AMD/ATI Commits to open source drivers

[macemoneta]

This was just announced a few days ago:

"AMD will soon deliver open graphics drivers, said Henri Richard just a few minutes ago, and the audience at the opening keynote of the Red Hat Summit broke into applause and cheers. Richard, AMD’s executive vice president of sales and marketing, promised: "I'm here to commit to you that it’s going to get done." He also promised that AMD is "going to be very proactive in changing way we interface with the Linux community."

It looks like Nvidia's is now the only closed source driver from a major manufacturer. From here (http://enterpriselinuxlog.blogs.techtarget.com/2007/05/09/amd-will-deliver-open-graphics-drivers/).

[retsam]

cough....i predict game cheat will explode because of this...

[at0mik]

I've juste bought an nvidia... My futur card will be an ati if there's no nvidia open source drivers...

[lightman]

It looks like Nvidia's is now the only closed source driver from a major manufacturer.

Sorry ? As of now, ATI has no open source driver. I will believe it when I see it ;)

Not that I care, though. I'm not going to spend half a grand on a card just to use it up to 70% (or 50, or 80, or even 95) of its potential just to be able to say "hey, but I'm using an open source driver". I use my cards for work, and that means that I want the maximum possible performance from them, if this means that I have to use a closed source driver then be it.

[grey_1]

My question is "So What"? Who really cares other than to be able to say "I only use open source software".

I have never had an issue with nvidias drivers for linux. Amdti needs to drastically improve their quality and the install process. I really don't care if it's open or closed source just so long as it works.

[hvengel]

AMD has always been OSS friendly. One of the first things that came into my head when I heard that AMD was buying ATI was that perhaps AMD might do the right thing by moving ATI to an open specs/source model once the merger had happened. Now we are seeing that perhaps this might actually happen.

AMD has had very good support for and from the Linux community over the years and has also had a good working relationship with the Linux kernel team. Intel has also had a good relationship with the kernel AND x.org teams. The natural question is who does AMD view as their main competitor, Intel or nVidia, and does this play into how they will handle the closed vs. open approach with regard to the GPU drivers and specs? My gut tells me that they view Intel as their main competitor and that they will eventually go open source/specs like Intel. The advantage that AMD/ATI have is that their graphics hardware is far more powerful than anything that Intel has available and if they do go the open specs/source route they will have a significant advantage over both Intel (because of having better hardware) and nVidia (because of the better relationship with the community) in the OSS market segment.

I have worked with another set of hardware vendors that make color measurement equipment and were competitors in their field. One of these also had a tradition, like AMD and Intel, of providing specifications that would allow OSS projects to interface to their devices. The other company had also done this in the past but had stopped doing this a few years ago and only offered closed source interface libraries. In addition, those libraries were only offered to the public for Windows and the Mac even though they had linux versions in house. About a year ago they merged and late last year dropped support for all of the devices that had open specs and basically became a closed source/specs shop. Their sales people were very surprised when I and other folks working in the application space that used these devices reacted negatively to the news of this change in company policy and they appear to be rethinking this at this time.

What most of us don't seem to realize is that the market share for Linux end user machines (non-server) has grown significantly over the last few years. In fact the installed base of end user (non-server) Linux machines was the same size as the Mac installed base in Dec of 2004 at about 2.8% of the installed base and was growing at a faster pace than the Mac user base. Projections at that time were that this would grow to 6% by the end of 2006 and 10% by 2010. My point being that many think that Linux on the desktop is a less than 1% market share sort of thing when in fact there are now more end user Linux boxes then there are Macs by a considerable margin. In spite of this all of these hardware vendors supply high quality Mac drivers but fail miserably when it comes to Linux. Seems kind of screwed up to me.

In any case I am optimistic that AMD actually understands this and that in the long run they will do the right thing and open up their GPU code and specs. Since this can only help them in the long run.

[lightman]

Slightly OT, but,

I have worked with another set of hardware vendors that make color measurement equipment and were competitors in their field. One of these also had a tradition, like AMD and Intel, of providing specifications that would allow OSS projects to interface to their devices. The other company had also done this in the past but had stopped doing this a few years ago and only offered closed source interface libraries. In addition, those libraries were only offered to the public for Windows and the Mac even though they had linux versions in house. About a year ago they merged and late last year dropped support for all of the devices that had open specs and basically became a closed source/specs shop. Their sales people were very surprised when I and other folks working in the application space that used these devices reacted negatively to the news of this change in company policy and they appear to be rethinking this at this time.

let me guess ? Monaco Systems and X-Rite ? ;)

[MamiyaOtaru]

My question is "So What"? Who really cares other than to be able to say "I only use open source software".

