Mesa 21.2 Lands NVIDIA's Code For Handling Alternate GBM Backends

[shmerl]

What interfaces? How are they resulting in poor integration with any Linux desktop? Please show examples.

Tons, from PRIME, vsync and what not to dma buf and the like. Blob will never be integrated because it's a blob. Good thing kernel developers don't tolerate this garbage. And Nvidia dances around it won't really fix their issues.

I assume if Linux users here need to even ask such questions today they are new to Linux or came from Windows recently, so such questions can arise. Windows users don't see any problems with the blob and don't know that isn't how drivers for Linux should be made, so that's understandable.

[sarmad]

Nvidia are trying to pretend they are still relevant for the Linux desktop. It's still good though for those who are stuck with their hardware.

Of course they are still relevant. The options for getting a laptop with a dGPU is currently pretty much limited to only nVidia. There is some random laptops here and there with AMD dGPU, but you'll find something around 1 choice with AMD for every 100 laptops with dGPU, the rest are all nVidia. Even Linux laptop vendors are still exclusively selling nVidia. Yes, that is 100% nVidia from ALL Linux hardware vendors; not even a single laptop from any of them is available with AMD.

This hardware situation may change at some point, but so can nVidia's driver situation.

[Myownfriend]

Still a sizable amount, but a minor percent of total desktop Linux users.

Still a sizable amount? If the Linux marketshare grows at all, then the vast majority of those people will be coming from Windows. We also don't know how many people went back to Windows because they refused to buy a several hundred dollar graphics card or a new a whole new laptop so that they can move to a free OS.

The only place where Nvidia blob will remain relevant is compute / AI because of CUDA lock-in. That will take a while to break, because of how much control they have now over that market.

Which applies to desktop users as well. Just because someone's running compute work on a GPU, it doesn't mean that they're server users.

[shmerl]

Of course they are still relevant. The options for getting a laptop with a dGPU is currently pretty much limited to only nVidia. There is some random laptops here and there with AMD dGPU, but you'll find something around 1 choice with AMD for every 100 laptops with dGPU, the rest are all nVidia.

Hardware trends in laptops are laggy in general to pick recent developments (Intel and Nvidia indeed had a lot of influence on that segment), but it's getting there. I don't see Nvidia having any place on Linux desktop once such kind of issues catch up.

Even with CPUs, you could barely get a good laptop with Ryzens not so long ago. Now more and more vendors make them.

[shmerl]

If the Linux marketshare grows at all, then the vast majority of those people will be coming from Windows.

And I already explained how it will work.

Longer term Linux users: AMD.
Newcomers from Windows with Nvidia: AMD on next hardware upgrade cycle.

So Nvidia will linger becasue of the newcomers, yes. But overall it's pretty much a goner for Linux users besides for the CUDA segment.

[Myownfriend]

Tons, from PRIME, vsync and what not to dma buf and the like.

Nvidia's driver supports all of those. DMA-buf was added in the most recent beta driver. Those are all supported on AMD's proprietary driver.

Blob will never be integrated because it's a blob. Good thing kernel developers don't tolerate this garbage. And Nvidia dances around it won't really fix their issues.

You're not providing any real detail here. What kind of integration are you talking about?

I assume if Linux users here need to even ask such questions today they are new to Linux or came from Windows recently, so such questions can arise. Windows users don't see any problems with the blob and don't know that isn't how drivers for Linux should be made, so that's understandable.

You can't seem to explain it to them either. I would prefer that Nvidia maintained an open source driver, but I don't go making shit up. I think you're a very passionate about this topic but don't really have an knowledge of what you're talking about. At least that's the impression I'm getting based on the pure vagueness of your posts.

Also, are you aware that AMD proprietary driver has OpenCL features that the open source one doesn't?

[oiaohm]

First off, it's not AMD + Mesa vs Nvidia + blob.

It's AMD + Mesa vs AMD blob vs Nvidia blob.

After Nvidia drops a driver that support GBM, it would be AMD + Mesa vs AMD blob vs Nvidia blob + Mesa.

This is wrong. AMD closed source driver uses the open source amdgpu or open source Radeon driver for the kernel mode. AMD used GBM interface from kernel directly. Fun point AMD you can use their closed driver(userspace) and the open source mesa solution at the same time due to the low level interface to the GPU being the same driver. Yes AMD allows one program running using the mesa stack and one program running using the closed source user space stack of AMD outputing on the same screen. Nvidia design you cannot be mixing the closed source and the open source userspaces with each other.

[shmerl]

Nvidia's driver supports all of those. DMA-buf was added in the most recent beta driver. Those are all supported on AMD's proprietary driver.

In some half assed and not functional manner. Once something like Firefox will work with the blob for video acceleration, then you could say it got better. But even if they do mange to use duct tape and voodoo dancing to work around proper usage of kernel interfaces, consider how long it took them to add that, years if not decades. Such kind of support pace is not something any user should appreciate.

[Myownfriend]

Longer term Linux users: AMD.

Or Intel or Nvidia. Again, even the info your provided shows that Linux gamers still use Nvidia cards. The trend doesn't matter. You can't just assume that allow of those people only just switched from Windows. If someone is fine using X.org and they just game, then they're still getting great performance and access to those proprietary technologies like DLSS and Cuda.

Newcomers from Windows with Nvidia: AMD on next hardware upgrade cycle.

Not necessarily. It completely depends on how useful those proprietary technologies are to people. If someone is using Optix or Cuda in Blender then they seem to be getting better performance right now. For their use case where work can be very well parallelized then it might make more sense for them to keep their current GPU during their next upgrade and just add another but that only works if they stick with the same brand.

So Nvidia will linger becasue of the newcomers, yes. But overall it's pretty much a goner for Linux users besides for the CUDA segment.

Again, that's stupid. "Pretty much a goner"? The hell are you talking about. Just hold the fanboy shit back a little bit. You'll be doing yourself a favor.

[shmerl]

Again, that's stupid. "Pretty much a goner"? The hell are you talking about

The trend.