AMD: Accelerating Open-Source Drivers?

[phoronix]

Phoronix: AMD: Accelerating Open-Source Drivers?

So far today at Phoronix we've published the AMD 8.41 Display Driver Preview, ATI R300/400 Linux Performance, ATI R500 Linux Performance, and ATI Radeon HD 2900XT Linux Performance -- yet, it's only ten minutes past midnight! However, looking at AMD's official statement on the Linux driver, there is a rather interesting sentence about good things to come...

http://www.phoronix.com/vr.php?view=10977

[miltiad]

Can't wait !

Keep up the good work Phoronix guy(s?) !

[schestowitz]

Hey guys,

You can help by Digging this (if you have a Digg account and only if you think it's Digg-worthy).

http://digg.com/linux_unix/Breakthrough_ATI_AMD_Driver_Has_Just_Become_up_to_ 10_Times_Faster

It seems like it'll make it into the front page.

Thanks!

Roy

[hmmm]

I just hope when the code is released, it's in a form that can be easily be integrated with current open source drivers. The recent work being done on AVIVO and the ati open source drivers is very good.

[chrisr]

That's the obvious inference, if AMD would rather concentrate on the R500/600 chips these days. I am assuming that R300/400 is purely "legacy" hardware now.

[R3MF]

there is now a chance that i will purchase an ATI/AMD video card.

i have always preffered AMD and I have wanted to purchase ATI since the acquisition, but found myself totally unable to do so in the face of awful linux drivers.

i know the hardware is equivalent to that of nVidia, so with the driver issue out of the way i may well be tempted to buy ATI vid-cards in future.

[yoshi314]

I am assuming that R300/400 is purely "legacy" hardware now.it will be legacy when more people will have r600/r500 than r300/r400. currently it's far from it.

[Michael]

AMD's efforts are indeed focused on the R600 (and R500) series.

[chrisr]

AMD's efforts are indeed focused on the R600 (and R500) series.

So it would make good business-sense for AMD to offload the on-going driver support for any R300/400 based hardware to the Open Source community, and then delete that support from the shiny, new fglrx code-base?

[rolz]

it's nice that you guys know.

But i feel like a horse running after a carret suspended 10 feet ahead from a fishing pole.

It would be epic if it's an opensource hardware accelerated 3D driver, but then again, maybe we will just get 7 more slides about the release train ...

I'm just impatient, have struggled quite a bit... currently have no 3D on xorg 7.3.. and even building it against my current kernel is a challenge. o and that control pannel has a nice feature of segfaulting

But if I would have to make a wish list:
- opensource (i'm not too fanatic on this one, but would be brilliant, no more binary patching away watermarks or other crashes)
- aiglx support
- performance improvement
- when loading display the text "rolz is great" in big golden letters instead of logo's

I'd say there allowed to drop one of those, and i would still be happy ;)

[Michael]

So it would make good business-sense for AMD to offload the on-going driver support for any R300/400 based hardware to the Open Source community, and then delete that support from the shiny, new fglrx code-base?

No, they will not be discontinuing R300/400 support with this new driver.

[ivanovic]

I have a dream of a world without a need to check if your graphics driver will work with the new kernel, because it is directly included in the kernel.
I have a dream of being able to upgrade my xorg server without first having to check if all the drivers will make the transition...
I have a dream of really good randr support no matter if I want to use binary blobs or not.
I have a dream where hardware vendors would directly hand over the specs of the hardware to a customer, so that he can code his own drivers if he has the knowledge to do that.

I hope AMD/ATI will help to make at least some of those dreams come true. At least the first dreams are already possible when using open source drivers, so a big company like AMD/ATI should be able to manage this, too, since there people do have the specs, the knowledge and the money to concentrate on driver development where open source devs atm have other things to do for a living, concentrate on reverse engineering to get some specs and only after that can work on the drivers.

After all the things you could read in Michaels posts in the last days/weeks there probably is something nice to come from that announcement. I would suspect that it could be at least about specs for the r300/r400 series up to specs for all the cards available including open drivers and the ATI/AMD devs working together with the radeon xorg devs on getting good open drivers. I hope they take Intel as example where the people from Intel directly work on the opensource driver.

Some things that made it appear to me like something like this is possible:
* Before ATI did state they could not open their drivers due to intellectual property from other companies being included. The new driver was completely rewritten, this could be taken care of.
* AMDs main competitor is Intel, not nVidia. Intel does offer open drivers and they seem to work on standalone graphics, too. Soo maybe AMD wants to get at least on par with Intel.
* AMD wants a bigger piece of the workstation cake. There often linux is being used. Better and open drivers are a selling argument over there.

Of course there are also some reasons not to open their drivers:
* nVidia does not open their drivers...
(You do see the vicious circle behind this?)

