NVIDIA's Proposal For A New API Better Than GBM Has Already Made Some Progress

[starshipeleven]

Be realistic, here. What you describe as "the world" consists of drivers for just two significant hardware manufacturers, AMD and Intel.

No it consists of all applications that have some kind of GPU acceleration that would need to add another rendering backend just for NVIDIA. They took years to get to this point with GBM and things are starting to kinda be ready for prime time.

So if NVidia can't copy the approach used by those drivers, and wants to collaborate on a third way - this is *not* a huge campaign to screw over the community. Yes, it creates a bit more work for everyone - but if it helps resolve the situation where one of the three major GPU platforms cannot cope with Wayland, it's worth trying to make it work. Afterall, if you want people to move to Linux as you suggest - surely you don't want to exclude those with NVidia hardware?

one of the three major GPU platforms cannot cope with Wayland

WAT?

Are you seriously saying that NVIDIA cannot cope with wayland?

For real?

And that linux ecosystem should stop and rewrite all their graphics backends wasting like 3 years because NVIDIA cannot "cope with wayland"?

[Delgarde]

Are you seriously saying that NVIDIA cannot cope with wayland?

For real?

And that linux ecosystem should stop and rewrite all their graphics backends wasting like 3 years because NVIDIA cannot "cope with wayland"?[/SIZE]

Well, that's the evidence before us, isn't it? I mean, I agree it's stupid that things have gotten this far in rolling out Wayland, before Nvidia have decided to get off their asses and start figuring out how they're going to support it. Their lack of engagement is why I run AMD hardware.

But that's the way things are looking. Right now, Nvidia have basically rejected the idea of implementing GBM, and I can only assume that this is because they don't think it's practical to do so - either outright impossible given their driver architecture, or as someone else suggested, because getting acceptable performance would be impossible within the constraints of that architecture. And since the Linux driver code is largely share with the Windows driver - something that normally benefits us - they're not going to re-architect their majority driver for the sake of a minority one.

So frankly, yes. Unless something changes, you've summed it up neatly - one of the three major GPU platforms, as you say, cannot cope with Wayland. For real.


So what can change? Well, if Nvidia could implement GBM, they'd have done it by now, so I assume that's not going to change. And Nvidia could throw resources behind Nouveau to make that more usable, but if they've not done that already, I doubt that's going to change any time soon. It's equally clear that nobody else is interested in ELGstreams, so they're not going to converge on that as a common platform. But right now, Nvidia *are* pushing the idea of a new API pretty hard - presumably because they'll be able to ensure that the design can be implemented on top of their existing code.

And you're right, developing and moving to such an API is a lot of extra work for people who were already happy with GBM. But can you see any way forward, other than this?

[patstew]

If you watch the Nvidia/Jason presentation and the one just before it on the XDC livestream then the Intel guys say that they always use the slowest/most pessimistic buffer allocation methods due to the current API. And it's even worse with the mobile GPUs. This isn't a particular problem with Nvidia's drivers, the problem is that the GBM interface isn't expressive enough to give good performance (This is pretty obvious by the positive response to Nvidia's presentation). The difference is that the open drivers are concentrating on making a working driver, not on performance, whereas Nvidia's customers (not linux gamers, big customers) require top performance so putting out a version of the closed Nvidia driver with shitty performance isn't an option.

[bulletxt]

The difference is that the open drivers are concentrating on making a working driver, not on performance, whereas Nvidia's customers (not linux gamers, big customers) require top performance so putting out a version of the closed Nvidia driver with shitty performance isn't an option.

People here don't understand this. They think Nvidia just have to come out with a driver, then performance will magically improve. Nvidia has a business and they are the GPU leaders. Nvidia has the know how, they know what they are doing and if they are refusing GBM and want to create a new api it means that GBM is shit for them.

[starshipeleven]

So frankly, yes. Unless something changes, you've summed it up neatly - one of the three major GPU platforms, as you say, cannot cope with Wayland. For real.

Wrong. "Cannot" implies they really cannot. In this case it's their own decision to not support GBM because of cost reasons and closed-source preference, while other companies do that fine as they use a less-frail approach.

And you're right, developing and moving to such an API is a lot of extra work for people who were already happy with GBM.

That "people who were already happy with GBM" are pretty much everyone in linux graphics ecosystem apart NVIDIA. Let's not understate this point.

But can you see any way forward, other than this?

Is eventually dropping NVIDIA from linux desktop a bad thing from a moral standpoint?
I mean, that's an adult company that made its own choices, and it's not everyone else's job to cushion its fall if they screw up, since this we aren't in a communist utopia, God Bless 'Murrica for this.

And this is assuming everyone else actually has the resources to cushion its fall, and that isn't a given.

[patstew]

That "people who were already happy with GBM" are pretty much everyone in linux graphics ecosystem apart NVIDIA. Let's not understate this point.

Watch the videos, none of the devs seem to think that GBM is sufficient in its current incarnation.

[starshipeleven]

Watch the videos, none of the devs seem to think that GBM is sufficient in its current incarnation.

If it was like that, how do you explain none has jumped on EGLStreams in the last 2 years?

Afaik an answer that is thrown around a lot is that GBM can be expanded. Has any dev in the video talked about expanding GBM to overcome its shortcomings?

[starshipeleven]

sigh, unapproved post for @pastew above.

[Delgarde]

Is eventually dropping NVIDIA from linux desktop a bad thing from a moral standpoint?

Go ask a philosopher. Personally, I'd rather have everything open - but that's out of pragmatism, not idealism or morality. And from a pragmatic standpoint, excluding Nvidia is an absolute non-starter...

[starshipeleven]

Personally, I'd rather have everything open - but that's out of pragmatism, not idealism or morality.

Yeah if it wasn't a tradeoff I'd agree. Point is, it is a trade-off so you cannot say what is better without keeping in mind what is the cost of that "keeping everything open" for your "pragmatism".

And from a pragmatic standpoint, excluding Nvidia is an absolute non-starter...

Just as it is changing GBM now. You'd be freezing graphics development and wayland porting for quite a few years, and slowing it down afterwards too as there would be 2 backend protocols. That's not acceptable.

Really, you cannot achieve high ground here as there is none. You must take sides, and it all comes down to what you want most.