Nvidia working on Wayland support for binary driver | Hacker News Reader

Nvidia working on Wayland support for binary driver

lists.x.org

107 pointstbrock69 comments

> we are actively working on support and this certainly throws a wrench into things.

That's what happens when you do work in secret.

More importantly, why does NVidia, who contribute practically nothing to Free graphics development (drivers or the rest of the graphics stack) think that the Wayland developers have any obligation to make Wayland work well with their secretly-developed blobs?

If NVidia wants to have input, they could hire people to work on Wayland. Otherwise: https://www.youtube.com/watch?v=YxHqi8tx048&t=0m28s

Exactly. Nvidia has got away with crap for far too long. And given the rise of Intel and AMD GPUs which come with fully OSS drivers, they need to get their heads out of sand and realize that they can't get away with it anymore and they need to either invest resources in getting their GPUs supported under Linux - be it Android or Desktop - or at least make their specs open like AMD. (The way they supported Tegra on Android is also similarly crappy - look at how many popular Tegra Android phones or tablets were released post the Nexus 7 2012.)

The last time I tried to use an AMD video card with 3d acceleration under Linux it was a horrible experience. Nvidias binary drivers just work out of the box even with the most recent hardware and have OpenGL 4.3 support. Has the situation changed with AMD cards?

It works pretty well for me using the OSS drivers. Better experience than any of the binary drivers (NVIDIA and AMD), and the performance and features are good enough for gaming via WINE (though still not quite as good as the binary drivers AFAIK). I think the current pain points are support for the latest hardware and power saving for laptops.

Kernel 3.13 got full dual graphics ACPI shutdown on the discrete part, and they have had DPM enabled on all parts since 3.12. Should be good now on the power saving end.

They can get away with it because their drivers actually work. The OSS drivers are improving, but it's taken them a half-decade or more to come even within 70% of Windows performance parity. nVidia's drivers have performance parity on day 1, and generally, far fewer show-stopping bugs.

Really, FOSS driver development has been confined to the last 2 years. AMD has a half dozen developers in Mesa, and I think Intel has around 20. They have pushed it from Core Profile 2.1 to 3.3 in 2 years, and will probably have it at 4.0 or 4.1 by years end.

Before that, AMD wasn't making free drivers, and Intel didn't care about their integrated graphics as much and never had higher GL support on them. They have come pretty far in a short time frame, and by the time GL 5 comes out (I assume, the one with the low level primitive support to match DX12) Mesa should only be a version behind or so.

Also, I wouldn't really agree Nvidia drivers work. I had a GTX 285 in an older machine I put Ubuntu on, and their blob didn't work with resume / suspend at all. Dunno if they'd work now, I just switched that system to Nouveau, which... works with suspend flawlessly. And manages fan speeds a lot better than the blob, too.

AMD's been making free drivers for seven years. http://www.phoronix.com/scan.php?page=article&item=826&num=1

I said the last half-decade to be generous. It's taken this long to finally get some of the older cards to a mostly usable state for 3D gaming (usually the point of a consumer-grade 3D graphics card).

Mesa has been picking up new GL versions which is good, but it's taken a really long time.

>AMD's been making free drivers for seven years.

It takes lot of time and effort to get GPU drivers usable with all features. I would say they actually did a great job hiring OSS people, contributing documentation after lengthy legal reviews and they kept up with most older and newer GPUs. And they now have dynamic power management! That's great progress for me.

You're fooling yourself if you don't think they could've done better. There's no reason they had to start from scratch. Bridgman constantly stated the the free drivers were intended for basic functionality and free software development purposes only, and that AMD would continue to provide proprietary drivers for people who needed performance.

ATi started the free drivers in a last-ditch move of desperation to inflate sales in whatever small way possible when they were on the ropes. They'd hoped that this would get some of the *nix geeks to make significant buys. They only sponsored 2 developers if I remember right.

I don't mean to look a gift horse in the mouth, but I don't think they really deserve all that much credit. Seven years is far too long to start skirting Windows performance parity, and please note that most newer hardware is still way behind. The open-source AMD drivers primarily remain of interest to people who want to develop free software or whose cards stopped working on the proprietary drivers. When I tried last September, you still couldn't reasonably play games with open-source radeon drivers on newish hardware.

