NVIDIA Denies Opening Up Its Driver

[Redeeman]

Once the GLSL is in place with an optimizing compiler the performance will majorly improve. But, that's a bit off yet, I fear- at least another 6-12 months before we start really seeing the results of the work in progress with Gallium3D, etc.

will we really see gallium being put to use in this timeframe? to be honest i would expect it to take longer?

[Chris]

Another issue people don't seem to fully realize with completely GPL'd FOSS drivers is patented (evil word) technology. For example, floating point textures and rendering (more or less a requirement for HDR rendering) is patented by SGI. You must get a patent license to implement it, and unless SGI decides to allow FOSS developers to implement it royalty-free, you'll never see that supported in GPL drivers (GPLv3, at least).

Another is DXT compression, patented by S3. Will also likely require license fees, and will not see the light of day in a GPL(v3)'d driver, unless S3 is feeling rather generous. Not to mention the various NV and ATI-specific extensions, some of which no doubt have patent encumbrances. And of course, there's the whole thing with MS claiming patents on the programable/shader pipeline..

[hubick]

Workstation customers make purchasing decisions based on proprietary software features and performance, which HW vendors spend a ton of money developing and don't want to give away to their competitors.

Workstation seems to be an exception to the typical user base, since the "Linux workstation market" is really the old "Unix workstation market", which didn't have a problem with closed source drivers.

the Linux workstation drivers are going to share a lot of code with drivers from other OSes, which in turn means that "opening up the drivers" puts your competitive edge at risk in all OSes not just Linux.

but every one I spoke with (including both ATI and NVidia customers) said that what they cared about was performance and stability on a standard commercial OS distribution (typically RHEL or SLES/SLED), not open-ness or ability to work across arbitrary distros.

GAH!

It seems that maybe you and/or your customers don't yet "Get It" when it comes to the Free Software model and all the reasons *why* they are using Linux today. Linux and the Free Software model has basically killed off Unix, for good reason, and it's now obvious which model has won out in the market. I think Nvidia, and apparently still AMD, need to realize that they are primarily a hardware company, and thus need to compete with the quality of their hardware, and to STOP looking for a competitive edge in software. Make the software a *shared* commodity so *nobody* has any advantage and you can all work together on it - that's what Linux is all about and you can only fight the model for so long before you either get with the program or someone who does eats your lunch (like maybe Intel) and you head down the same path proprietary Unix did.

[bridgman]

I think you just touched on the real issue -- the fact that there are at least two totally distinct groups using Linux, with totally different reasons for using it.

One group strongly embraces the "Linux is better because it's free and open" thinking, but the other group either feels there is a simpler explanation (the commercial Unix vendors underestimated how quickly PC hardware would progress, dragged their feet on porting to PCs and got their butts kicked out of the market as a result) or simply don't care.

To that second group, the "free and open-ness" of Linux is much less of an issue than the ready availability of a commercially supported "*nix" system for PC hardware, leveraging the work done by Red Hat, Novell and others to use Linux on commercial servers. The Unix platforms which embraced PC market are doing OK (BSD-based MacOS had almost 5% market share last time I checked, and Solaris seems to be showing renewed signs of life) but most of the other Unix vendors missed the boat.

I agree that for the Linux-specific portion of our market we should be embracing open source and that is obviously what we are doing -- I expect that in the not-too-distant future the majority of our consumer Linux users will run the open source drivers -- however there is also a small but important segment of the Linux market (the "second group" above) which expects feature and performance parity with Windows and for *that* market we are continuing to offer a closed-source driver to leverage and protect the significant investment we make in driver code for other platforms.

[duby229]

what Linux is all about and you can only fight the model for so long before you either get with the program or someone who does eats your lunch (like maybe Intel) and you head down the same path proprietary Unix did.

Intel is already eating ATi's lunch... It is Intel that is making just about every major breakthrough in the linux graphics infrastructure. It's Intel that is shaping the future. Not ATi. The way ATi is going right now, the year or so after Intel releases a top end competitor ATi will be done.. It's a shame really, but the only one to blame is ATi.