A lot of people do. Otherwise no one would be working on Nouveau, no one would be waiting for the Open Graphics guys, no one would take this announcement by AMD and get excited by it, or by Intel's upcoming discrete card.

I'd love to see a driver in the kernel so I can compile it and go without dicking around with binary modules. Even forgetting inclusion in the kernel, I'd love to see drivers included with more distros and usable in a livecd without worrying about licensing.

Hey, nVidia can choose not to open up, ever, but I can choose to go to whoever else does first. I got an nVidia card for my new machine, but I also got it with an integrated Intel gfx chip that works no fuss in Linux, as a way of supporting them. I'd be more than happy to replace the nVidia card with something powerful from Intel.

[doctor_octagon]

My question is "So What"? Who really cares other than to be able to say "I only use open source software".
Huh? Perhaps you missed the whole philosophy of the entire open source software movement?

I suggest you read The Cathedral and the Bazaar again to rediscover all the advantages of open over closed source.

Doc

[grey_1]

Huh? Perhaps you missed the whole philosophy of the entire open source software movement?

I suggest you read The Cathedral and the Bazaar again to rediscover all the advantages of open over closed source.

Doc
Haven't missed it at all. Been pretty keen on it for a few years. I will admit that I'm probably missing something though, as a number of distros do include access to drivers. It just looks to me like a lot of people very upset over licensing which the manufacturer has every right to implement.

I don't mean this rudely, really. But my $400.00 video card isn't a bit of philosophy. It's hardware that I want 100% performance from. With that in mind, open-source is not always better just because it's open-source.


A lot of people do. Otherwise no one would be working on Nouveau, no one would be waiting for the Open Graphics guys, no one would take this announcement by AMD and get excited by it, or by Intel's upcoming discrete card.

I'd love to see a driver in the kernel so I can compile it and go without dicking around with binary modules. Even forgetting inclusion in the kernel, I'd love to see drivers included with more distros and usable in a livecd without worrying about licensing.

Hey, nVidia can choose not to open up, ever, but I can choose to go to whoever else does first. I got an nVidia card for my new machine, but I also got it with an integrated Intel gfx chip that works no fuss in Linux, as a way of supporting them. I'd be more than happy to replace the nVidia card with something powerful from Intel

And choice is what it's all about. My 8800 works with no fuss in Ubuntu btw, very little 'dicking around' :p needed once I worked at it a touch.


It almost looks like "open-source the religion". Me..I just want it to work, who ever the manufacturer of my current hardware.

[macemoneta]

This isn't a philosophy issue, and it's not a Linux issue. The benefits of open source apply equally to Windows and OS X.

Just ask the folks on this board who have video cards that are performing fine for them, but have been told that they must now use the "legacy" driver (and will no longer see any enhancements).

The laptop model that Toshiba released after my model came with a remote control for multimedia. Six month later, Microsoft released a service pak for Windows. The remote control no longer worked. The manufacturer said the unit was "no longer supported".

Open source protects you, as a consumer, from those events. It offloads development resource requirements from the manufacturer. Hardware manufacturers get to concentrate on making great hardware, not trying to write drivers for whatever platform their customers want.

If you were a computer user in the late 70's, you'd know that this was the way it used to be. Every piece of hardware came with full specifications (even schematics). You could easily write a driver for anything you purchased.

Things changed after the infamous Bill Gates posting (http://en.wikipedia.org/wiki/Open_Letter_to_Hobbyists), where he indicated that sharing software (the norm at the time) was "stealing". It's been all downhill since then.

[grey_1]

This isn't a philosophy issue, and it's not a Linux issue. The benefits of open source apply equally to Windows and OS X.

Just ask the folks on this board who have video cards that are performing fine for them, but have been told that they must now use the "legacy" driver (and will no longer see any enhancements).

The laptop model that Toshiba released after my model came with a remote control for multimedia. Six month later, Microsoft released a service pak for Windows. The remote control no longer worked. The manufacturer said the unit was "no longer supported".

Open source protects you, as a consumer, from those events. It offloads development resource requirements from the manufacturer. Hardware manufacturers get to concentrate on making great hardware, not trying to write drivers for whatever platform their customers want.

If you were a computer user in the late 70's, you'd know that this was the way it used to be. Every piece of hardware came with full specifications (even schematics). You could easily write a driver for anything you purchased.