> More importantly, why does NVidia, who contribute practically nothing to Free graphics development (drivers or the rest of the graphics stack) think that the Wayland developers have any obligation to make Wayland work well with their secretly-developed blobs?

NVidia could also just not support Linux at all, which would certainly create even more of an uphill battle for Free graphics development and Linux.

That would be a pretty terrible PR move by nVidia. While I hate the fact that they don't open source their stuff, I have to give them props because their blobs drivers work really well on Linux too.

Abandoning Linux is definitely the wrong move, especially now that the gaming community is thriving.

Terrible PR among the tiny minority of Linux users.

Most people are using Windows or Mac and most of those are either completely ignorant, completely indifferent or actively hostile to Linux

You're not too familiar with the gaming community and the noise they can generate, are you? Just look at the Oculus deal, if you're in the gaming industry you definitely don't want to piss off gamers, no matter how much of a minority they can be.

This said, Linux is thriving a lot these days and is gaining a huge momentum as far as gaming goes. The community is constantly growing, the tools are all coming to Linux, SteamOS is right behind the corner, more and more companies are targeting the Linux platform, etc etc.

Bailing out of the Linux ecosystem now would be a terrible and shortsighted move on behalf of nVidia.

What are we supposed to learn about the power of gamers from the Oculus deal? They can whine on forums and cancel their pre-orders? That has been the only visible "negative" consequence of the Oculus/Facebook deal thus far.

Cancelled preorders, permanent PR damage, several dozen projects have been canned & fear has replaced confidence in consumers & devs alike when talking anything Oculus now. Not the rule but it's prevalent. You downplaying the impact is just plain stupidity.

I mentioned canceled pre-orders and I don't think that was unanticipated. I don't know how many were canceled, but with Facebook's coffers backing Oculus VR now, I must say I don't think they're overly concerned. My understanding was that the DKs were being sold primarily to finance further development and prototyping (and in particular to provide proof to suppliers that sufficient volume would exist), which is now a non-issue since they now have essentially as much money as they want, and can order x million units to get bulk pricing regardless of consumer demand.

The only major project that's been canceled was Minecraft, which Luckey privately pointed out was complete and utter vaporware anyway. I have a DK1 and I can tell you that most "Rift projects" were one-off tech demos that mostly weren't maintained, and no one will miss. If OVR continues to develop a good product, people will continue to develop for their platform, whether Facebook owns them or not.

What PR damage, outside of the group that they expected would be unhappy? It's all over the head of people who didn't know what Oculus was before the acquisition. I don't think people who aren't part of the industry really care, because they don't know anything about it. To the extent that it was covered by mainstream outlets, it was because Facebook bought another company, not because they were outraged that the dream of Oculus VR was being destroyed by FB.

Fear hasn't replaced confidence in all consumers and developers. Logical persons may have their opinions and predictions, but no one is justified in anything beyond "wait and see" until something actually happens.

I'm not downplaying the impact of the ire of the gaming community. I just don't think there's been a noteworthy one other than the vast amount of fruitless rage that's filled tech forums recently. Perhaps we disagree, but I don't think my assertions are "plain stupidity".

Thankfully, Linux has alternatives right now. And if you want free drivers, they aren't on the table anyway.

If Nvidia dropped off the face of the Earth in Linux support everyone would just buy AMD discrete or Intel integrated parts and not bat an eye. And both of those companies have dozens or hundreds of developers working in the FOSS space. And for some reason, they also get my money, where I purposefully avoid Nvidia products (I don't buy their Tegra APUs, I'd never buy a computer with one of their graphics accelerators, etc).

I guess you don't consider gaming a worthwhile endeavor on Linux?

I have no problem running most games under radeonSI. Only ones I have that don't work are Natural Selection 2 and Metro Last Light, which is mostly waiting on GL 3.3 in LLVM 3.5.

For example, DOTA2 / Portal 2 / TF2 all run at ~60 fps with high / ultra settings on my 7870.