[mlau]

AMD announced that they're going to provide
specs and skeleton driver sources for >=R500 parts:
(and apparently also developer time)

http://lwn.net/Articles/248227/

[Michael]

AMD announced that they're going to provide
specs and skeleton driver sources for >=R500 parts:
(and apparently also developer time)

http://lwn.net/Articles/248227/

That's part of it... Not the complete package though of what they are doing. You'll need to wait for the Phoronix article :)

[Svartalf]

That's part of it... Not the complete package though of what they are doing. You'll need to wait for the Phoronix article :)

No, I want it RIGHT NOW! >:-) :D

In all honesty, I can't wait to see the article coming up. I want to see what they're actually planning on doing- with what HAS been disclosed, while it's not QUITE what I'd hoped for, they now have my full attention- IF they do what we've found out so far from the grapevine or more.

[Source Of Radiance]

The fact that Phoronix has so much info before everyone else has made me register here. Great stuff, and keep up the good work!

[Michael]

The fact that Phoronix has so much info before everyone else has made me register here. Great stuff, and keep up the good work!

Great, glad you enjoyed it. Be sure to stick around as we have many more articles (on ATI and other topics) in the queue! In fact, I think there will be another two to four articles that get published before the driver is even released on their website.

[danij3l]

hi

here is something more on this topic. it seems that AMD will provide some help, but not entire driver.
http://lwn.net/Articles/248227/

[immudium]

Very intriguing stuff here. I have to admit, it would take an awful lot to convince me to make my next video card purchase an AMD card over an NVidia card. A high quality performant open source driver would probably do it, but when Unreal Tournament 3 and Quake Wars roll around, I'm afraid I'll have little patience for ATI's traditionally quirky driver behavior and subpar performance on Linux. That said, I'm hoping for the best and I'll be watching what Phoronix has to say very closely as this is definitely the place to get the best information for both sides of the video card fence.

[Thetargos]

I was toying around with the idea of a high quality Open Source driver derived from full specs disclosure... What would that mean? Boy! Back in the days when I had an ATi card (now I own an nVidia card) and knowing how both companies handle their drivers and devices, I always thought ATi was more close to the true "nature of Linux" in the graphics department at least, by not requiring any special devices, and by using pretty much all the infrastructure already in place (no need to brake your X11 installation with the drivers), that is a much more elegant implementation than what nVidia had (albeit, nVidia offered what users wanted: speed and features).

So sort of "knowing" how are both drivers implemented, I always thought that ATi had a much better chance to actually open at least one part of their drivers: The DRM module, and keep the X11 parts and bits closed, as that's where the real "magic" happens, the DRM module is "simply" a pass-through to the hardware (sorry if I'm over simplifying things here). At any rate, this announcement of openning the specs for some parts of the graphics platform could lead to a high quality Open Source 3D driver, however I wonder what is AMD going to do with technology licensed from third parties, like S3TC and others? I thought they could:


Act as proxy for the Open Source Community to be able to implement these, on the AMD/ATi hardware only, effectively sponsoring the development of the drivers, since they have already paid the licenses for such technology (but that opens up the door for questions such as how "open" would the Open Source drivers be?); or
Support the development of a high quality Open Source 3D driver, stripped from such "licensed" technology (R200 all over again?), and still offer an X11 binary driver with such features, but using a common kernel interface, which would translate into:

Instant new kernel release compatibility.
High quality Open Source 3D drivers missing a few features (I just hope the licensed technology doesn't prove to be performance killer if the drivers are stripped from it).
Ease of deployment (due to kernel improved compatibility)
Better quality drivers, given the cooperation between the company and the community for maintaining common code for both the OSS and proprietary driver.
Happier users :D



That would be a dream come true, but until the final announcement, we can only guess and hope and dream. I'm eager to see what's in store for us!

[hmmm]

It feels like the whole state of ATI drivers on linux has just back flipped completely in less then 24 hours :D - this is awesome news for the AVIVO driver. Too bad it will pretty much obsolete Revenge.

[Michael]

It feels like the whole state of ATI drivers on linux has just back flipped completely in less then 24 hours :D - this is awesome news for the AVIVO driver. Too bad it will pretty much obsolete Revenge.

Not the Avivo driver, but the new yet to be named Linux driver...

[Kano]

Why introduce a new driver and not integrate into current one? Would ease up hw detection - just everything from ATI vendor pciid to ati driver. Currently you need a list with all supported old cards for ati and you use the rest for vesa. When you want to write a list for new driver then good luck... It is enough to update my own pci id list from fglrx driver.

[Michael]

The article will be published at noon EST tomorrow.

[Speedator]

Waht about specifications for R300 and R400? There are open source drivers, but they also need more development(slow, not everything works). Will there be any progress for those, too?

[Thetargos]

If you read the press release for the Kernel Summit, they pretty much stated that they want to support Open Source drivers for across the board ATI products, so most likely, yes, that includes R300/400 products too ;)

[chrisr]

so most likely, yes, that includes R300/400 products too ;)And what about R100/R200 based products? I assure you that these cards are still out there, and are in active use...

[Thetargos]

And what about R100/R200 based products? I assure you that these cards are still out there, and are in active use...
MAYBE they'll merge the current code for R100/200 and the new once they figure out if the new code can drive these cards or if they need some other means to unify them under the same code base, but I'm not sure it'll be a high priority in the devs TODOs, most likely not.