ATI Drives Graphics Performance for Linux Users With New Unified Driver [LWN.net]

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 21, 2002 22:33 UTC (Thu) by Peter (guest, #1127) [Link]

Another big and ugly binary only kernel driver.

Yup. They do claim to support OpenGL 2.0 though. I must have missed the latest news; I thought OpenGL 2.0 was still in the standardisation phase.

I also wonder about their claim that the unified driver will handle not only various and sundry Radeons but the "FireGL line". Because it's not really "a line" - first there was the FireGL 1, bought from Diamond / SonicBlue, then the FireGL 2/3/4 from ATI itself, all sporting IBM GPUs. As I understand it, these have nothing in common with the FireGL 8700/8800, which are based on ATI's own Radeon R200. So when they say the "FireGL line" do they mean all the FireGL branded boards or just the 8x00 series?

Hmmph, I suppose I should just download the stupid thing and check the source stubs. Naahh, too much work. (:

But then all performance will be useless if the driver is not stable.

Definitely. So why not open source the driver? *sigh*

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 21, 2002 22:49 UTC (Thu) by jwharmanny (guest, #971) [Link]

> Definitely. So why not open source the driver? *sigh*
>

Because they have, more then any other IT-tech company, some very important secret technologies to hide. The code of those drivers is worth very much money.
At least ATI and nVIDIA provide Linux drivers and support. Some other companies could learn much from them.

On the other hand, I believe I have the right to know what kinda card I buy, and the right to find out how it works. Freedom doesn't matter only for software.

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 22, 2002 0:44 UTC (Fri) by job (subscriber, #670) [Link]

Binary-only drivers is hardly "support". That crap taints your kernel so no kernel developer wants to touch it, which means no help for you on linux-kernel or any other of the usual places.

ATI! Please help the DRI effort! You would sell a lot more cards that way!

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 25, 2002 0:56 UTC (Mon) by zander76 (guest, #6889) [Link]

Do you have any idea what your talking about? Some of the idealistic
people will not touch a binary driver but they are turning linux into a
religion and not an operating system. Linus has stated on serveral
occations that we should not use inferior products just because of an
ideal. Remember its RMS (the dictator) that is all about free software
only, and not linus and not all the developers.
Free software is about having the best software not about some
communism/religion BS that you are passing around. You communists are
going to far with the tainting the kernel with binaries stuff. I have been
around linux for awhile now, and I am more then happy to have drivers that
work for my video card, that a company has been nice enough to provide
drivers for. There has been a lot of problems with video cards in linux
and now that 2 companies are producing drivers you're complaining because
they don't fit your religion. Why don't you write the right winged drivers
so there is no kernel tainting.

Zander

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 22, 2002 0:47 UTC (Fri) by coriordan (guest, #7544) [Link]

> Because they have [...] some very important secret technologies to hide

okay, then release a GPL version that does most of the good stuff and
release a binary-only version with these secret technologies. I don't need
their secrets and I would definitly buy video cards from a company that
releases GPL code.

I sure as hell won't be linking a binary to my kernel. I trust code that
Torvalds or Cox have accepted, I know it has passed a lot of peer review.

I don't mind if the free software communitys' driver is a little slower at
times. The lowest performance you will ever see is in a computer that has
crashed.

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 22, 2002 0:56 UTC (Fri) by stuart (subscriber, #623) [Link]

>I don't mind if the free software communitys' driver is a little slower at
>times. The lowest performance you will ever see is in a computer that has
>crashed.

Couldn't agree more.

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 23, 2002 0:02 UTC (Sat) by proski (subscriber, #104) [Link]

I trust code that Torvalds or Cox have accepted, I know it has passed a lot of peer review.

Agreed. Just check the driver - all the files in the RPM package for XFree86 4.2.0 have executable permissions - drivers, sources for the open part of the driver, pixmaps and even tar archives. That's an indicator of their (not) understanding of UNIX permissions.

If the ATI engineers are so careless of their programming as they are of their packaging, you should never run that driver as long as you care about security of your system.

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 25, 2002 0:59 UTC (Mon) by zander76 (guest, #6889) [Link]

Do you really mean to tell me that you don't run windows or anything else.
Do you actually read the code yourself, there is over 600 packages in most
distributions and I know a few people that have found compramizable bugs
in linux software and have not released it and I have not hurd other
people finding it either. So do you really thing that all the software on
your machine is actually being audited.