There is nothing about driver openness that would impede their ability to run OpenGL games besides the version support, and that is rapidly being remedied.

or laptops apparently

I don't see where nvidia acts as if Wayland devs have any obligations here. All he said was that their driver would not work after this change. The only thing this obligates Wayland devs to do is not change this if they want nvidia to support wayland.

Could someone quickly recap what DDX and glamor are, and how they relate to XWayland, Xorg, Wayland, and nVidia binary drivers.

Disclamer: I'm not a kernel/driver developer, so take everything with some salt

Right now, every driver provide a different DDX (think about it as an API), which the X.org server uses to offload 2d drawing to the GPU.

Glamor is a way of doing 2D rendering on the GPU through OpenGL. It's currently used by RadeonSI (AMD 7xxx cards and newer), and can optionally be used by other free drivers (Intel, r300/r600, Nouveau).

If XWayland drop supports for DDX, then drivers that want to offer accellerated 2D on it have to support Glamor.

Free drivers can already use Glamor, so it's not a problem, but closed source drivers have yet to implement it (and I'm not sure if it's merely a manpower problem, or there are also licensing problems)

Just to simplify Spittie a bit for clarity, under X you have dedicated 2d drawing APIs. Wayland uses EGL, and just drops 2d entirely, so the entire desktop is running top of GLES. In order to run legacy X applications, Glamor is being baked into the X server so that X apps using Xrender can still work, but the binary drivers aren't using it yet, for a myriad of reasons.

So this in effect makes XWayland just another Wayland client, rather than what it used to be - the usual Xorg running in a very careful way not to step over Wayland and drawing on Wayland surfaces? And now nvidia are complaining that unless XWayland can have direct hardware access, they don't really have a path planned to supporting OpenGL and the like on XWayland, presumably due to them not supporting Wayland?

Perhaps their time would have been better spent finding ways to support Wayland proper and not coming up with proprietary workarounds that will stop being feasible the moment someone makes a small change.

That's not a small change.

As a long time Linux user, I have loved Nvidia for nearly a decade and continue to do so. They were the pioneers of serious 3D gaming in Linux and still the guys with the best driver.

Thanks guys, you rock! Keep it up!

I remember three or o years ago when NVidia showed no interest in making their driver separate from X on linux for use with Wayland.

It behooves nvidia's Linux division to be conservative when it comes to driver development. Things change fast in the OSS community, and you never know when someone is going to pull a TTM->GEM. It's best to have reasonable assurance that something is going to stick around before you invest the manpower into supporting it, lest you have to report to the bigwigs that you worked 3-6 months to support something, and ended up having to throw it away because a maintainer decided he liked a certain approach better. Incidentally, that seems to have happened here to a small degree.

What about Mir?

What about it? Why should they care?

Ubuntu/Unity is the best desktop environment I've ever used, but Mir can go die in a fire. Completely unnecessary. Everyone who looks like they know what they're talking about thinks so, at least, and I'll go with them.

Personally I am happy to see competition also in the display server protocols. The absence of it made X so stagnant

This is not a business. You don't want to see "competition" (competition in what?)

Variety is nice and having choices is nice, but when there's compatibility and man-hours to be spent in ensuring that compatibility/support across all the possible flavors of Linux distributions, I'm not sure we really want "competition".

It's APIs we are talking about here, standards. How would you feel if there were dozens of TCP/IP protocol standards (not just implementations, mind you) and if you wanted to develop any simple networked application you'd need to write a compatibility layer for each and every one of them?

Competition is definitely not what we need in this case, sorry.

I would agree with your position only if we could assert that Wayland is the best protocol we can ever have.

But can we affirm that? Can we suppose that there is no way to make a better protocol? What if Mir in the end will become better or will work better or will be more adopted? What if the guys of wayland will fail in making something really good? What if they take some wrong decisions and we will have no alternative in which rely?

We need everywhere competition.

Wayland exists because it was proposed with concrete reasons why X was rubbish and needed replaced.

So everyone said "yes, true, valid, you speak honestly here" and they proceeded to develop Wayland and everyone was on board with transitioning to it when it was ready.

