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

[aufkrawall]

vsync

Not a fan of the Nvidia Linux driver, but they managed proper fifo vsync for their Vulkan driver. Unlike AMD, who simply don't care for fifo vsync being borked in amdvlk since it exists on both Windows and Linux (yes, thankfully there's radv for the latter). You can really only buy AMD for Linux only (if you can live without OpenCL and lagging behind for raytracing), they seem to somewhat disdain their Windows customers.

[Myownfriend]

The trend.

It's not a trend to zero. It's a trend to no longer being the majority specifically amongst 10,507 Linux gamers from one specific site. Nvidia and AMD's place on that chart represents 1030 and 838 people respectively. That "trend" that you're looking at is the result of 264 people no longer using an Nvidia card. We can't even say that they switched to AMD because AMDs rise only represents 130 people and there's no rise in Intel's share to make up for those numbers. For all we know, the missing 134 people just switched back to Windows.

Fact is, these numbers are two small of sample size to mean anything and they don't even claim to be definitive. They even say "As with any survey, this won't be 100% accurate and should be taken with a pinch of salt." but in your world that means they're done for.

Maybe you should just stop with the fanboy stuff and realize that you don't know what you're talking about and don't know how to gather and understand data.

[shmerl]

It's not a trend to zero

I don't see why it wouldn't be a trend to low percentage. As above, Nvidia offers inferior support (features take decades to add), poor integration with the desktop, poor performance of simple desktop interactions, no advantage of gaming performance (with last generation of GPUs) and so on. So only negatives, no positives. Negative trend is expected in such case. Influx of former Windows users will keep it around, yes. But won't reverse the trend.

You can be in denial about the problems, but Nvidia shows no interest in solving the root of them - refusal to upstream their driver. So as above, good riddance, Linux users don't need them. And you can stop whitewashing Nvidia using the blob as something that's good for Linux in general.

[fluke]

Now unlucky notebook users need Wayland+PRIME offloading working.

It does, with 470 driver. I've been gaming in wayland with nvidia since it released.

[Zan Lynx]

1. Nvidia kernel driver does not use GPL symbols. If it used, kernel wouldn't load it. Kernel has protection against this.
2. It's your distro's fault. Arch based distros are shipped with precompiled module for default kernel.

Then Arch and other distros that do this are committing a GPL violation by distributing binary software that does not contain or offer the source. Linux kernel code is GPL v2, not LGPL. It does not contain a linking exception. So any code linked to it must also provide source.

They're just hoping no one involved with Linux is bothered enough to sue them.

[abu_shawarib]

It would be great if the stars align and Ubuntu 21.10 had all the requisites to run Wayland with Nvidia's drivers by the time it comes out.

[evasb]

It would be great if the stars align and Ubuntu 21.10 had all the requisites to run Wayland with Nvidia's drivers by the time it comes out.

Yeah, and it'll will be hopefully stabilized on Ubuntu 22.04 LTS that is the most important release.

[d3coder]

Then Arch and other distros that do this are committing a GPL violation by distributing binary software that does not contain or offer the source. Linux kernel code is GPL v2, not LGPL. It does not contain a linking exception. So any code linked to it must also provide source.

They're just hoping no one involved with Linux is bothered enough to sue them.

You don't know what you talk about, don't you? Kernel has GPL and non-GPL exported symbols. Nvidia kernel module does not use GPL symbols. Kernel module license is proprietary, but source code is available.

[Myownfriend]

I don't see why it wouldn't be a trend to low percentage.

Again, the numbers you're using are too small of a sample size to mean anything. When you're talking about a sample size of less than 1/5th of a 10,500 people, you're gonna get wide swings in the numbers when a small amount of people change their hardware. Those numbers aren't indicative of Linux or Linux desktop as a whole.

As above, Nvidia offers inferior support (features take decades to add),

But again, we're having this discussion after they started supporting dma-buf and just as they got a PR merged into mesa to allow Mesa to run on stripped down version of their driver. What's the long list of features that they wouldn't be supporting after that? Considering you weren't aware that Nvidia already supports v-sync, Prime, and DMA-buf, I don't think you'd be able to do that.

poor integration with the desktop

Which goes away when they release the driver that supports Mesa. We said this already.

poor performance of simple desktop interactions

Like what?

no advantage of gaming performance (with last generation of GPUs)

Not a disadvantage either. I've seen the same Phoronix benchmarks that you have and they're competitive. They trade blows and neither really has a huge edge over the other though Strange Brigade didn't seem to run on RADV in this test.

And again, people do more with their computers than playing games. The Linux desktop community isn't just made up of gamers.

So only negatives, no positives.

Again, Cuda and Optix.

Negative trend is expected in such case. Influx of former Windows users will keep it around, yes. But won't reverse the trend.

Your view of the world is entirely too simple. I'm going repeat yet another thing that I said before. The numbers you shared, as ultimately meaningless as they are, seem to infer that half of Nvidia's drop in that graph came from people keeping their card and switching back to Windows. Either that or fewer people reported on their hardware.

There's no information that we have that all or even most people coming from Windows are upgrading their computers to get rid of their Nvidia cards due to any weirdness they experience from Nvidia's Linux drivers. We don't even know if the majority of Linux users coming from Windows are even former Windows users. They may just be dual-booting and you can't assume they're all gonna ditch Nvidia just for their Linux partition. If they're gamers, they might decide to do all their gaming on Windows where Nvidia allows them to play games with much better ray-tracing performance than AMD. I don't know any of that and you don't know any of that. You're not making sensible projections you're just stating an outcome that you're hoping for.

You can be in denial about the problems, but Nvidia shows no interest in solving the root of them - refusal to upstream their driver. So as above, good riddance, Linux users don't need them. And you can stop whitewashing Nvidia using the blob as something that's good for Linux in general.

How the fuck did you read my posts and gather that that's what I was saying? Can you read?

I was criticizing you for being fanboyish and making insane claims and counterbalancing your fantasy with reality. Saying that Nvidia is done for in the Linux community when they majority of Linux users still use their cards is insane and unrealistic. You can criticize Nvidia's lack of open-source driver, but the fact that the majority of Linux users, people with Nvidia cards, will be able to use Mesa is a great thing for Linux as a whole. "The blob" doesn't negate that. I never said that the binary drivers are better for Linux than open-source drivers, you made that up yourself. I was just saying, and read this slowly so you take it in, that the news that spawned this thread is a good thing for Linux.

[shmerl]

Shrug. Nvidia is going to be irrelevant for the Linux desktop, whether you ignore the trend or not. And it's a good thing. The less developers need to deal with idiosyncrasies of the blob, the better.

CUDA has nothing to do with desktop use case. It's a lock-in garbage that Nvida uses to control AI market for the most part. Some like Blender that were using CUDA are switching to Vulkan and other portable ways of GPU compute. CUDA is a dead end.