Things changed after the infamous Bill Gates posting (http://en.wikipedia.org/wiki/Open_Letter_to_Hobbyists), where he indicated that sharing software (the norm at the time) was "stealing". It's been all downhill since then.
This makes more sense to me. I've never really delved into the history of open source, more of a rather nuetered opinion based on others comments.

I've been one of the lucky ones then as I have never had an issue the way your explaining it, but thanks for the patient :p clarification.

[hvengel]

Slightly OT, but,



let me guess ? Monaco Systems and X-Rite ? ;)

X-Rite and GretagMacbeth actually. They merged in June 2006. X-Rite had up until Dec 2006 always released device specific interface specs for their hardware to developers. No need to sign an NDA or even a license for that matter. Just sign up as a developer. On the other hand GretagMacbeth required developers to sign a license agreement that even precluded making calls to their interface libraries in open source software. GretagMacbeth stopped releasing hardware interface specifications for all hardware released after the Spectrolino and it's derivatives.

What makes this particular case so absurd is that the interfaces to these devices are actually very simple compared to something like a graphics card and all of the previously GretagMacbeth devices that are now shipped by X-Rite have had the low level protocols reverse engineered and there is now a GPL interface library available. I am now using this library in my software and it now has support for using a number of current and discontinued X-Rite/GretagMacbeth devices for calibrating and profiling displays in CVS. This has only been tested on Linux and BSD platforms so far but testing should begin on Windows and OSX soon.

I was told by the person who is the head of their software division that they would not release the low level interface specs since this would compromise their "IP" but they also refused to release the interface libraries for Linux/Unix platforms even though they had these available in house since at least 2004. So they basically backed the OSS community into a corner by refusing to provide any support even though I and others literally begged them to release the existing Linux/Unix libraries.

I know the person who did the reverse engineering on these devices and he told me that it took him several hundred hours to have the whole product line reverse engineered and working in his GPL libraries. He also told me that if he had access to the interface specs that he would ahve had a working library for these devices in perhaps 20 hours. So it appears that their notion of how to protect their IP was not very well connected to reality. In the end all they did was create a bad relationship between themselves and those of us in the OSS community who work in that field.

What this really points out is that hardware vendors can take a number of different paths.

1. The path that GretagMacbeth took - which is to basically say to the Linux/Unix/BSD community that you don't want to be bothered and that you will not even supply basic device information to allow them to write their own interfaces.

Clearly this approach does not work well for either the user community or the hardware vendor since in the end the user community will either reverse engineer the hardware anyway exposing whatever IP the vendor was trying to hide or it will walk away from the hardware vendor (if there are alternatives). In addition, it will also leave a bad image of the vendor in the user community.

2. You can take the approach that most of the video card vendors have taken which is to supply just enough in the way of binary drivers/interfaces to prevent the community from creating their own open source drivers.

Nvidia is one of the better examples of how this could be handled. But even at that there are lots of problems that still exist. For example, only a very limited number of hardware platforms are supported. In addition, it appears that most vendors can not pull off doing this successfully. All we have to do is look at ATI and Matrox for two examples of vendors that have failed to be successful at this.

3. You can take the old X-Rite approach and supply hardware specs and let the community take care of writing and supporting interface libraries/drivers. Not a bad approach really.

It does require the hardware vendor to write the specification documents and make them available. So there is some additional work involved on the part of the vendors compared to #1. But beyond that there are no significant negatives for the vendor compared to $1 and #2 above. And compared to #2 it likely results in much lower support costs for the vendor and in the long run should result in better overall support for their devices on these platforms.

4. The approach Intel has taken which is to embrace the community and release your drivers as open source. I personally think this is the best approach.

[atriq]

Umm... anyway back to the thread topic:

Unfortunately, Henri Richard never explicitly said "open source graphics drivers", he said open graphics drivers, which could mean a number of things. Nor did he specify how "open" they would be; it's possible only portions of the code will be available which interface with binaries.

As said earlier, if they aren't actually available, the statement is worth nothing.

[amp_man]

I'll believe it when I see it. Linux support is the reason four of my last five video cards have been nVidia cards, the fifth having been a radeon 9500, and it's been long replaced. My laptop's built-in video is a Radeon x200, which isn't supported correctly through ANY driver. If that worked correctly, I might consider buying another Radeon for my desktop, but unless that happens, I'll make sure my next laptop has an nvidia chip too.

[Q]

Open sourcing Linux drivers makes sense, really, especially for ATI. They've always half-assed any Linux driver they put out. They've probably just given up for the most part and are going to let the vocal Linux users take the driver where it needs to go.