Mir has never produced any arguments for its adoption and has only spread false information claiming why it has merits over Wayland.

Free software will usually jump on any software that can clearly state "this replaces X, for reasons Y, because X is insufficient for purpose in these domains Z and can't be fixed". You get situations where projects like pulse and systemd come under fire over disagreements over what Z domains X is appropriate for, not because there are arguments against Y being lacking functionality.

Wayland at least has (and strictly speaking is) a protocol, you can make a Wayland competitor speaking its protocol. In Mir the protocol is an implementation detail bound to break between versions.

Competition isn't the solution to every problem. Sometimes cooperation is the correct choice.

https://xkcd.com/927/

wayland is less negative about network transparency than mir. I am ALWAYS using network transparency (none of my X clients are running localy). When I am over "slow" network (for example to provide support to Portugal from France), I use vncviewer, but I do not even bother to run the x11vnc command on the same host than the display.

We have already lost so many years of work in the competition between Gnome and KDE.

OTOH, without competition we wouldn't have had XFCE to retreat to when both GNOME and KDE went weird.

I don't know what you're talking about. I've been using Blackbox/Fluxbox since 1998.

God heavens, why all the bloat, just what is wrong with TWM?

Seriously Canonical needs to market Ubuntu to device manufacturers to generate an income stream. I really hope they succeed although I think it is a shame that they have felt the need to fork the Graphics sub-system in this way.

You would have still had Openbox. And Enlightenment. And Xmonad...

One naked xterm - none of this girly-man "window-manager" rubbish - from which I start every other application, specifying -geometry. If I'm feeling effete that day.

I don't think that that's the case at all.

Even if we ignore the obvious differences in philosophy and implementation between the two, we're better off that they both exist solely because they've acted as hedges for one another.

When KDE 4 was released, it wasn't exactly ready for widespread use. While things did rapidly improve, GNOME at least provided an option for those who wanted a more modern desktop environment than KDE 3.5 in the interim.

The opposite situation has happened more recently. GNOME 3 was a complete disaster at first, and in many ways still is. Many former GNOME 2 users switched to KDE 4, by that time a mature and very usable desktop environment. Since we have seen only minimal effort from the GNOME 3 community to address its severe problems, the existence of KDE has become more critical for more users.

Do you remember how stagnant GCC was before Clang came along?

gcc is not a standard though, it's a product. It's a bit different in this case and I totally agree with you.

Well, that's mincing words. GCC is as much of a standard and/or product as Wayland or X are.

It is an implementation of the C, C++ etc standards, and it was also, for long, the canonical such implementation on Linux and Open Source OSes.

GCC wasn't stagnant at all. There were relatively frequent major releases including ports to additional platforms, very good progress implementing stuff like C++11, better optimizations, and many other important improvements.

Clang has helped show some additional areas where GCC could perhaps be improved, but GCC was surely not "stagnant" at the time. It wasn't like during the EGCS situation in the 1990s.

I remember how stagnant it was before EGCS came along, that's how old I am. :) EGCS was clearly technically superior to GCC (at least in some ways), and when the FSF gave up or came to its senses then the EGCS and mainline GCC projects were re-merged. (Compare also the Rails/Merb fork/re-merge.) By contrast, no-one outside Canonical seems convinced that Mir has important technical advantages over Wayland, and (as far as I've heard) Canonical seems to have little interest in preventing or minimising balkanisation, so it's hard to see the justification. The more efficient way for Canonical to speed up or shake up Wayland would seem to be to contribute to it, or failing that to maintain a fork designed to be a candidate for full or partial re-merging, no?

(I am not an expert on anything.)

How many key elements of a desktop Linux are there three or four of? How many browsers?

On top of which, everyone is ignoring the Android graphics stack architecture, which is by far the most common Linux-based, open source basis for an interactive userland, with the most apps.

Isn't one of Canonical's stated reasons for doing Mir to be more compatible with Android drivers?

Yes, but Jolla did the same with Wayland. Perhaps Mir is designed to be more "production-ready" for that approach, but it isn't unavailable with Wayland. I have not read the code, but everything I have read about Mir and Wayland tells me you can't slip a dime in the gap between their capabilities. You can, however, fit as much drama as the open source community can generate.