[Melcar]

^
Funny how a thread about nvidia is turning into a "it's ATI's fault" thread. AMD is trying to please both sides here, and I think they are doing a good job at it. They acknowledge the needs of the OSS community, while still maintaining their commitments to the small (but influential) UNIX crowd that still desires proprietary stuff. This is far more than what nvidia or Intel are doing; nvidia seems to simply not care about Linux proper, and Intel simply does not have the same obligations ATI and nvidia have as far as high power workstation graphics go (until Larrabee pops out that is).

[duby229]

From the article this thread is about.

however, there is a (smaller and largely unspoken) set of people outside of NVIDIA that want the binary curtain to remain. They fear that if NVIDIA provides an open-source alternative, kernel developers may become even more hostile towards binary blobs.

In this theoretical but possible situation, NVIDIA and AMD would continue producing their binary-only drivers while at the same time supporting the alternative open-source drivers that are focused on the "out of the box" experience and not the high-performance full-featured driver that would remain closed-source. With there being open-source alternatives for all major graphics hardware, free software zealots could make the decision (or several decisions over time) that would block critical elements of the kernel that are necessary for the binary drivers to work. If the kernel developers do this and there ends up being no revert of action, they have just severely stabbed Linux.

This would likely result in a fork, legal action, or other challenges in order to overturn such a decision to reach a new common ground. If no common ground was reached, NVIDIA and AMD would be put in an awkward position of not being able to support their workstation customers with the high-performance drivers they need as they can't have some of their third-party intellectual property and other work exposed in an open-source driver. If it were really severe, the workstation business could just pack up and move to another operating system. That's what some fear at least.

I've taken the liberty of bolding what I have issues with, and underlining what pertains to AMD (aka ATi)

I just wanted to make it perfectly clear that this snippet from the front page article this thread references, is what I am talking about. Micheal references both nVidia and ATi, and while nVidia may be more guilty then ATi at this poitn they are both guilty. I'll give AMD some credit that at least they are trying. Nvidia isant even trying. And lets face the facts here, when (not if) binary blobs are denied access to the kernel the only company that will be prepared for it is Intel. If ATi continues down the path that they chose it is doomed to fail. And the only one at fault is themselves. The way I see it is when Intel releases a top end competitor nVidia is doomed. ATi still has a chance if the open driver is mature enough to take on the full load of supporting the Linux community. If not then ATi is doomed as well.

[hubick]

however there is also a small but important segment of the Linux market (the "second group" above) which expects feature and performance parity with Windows and for *that* market we are continuing to offer a closed-source driver to leverage and protect the significant investment we make in driver code for other platforms

Sure, but what happens when a competitor comes along who gives consumers _both_ an integrated open source solution *and* high performance? You just don't think it can happen? Or think you are capable of reacting quickly enough if/when it does?

[ethana2]

I'd just like to say, and this may not be particularly /informative/, but you folks at AMD and intel are doing an awesome job, and I would feel no qualms at all about buying the newest most powerful gpu you offer. Keep up the great work. We love you guys, you rock.

I'm of the opinion that hardware vendors -only- obligation for data provision to customers is hardware specs.

Providing drivers for our OS /and/ helping code free ones is just amazing. Think about it. That's a lot of stuff for free. AMD is going above and beyond the call of duty as I'm concerned, and I really appreciate that.

[remm]

We adopted a two driver strategy because we felt that workstation absolutely needed a closed source solution (because of the market pull for parity with Windows and the competitive challenges that brings) while most other Linux users could be satisfied with an open source stack that had all the important functionality but not the proprietary features or performance work which feeds the workstation market.

The reason they "need" a closed driver is first of all that, when they buy their (new) hardware, there's no 3D available at all with the OSS driver. Before making claims that performance is king (which is fine by me, but you'd also have to demonstrate that an OSS driver cannot have good performance), I'd like to point out that, as of now the ATI OSS drivers have: very early 3D for R500, no 3D for R600, and nothing for R700. How could a user possibly choose to use the open source driver exactly ?