Terry Makedon (AMD/ATI): "We can't open our drivers"

[ivanovic]

The german newssite golem.de did have an interview with Terry Makedon, chief developer of graphics and chipset drivers for AMD/ATI. In short he did state the AMD/ATI will not open their drivers in the near future ("In the next six to twelve months this (opening the drivers) is not the road we will take") due to "proprietrary technics from 3rd party companies".
You can read the whole interview at golem.de, but keep in mind that it is in german, all statements I did qoute in here are translations from the german text: http://www.golem.de/0703/50995.html

IMO it is not necessary to open the existing drivers. I would already welcome if the specs for the cards were published. I don't think it will reveal any secrets when you tell someone in which register you have to write a value to achieve something. You can't tell me that these specs will really tell your competitors much about the internals of the hardware. Now I only got to hope for Intel releasing some add on card with free drivers since they seem to be the only real alternative...

[d2kx]

For those of you who don't speak German:

he said that the Linux CCC is coming in 8.36 or 8.37.

[Svartalf]

Originally Posted by ivanovic

The german newssite golem.de did have an interview with Terry Makedon, chief developer of graphics and chipset drivers for AMD/ATI. In short he did state the AMD/ATI will not open their drivers in the near future ("In the next six to twelve months this (opening the drivers) is not the road we will take") due to "proprietrary technics from 3rd party companies".
You can read the whole interview at golem.de, but keep in mind that it is in german, all statements I did qoute in here are translations from the german text: http://www.golem.de/0703/50995.html

IMO it is not necessary to open the existing drivers. I would already welcome if the specs for the cards were published. I don't think it will reveal any secrets when you tell someone in which register you have to write a value to achieve something. You can't tell me that these specs will really tell your competitors much about the internals of the hardware. Now I only got to hope for Intel releasing some add on card with free drivers since they seem to be the only real alternative...

Indeed. I do remember stating at the time AMD took over that I didn't see them opening any info anytime soon... Keep in mind that while you might get specs, they will not easily enable you making a driver. These days, it's not some custom ASIC with fixed functionality rendering operations- it's a full-fledged stream processor. While I have high hopes for TG and the Intel drivers, until we see something resembling the on-paper capabilities of the X3000, we're not QUITE ready for the AMD or NVidia stream processor information.

Regardless of the aforementioned statement, Intel seems to be shaping up nicely all the same. I'm probably going to get one of the 965G motherboards shortly for testing work for LGP so I've got another reference platform (Maybe two- that way I can have an open one with the Intel Drivers and one that switch-hits between providing an NVidia and AMD PCI-E test machine. I'm really needing several upgrades since it's been a bit since my last cycle... )

If Larabee ends up being usable and they do the right thing (If the rumors are true, it'd be silly for them NOT to do the right thing for us and themselves at the same time...) it's going to get interesting real quick.

[Synergy6]

I can understand not opening the Xxk series, but I would have thought the previous ones (X800, 9700 etc) could be opened with little competitive disadvantage. A lot of people run Linux with such cards, and would be appreciative of the full compatability when choosing their next card, even if it had lesser support.

[yoshi314]

i read somewhere about a guy who said he had worked for a while in nvidia or ati. he claimed that graphics drivers are written in such a way that makes them very difficult to maintan. the code is full of hacks, and often relies on undocumented platform quirks to behave faster.

wonder if that's what nvidia guy meant by driver being hard to write :]

also ati some time ago gave out a statement about opening the drivers - they said they would not because of something along the lines of "patented optimization algorithms" :] (i believe it was sometime around the time of amd's takeover).

[Michael]

Originally Posted by yoshi314

also ati some time ago gave out a statement about opening the drivers - they said they would not because of something along the lines of "patented optimization algorithms" :] (i believe it was sometime around the time of amd's takeover).

The quote at Linux.com like back in August from Portland? What I believe that portion comes down to is that they have purchased / licensed code in the drivers from third parties, and part of the terms were that they would not give away that licensed code. It would be like a vBulletin owner giving away the code and allowing other non-subscribers to use it.

[yoshi314]

if i could only find that link :/ i don't really remember. it could be from linux.com indeed.

i wonder why ati can't just open the parts of the drivers they have rights to, and leave the 3rd party stuff in a blob.

you may have heard about xara lx - the company decided to leave its core graphics library as a blob (because it's high performance gives them certain advantage), and opensource the rest of the application under non-windows systems.

now people started porting xara to cairo to make it independent from the blob library. the original authors of xara lx have nothing against it, as long as it does not get ported to windows (that was the condition when they opened up the code).

why can't ati do the same? they could leave all the problematic stuff in a smaller blob, and open what they can. it would just be a matter of time when somebody would start developing an open replacement to those blobs.

i wonder what could that 3rd party code in the ati driver be. only drm solutions like hdmi come to my mind.

[Michael]

Was it this link: http://www.linux.com/article.pl?sid=06/10/11/1355201

[yoshi314]

i meant this one when i thought about 'patented optimizations'. it was probably repeated on other sites as well. don't remember on which one i read it originally, though.

http://news.com.com/2061-10791_3-6104655.html

[Svartalf]

Originally Posted by Synergy6

I can understand not opening the Xxk series, but I would have thought the previous ones (X800, 9700 etc) could be opened with little competitive disadvantage. A lot of people run Linux with such cards, and would be appreciative of the full compatability when choosing their next card, even if it had lesser support.

In reality, anything in the R300, R400, and R500 series of chips are going to be intrinsically identical at the programming level, with each chip in question being a scaling up of the R300 architecture or a process improvement. If they revealed the details of the X800 or 9700, you might as well be opening the whole kimono for them.