NVIDIA Driver Developer Discusses Linux Graphics

An anonymous reader writes "Andy Ritger, who leads the NVIDIA UNIX Graphics Team responsible for creating drivers on Linux, FreeBSD and Solaris, has answered many questions at Phoronix about the state of Linux graphics, gaming, and drivers. Ritger shares some interesting facts, such as: the Linux graphics driver download rate is 0.5% that of their Windows driver downloads at NVIDIA.com; how the Nouveau developers are doing an incredible job; creating an AMD-like open-source strategy at NVIDIA would be time intensive and unlikely; and development problems for the Linux platform. Also commented on are new features that may come to their Linux driver within the next twelve months." Like all stories at Phoronix, in common with most other hardware review sites, this one is arbitrarily and maddeningly spread across 8 pages.

Comments

Measurement from the NVIDIA site?

[Lord Lode]

I download my Nvidia drivers from the Archlinux package repository. How many Linux users manually download them from Nvidia? The 0.5 percentage could be a big understatement...

Re:Measurement from the NVIDIA site?

[Miamicanes]

> So why are those goodies disabled under Linux?
> Come on NVIDIA, just release the specs and let us write our own damned drivers eh?

It's not so much that they're "disabled", as a case of being "not implemented". The problem is that the line between what's a hardware capability and what's implemented mainly at the driver level through software is increasingly blurred. Just to give a familiar example, look at a PCI Winmodem. At the end of the day, a Winmodem is basically a PCI soundcard that's hardwired to a phone jack and optimized for PSTN-level voltages & impedance. Someone like Conexant could flawlessly document how to use the chips on one of their Winmodem cards to generate and sample audio, and it wouldn't do a thing to help anyone actually make the card act like a 56k modem under Linux. Someone has probably done it by now, but back when it would have actually still mattered (circa 1999-2000 or so), there was no such thing as an open-source Linmodem driver for that precise reason. Documenting the hardware was necessary, but even fully-documented, it would have only gotten you ~2% of the way towards the ultimate goal of *being* a software-defined modem.

There's another problem with video drivers -- patents. As a practical matter, everyone in the industry violates at least one patent belonging to the other big players, and they're *all* sitting ducks for every patent troll who comes wandering along. If NVIDIA were 100% altruistic, fully implemented every Windows feature into their Linux drivers, and released the full documented source code to their proprietary Linux drivers, they'd essentially be painting a red target on their forehead and making the patent trolls' fishing expeditions that much easier. It's sad, but it's true.

There's only two questions that matter

[QuantumG]

Q: Are there any plans in place to provide new features within the xf86-video-nv driver or to better engage with the Nouveau developers for some open-source support?

With the nv driver, we've always tried to provide something minimal that just works out of the box and requires the least maintenance. For that reason, feature set in the nv driver has stayed pretty slim.

The guys working on nouveau have done a really incredible job so far. However, our policy remains the same: we won't try to hinder their efforts, but we have no plans to help them.

Scumbags.

Q: AMD was able to open source and/or document a lot by separating out the parts they couldn't legally disclose. Similar problems have been cited as preventing NVIDIA from open sourcing their driver (licensed 3rd parts code, etc) or documentation. Could nVidia use the same strategy?

A similar strategy might be technically possible for NVIDIA, but for better or worse I think it is quite unlikely. There are several reasons for this:

- For competitive reasons on other platforms, I don't think we would ever open source any of our cross-platform driver source code (which is 90%+ of the Linux driver... see my earlier description of code sharing). The Linux-specific pieces of the driver code base don't really stand on their own, and generally need to change in sync with the cross-platform code, so I don't believe it would be practical to just open source the Linux-specific pieces.

- We have developed substantial IP in our graphics driver that we do not want to expose.

- Unfortunately the vast majority of our documentation is created solely for internal distribution. While at some point it may be possible to release some of this information in pubic form it would be quite a monumental effort to go through the vast amounts of internal documents and repurpose them for external consumption.

Yes, and there's a whole community that would like to help you do that. That second answer is the real point here. They don't want to open source it because they don't believe in open source. It's that simple. Hopefully this will kill the last of the NVIDIA apologists.

Re:There's only two questions that matter

[Anonymous Coward]

Yes, and there's a whole community that would like to help you do that. That second answer is the real point here. They don't want to open source it because they don't believe in open source. It's that simple. Hopefully this will kill the last of the NVIDIA apologists.

No it won't. I have dual 9800s and it runs WoW like a champ (read: no spontaneous system resets or strange/spurious bugs.) I'm unlikely to change to a card with an open source driver anytime soon, because what I have right now works.

I'm really mystified by this attitude - if a company produces a stable, reliable product with closed software and the market is willing to pay for it, what difference does it make? It's not like they are charging $$$$ for crappy product, like, say, Windows. And if you can't understand nVidia's position - e.g. maybe they really DO have some novel graphics processing pipeline in their software that provides them with a competitive speed advantage over the competition - then it's worth keeping it obscured. That's capitalism. (Speculation: some of their binary code dynamically optimizes an FPGA on board for better performance.) You can be damned sure that if I suddenly managed to come up with a novel algorithm for faster database transactions that I'd keep it secret, too, and then sell the hell out of it in competition with Oracle and DB2. Again, that's not bad, that's competition.