Zander

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 22, 2002 2:52 UTC (Fri) by felixfix (subscriber, #242) [Link]

> Because they have, more then any other IT-tech company,
> some very important secret technologies to hide. The
> code of those drivers is worth very much money.

Nonsense. I have worked on lots of firmware, and believe you me,
we bought all the competitors' products and investigated them
If there was anything interesting in the drivers, we disassembled
them. Their competitors have long since figured out the secrets.

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 22, 2002 4:32 UTC (Fri) by Russell (guest, #1453) [Link]

So if they GPL the code, that kind of protects the secrets anyway. The competition can't use the secrets unless the competition shows you theirs?

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 22, 2002 11:37 UTC (Fri) by pontus (subscriber, #3701) [Link]

To release code developed in-house under a free software license is a very difficult thing for a company like ATI. An important consideration is the patent situation.
1) By releasing the driver, the internal workings become much clearer, and it becomes much easier for other companies to find or imagine patent infringements.
2) They likely implement their own patented technologies in the driver, and so they would have to license their own patents to everyone at no cost, making them worthless.
3) If there are ATI copyright notices in the source files, and somebody else adds code that infringes patents, ATI are potentially liable.

All this exposes ATI to severe risks if they should release source code. Trying to safeguard this kind of legal situation can be terribly expensive.

Consider instead that the Radeon series cards probably give the best 3D performance of any free software 3D drivers available, thanks to ATI supporting the DRI project.

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 22, 2002 12:32 UTC (Fri) by Peter (guest, #1127) [Link]

1) By releasing the driver, the internal workings become much clearer, and it becomes much easier for other companies to find or imagine patent infringements.

Kinda ironic, eh, considering that the entire purpose of patent law was to encourage publication of inventions?

2) They likely implement their own patented technologies in the driver, and so they would have to license their own patents to everyone at no cost, making them worthless.

Nonsense. First of all, unless they wish to start charging for the driver, they already have to license the patents for free to anyone who uses it. Second, and more importantly, a copyright license on a copyrighted work does not imply a patent license. For a relevant example, see the patent license language in some of the Linux kernel MTD code - it is separate from and more restrictive than the GPL terms. Patents cover use of something; copyright covers distribution. The two are often related but do not have to be.

3) If there are ATI copyright notices in the source files, and somebody else adds code that infringes patents, ATI are potentially liable.

Only if ATI accept the enhancements. How can you be liable for someone else's infringement action? Sure, ATI might get sued for this, but they have a pretty solid defense and it wouldn't go anywhere.

Consider instead that the Radeon series cards probably give the best 3D performance of any free software 3D drivers available, thanks to ATI supporting the DRI project.

That's good, and they should continue this. In fact what they should do is contribute their OpenGL 2.0 support code to DRI / XFree86 / Mesa, so they won't have to ship this binary crap.

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 25, 2002 14:16 UTC (Mon) by pontus (subscriber, #3701) [Link]

Sorry for some bad English there; when I said 'license patents to' I meant 'grant a patent license to'. You will agree that ATI is under no obligation to grant patent licenses for free when they distribute their binary-only drivers of course. However, if they should distribute the driver under a free license, without a patent grant, for example any Linux distributor shipping an X server with the code would be infringing.
You say 'Only if ATI accept the enhancements', but if they release it under a Free Software or OSI approved license, they can have no control over the modifications.
Whether they release it under their own free software license, or contribute the code to the DRI/XFree86 projects, the problem remains: if there are copyright statements by ATI in the source file, and the source file infringes patents, then ATI are on shaky ground. *I* wouldn't bet *my* company on the premise that a judge and jury will understand the subtleties of copyrights on single lines or characters in a source file. Let alone if modifications are separately innocent, but infringe patents if they are used together. Being in the right doesn't mean that you will win in court.
I can understand ATI's position here, and I believe they are doing as much as they reasonably can.

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 26, 2002 15:56 UTC (Tue) by Peter (guest, #1127) [Link]

Sorry for some bad English there; when I said 'license patents to' I meant 'grant a patent license to'.

Don't worry, I knew what you meant.

You will agree that ATI is under no obligation to grant patent licenses for free when they distribute their binary-only drivers of course.

Binary-only has nothing to do with it. They are under no obligation to grant patent licenses at all. It's just that if a driver uses patented technology, it's a pretty useless driver unless the user is granted a license to use the patent.