It was, but the way they achieved it was by reusing components Jolla developed for wayland. As it stands I believe their reason for developing Mir is non-technical. It might be that developing Mir increases the company's perceived value for investors.

It's about control (license-wise) over what's going to end up in Ubuntu Phone.

Source: Close friends work at Canonical. No proof, sorry - believe at your own risk.

I guess we don't need BSD either then (or the other way around). And directx. And so on. One tech standard to rule them all!

Competition certainly sounds nice on paper, but think of the problems that the various sound servers/api's have caused. OSS(+ 4Front's version), ALSA, PulseAudio, Jack.

Currently, OSS compatible APIs are what you have on (*nix && !Linux).

ALSA is monstrous to configure, but capable.

JACK serves a very specific purpose for music production.

I'm still not sure what pulseaudio is for, but I don't think I'm the target audience for it.

Point is, they don't exist contrary to each other, but in parallel and often in concert.

Pulseaudio is typically used with ALSA as a sink. You know how ALSA is a bitch to configure? Pusleaudio makes you not have to give a shit about it.

I get that, but Pulseaudio has also ruined my day more times than ALSA has... so as far as my specific requirements go, I'm not sure what exactly it's supposed to achieve. It's just sort of foisted on me if I pick the wrong distro.

Competition usually causes improvements in competing entities. However, this time, it might hurt.

You see, there's a compatibility issue. The display server has got to be 'supported' by everyone, from applications, toolkits, desktop shells, righ down to the GPU drivers.

That's why the wayland project took so long. Creating a display server is one thing, adopting it on the whole stack is totally different.

Free software, specially on the desktop side, simply lacks the manpower to maintain support for multiple display servers.

If Mir was a drop-in-replacement for Wayland, noone would've been mad. But this situation may cause a huge fork in desktop applications, shells, etc.

Ubuntu is on its and the whole community has adopted Wayland. Im sure sooner or later they will abandon Mir (just like Upstart), because Ubuntu is too small to patch everything just to support Mir.

This is excellent news. It's amazing to see the driver situation of Linux progressing so quickly these days.

I have to ask, did you even read the first sentence of the email from James Jones of NVIDIA? It says, "This direction seems to conflict with our plans to continue running our DDX driver under XWayland." The rest of the email isn't exactly positive, as well.

"Progressing so quickly" just isn't the first thing I think of when hearing about difficulties like those described in that email.

It kinda sucks if you want Nvidia, Wayland and *BSD. I really wish Nvidia would stop with the binary drivers.

Back when Matrox released their G200 card they used a little known fact in X11 that allowed them to do platform independent drivers. I'm not sure how it worked, but remember downloading the and using the "Linux" driver on OpenBSD to gain stuff like dual-head support. I'm not sure the same would work for hardware accelerated graphics, but it would be nice if Nvidia help getting their hardware supported on whatever platform people choose to use.

Gallium is actually designed to be cross platform too. Obviously AMD isn't using that (and they were the major developers of Gallium) but the only part of it that is platform specific is the winsys interface.

It does, however, mean you can port the Gallium version of the Intel Driver, or the Nouveau / Radeon free drivers, or the other Gallium drivers like Freedreno to any platform with just that one API switch.

I figure it was one of the justifications in funding Gallium from the AMD boardroom perspective - it isn't just a Linux driver, it is a pivot to easily get their driver on any other platform they need it on, whereas Catalyst proves to be very specific, requiring developers on both sides actively maintaining it to have a Linux + Windows version, plus the OSX one they certainly have.

> Wayland and *BSD

I believe Mesa does not even support EGL on BSD. Now I know the Nvidia driver doesn't use Mesa, but this indicates how little effort is being put into Wayland on BSD.

I can't say I blame them. BSD is primarily used on servers with no GUI, so the BSD community uses their limited resources to support that use case. It's the same reason BSD has poor support for WiFi and suspend/resume. But I don't think we're going to see Wayland on BSD for a very long time.

Reply on news.ycombinator.com