Don't get me wrong, I doubt that 95% of the code that is closed is actually worth closing. It really might be that last 5% scattered everywhere that warrants keeping it closed.

Re:There's only two questions that matter

[Kjella]

Scumbags. (...) They don't want to open source it because they don't believe in open source. It's that simple. Hopefully this will kill the last of the NVIDIA apologists.

Oh, STFU and volunteer yourself to go write open source AMD drivers. They've been running an open source strategy now for 2+ years and they're still short on manpower even though there's plenty specs out there and AMD is actively leading the development on top of the hours they've spent getting the documentation through legal review. There's plenty evidence to suggest the open source drivers would drop dead if AMD wasn't carrying them every step of the way, you think nVidia is impressed? The alleged army of open source coders waiting for specs is more like a handful, that's not a claim it's a fact. By all means they're making great progress and all that but they're way, way behind the blobs still.

Re:There's only two questions that matter

[petrus4]

I'm really mystified by this attitude - if a company produces a stable, reliable product with closed software and the market is willing to pay for it, what difference does it make?

The reason why you can't understand this attitude, is because you're not a Stallmanite freetard.

You're essentially correct; from any sane, neurotypical point of view, there's absolutely nothing wrong with nVidia's hardware or its' drivers being proprietary whatsoever.

Hopefully this will put an end to some trolling.

[serviscope_minor]

I'd like the authors of some common troll s to note:

a) The most high profile binary kernel module distributor considers the unstable kernel API to be very little trouble.

b) One of the most high profile X driver cerators thinks that X is well designed.

so there.

Lies, damn lies, and download rates

[Antique Geekmeister]

One powerful reason for the low Linux download rate is because the packaging for the NVidia Linux drivers is terrible. It doesn't upgrade properly, it replaces system provided OpenGL libraries with little warning, and it has lacked (the last time I looked) a way to detect if there is a more recent driver available. Instead, people install the freshrpms or atrpms or other repositories that report dependencies and available updates more reliably for RedHat based software,

I shouldn't have to compile a kernel module in order to install a software package: it should be published, or at least publishable, along with the updated kernel itself. But NVidia refuses to use licensing that would permit this, so they're going to continue to have people not only using alternative installation sources, but becoming quite angry when they update their kernels and their graphics drivers from NVidia stop working until they can be recompiled and a new kernel module built.

Re:Ran ran ruu!

[jim_v2000]

It is interesting then that ATI gfx drivers have generally been a pain in the ass to use in Linux, while Nvidia's work well most of the time.

I think the problem

[Sycraft-fu]

Is that you get people who've toyed with writing a driver for something simple, and get lulled in to thinking that means that all drivers are not a huge deal. Problem is that's not the case. Something like a basic SATA controller really doesn't have a whole lot in the way of functionality for you to implement to get a driver up and working. You can see this in terms of downloadable driver sizes too. Take a look at something like the MegaRAID cards from LSI. The actual driver is all of 25k.

Well that's not the case with graphics cards. They are extremely complex beasts, and getting more complex all the time. You are working to implement a very complex API (OpenGL). As such the driver is going to be much more complex. You can again see this in terms of driver sizes. The core nVidia driver for my 7950 here at work is 16MB. That's just the main driver file, there are other support files it needs to work, and then more files on top of that to really give you all the functions you want (like the custom control panel and such).

So it is a much harder job. It is also a continually moving target. As of this month, we now have a new generation of graphics hardware out that has major differences. The DirectX 11 gen hardware (Radeon 5000 series) is quite different from the previous gen in terms of what it can do. As such the drivers are going to be different. It isn't a case of "Just update the old drivers for the new hardware." It is writing drivers to support a whole new set of features.

Thus I think you get people who have this "Oh it isn't so hard," idea because they've played with the simple stuff. Ya well, sorry guys this isn't simple. In fact, I'd wager graphics drivers are the most complex drivers on systems these days.

As such I can see why nVidia isn't impressed. It isn't a case of "Just give us the docs and we'll knock out a dynamite driver in a week." They might like to pretend that is how it'll be but it's not.

Re:And don't forget the NVidia non-user base

[Chandon Seldon]

You're so convinced that preference for open source software is a question of "dogma" espoused by "purists" that you haven't stopped to consider the practicalities of the issue. When it comes to drivers on Linux, proper open source releases have huge practical advantages:

• They can be distributed with distros without anyone needing to jump through weird hoops.
• They can be maintained in-kernel, so they work with new kernel releases automatically.
• They can be fixed by the community, so they have fast turnaround on annoying bugs and favorite features.
• If the device manufacturer doesn't keep up with them, they don't instantly code rot.
• They can be integrated with other standard code, so they do all the normal stuff without anyone needing to re-invent the wheel.

When it comes to graphics drivers, these issues are mitigated to a large extent by the fact that Nvidia and ATI have very active driver teams that keep up with things. There are still some advantage to Intel graphics from open source drivers: you'll never have to worry about picking "old" or "new" driver packages like Nvidia for example. Having the option to one day run OpenBSD is another. But, in general, using Nvidia or ATI blobs on Linux is reasonably painless.

The same is absolutely not true for any other kind of hardware. Proprietary network drivers, RAID drivers, printer drivers, or webcam drivers are simply a nightmare - much better to get something with in-kernel drivers that will just work out of the box. The manufacturer *will* forget about you and leave you stuck on random old kernel revisions limping along with an unsupported driver.