There is nothing legally wrong with ATI distributing a GPL driver with a restricted patent license. They could say "you may redistribute this and make derived works from it to your hearts' content, but if you make use of X, Y and Z patents contained therein for anything other than driving an ATI Radeon series graphics card you are infringing the patents." The fact that the code is out there to be looked at and even copied is of no consequence: in fact the very purpose of patent law was to encourage open publication of inventions.

You say 'Only if ATI accept the enhancements', but if they release it under a Free Software or OSI approved license, they can have no control over the modifications.

That's true. It's also true that in many legal environments you are likely to get sued for getting up in the morning. The fact that you can get sued for something isn't really a deterrent nowadays because that's the case no matter what you do in life.

Look, if ATI publish XX code, and someone else makes a derived work that infringes NVidia's patents, but ATI refuses to distribute the derived work (by not accepting the third party patch), how can ATI possibly be liable? That's like saying, if I publish a paperback novel, and on the copyright page I say "this book may be freely copied, edited, and derived from", and you take my book, OCR it into your computer, do some editing, plagiarise a third party, and republish it (leaving my copyright on the front) ... that I would be liable for your actions. How silly is that? It's right up there with suing Xerox for someone's copyright infringement.

Being in the right doesn't mean that you will win in court.

You may as well give up and move to China where they don't care about IP law, then. (: That's true in much of the Western world no matter what you do. There is such a thing as being cautious of frivolous lawsuits, but I draw the line at watching out for getting blamed for something someone else does. It's like putting a sticker on a hammer, "WARNING: do not use to strike a human head, as permanent brain damage may result".

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 27, 2002 14:58 UTC (Wed) by pontus (subscriber, #3701) [Link]

I don't disagree with your arguments; there are good arguments in both directions. In the end, it's a matter of risk assessment. I believe that a big corporation's readiness to participate in free software or open source is determined largely by the competence and know-how of their lawyers.

The fact that you can get sued for something isn't really a deterrent nowadays because that's the case no matter what you do in life.

True, but only to a certain extent. The risk of getting sued is always there, but it can be large or small. The point with freely available source code that anybody may modify, is that ATI may have a hard time proving that their source code contributions are not infringing, if the copyright is shared. This seems to be a deterrent in this case.

It's right up there with suing Xerox for someone's copyright infringement.

Or right up there with suing Elcomsoft for someone's copyright infringement?

"WARNING: do not use to strike a human head, as permanent brain damage may result"

Or as in the real-word example on cigarette lighters, "Make sure flame is extinguished after use."? It's a crazy world...

ATI Drives Graphics Performance for Linux Users With New Unified Driver

Posted Nov 27, 2002 20:52 UTC (Wed) by Peter (guest, #1127) [Link]

The risk of getting sued is always there, but it can be large or small. The point with freely available source code that anybody may modify, is that ATI may have a hard time proving that their source code contributions are not infringing, if the copyright is shared.

I disagree. They should have an easy time demonstrating that the tarballs you download off their site do not infringe someone else's patents. The fact that tarballs are available elsewhere that do should be of no consequence.

Look, anyone can write something that breaks a law (copyright infringement, or patent violation, or whatever) and slap an incorrect copyright attribution on the top. Therefore a copyright line in a source file doesn't really mean squat. This is so blazingly obvious that I would expect even a legal court to see it. In which case, the only way to prove ATI infringed your patent would be to demonstrate that you could download the infringing code off their site.

In fact, if the source were freely available (open source or even just "community source"), that would provide an excellent defense for ATI, since it means anybody could easily have tampered with any incriminating source.

Of course, if ATI were in fact actively infringing someone else's patents, releasing the source code would be a rather dumb idea. I've been proceeding on the assumption that they aren't.

Or right up there with suing Elcomsoft for someone's copyright infringement?

Yes. (: But different, in that the wrath of SGI and NVidia (which ATI would theoretically have to face) is nothing compared to the wrath of Big Media (whom, by way of Adobe, Elcomsoft has had to face). And, until you start talking about TV-out ports and Macrovision, there are no DMCA issues with graphics card drivers. I understand that's why Matrox never released non-NDA'd info on programming their All-In-* bits. But really, I would much rather have an open-source ATI driver that disables the TV output port than a closed-source blob that handles Macrovision....