mature and performant, so I don't understand why they haven't just went with Zink and focused on their Vulkan driver instead.