AMD Dropping R300-R500 Support In Catalyst Driver

[RealNC]

It still seems progress is slow though. Do the devs working on then do so exclusively, or is it just one of the projects they're working on?

[Saist]

here's what I wrote on MepisLovers about this

March will be the last month that the R300 architecture, launched back in 2003 with the Radeon 9700, will be supported under AMD's Fglrx releases. As of the April Catalyst 9.4 set, all of AMD's OpenGL 2.0 cards will be removed from support, on both the Windows and Linux drivers. (so no, this isn't AMD picking on Linux, Windows is losing support too). These are all graphics cards from the Radeon 9500 - 9800 series, x300-x800 series, and x1x00 series.

Phoronix goes over how the news affects upcoming distributions : http://www.phoronix.com/vr.php?view=13559

The bad news is that AMD has no plans to update the final 9.3 driver to work against future X.org servers, a practice Nvidia carries out on the 71.86.xx, 96.43.xx, and 173.14.xx driver sets.

The good news is that improvements alread in development for the X.org Mesa drivers will probably bring the 3D support for R300-R500 GPU's in within 10% or so of the existing Fglrx driver, and 2D support is in many cases, faster on the existing Open-Sourced Drivers.

The other good news is that the decrease in product support will mean fewer regressions, and should simplify driver development for the RadeonHD series.


***

Okay... my take. I'm not really sure what to think. I'm a little uncomfortable... okay, a lot uncomfortable with the UAE taking an interest in AMD. If this change in support hadn't occured so shortly after the board of directors changed up, I'd be a little bit more willing to just pass it on as standard obselence.

Phoronix also makes mention that the hours had been cut on RadeonHD developers. However, what isn't mentioned is whether or not these shorter hours are by the request of AMD to Novell, or by Novell directly. What some people have also noted is that the X.org ATi driver has seen faster development than the RadeonHD driver, mostly due to Novell's approach being based on a complete reverse-engineering of the hardware and registers, rather than by using AMD tools such as AtomBIOS. Last year though, the RadeonHD developers pretty much gave in and started to use AtomBIOS : http://www.phoronix.com/scan.php?pag...atombios&num=1

Only time will tell if the new AMD after the split off of the Foundry, and the loss of Hector Ruiz to the new Foundry Company, will continue to support Open-Source... I'm just hoping that some of the moves over the past couple of weeks are just knee-jerk reactions by Novell and not indicative of a change in AMD...

otherwise, Intel's Larrabee.. and yes, this is ME saying this... Intel's Larrabee suddenly just a heck of a lot more attractive.

[sreyan]

@saist: RadeonHD also uses atom bios now for some cards

Honestly i assumed fglrx was not targeted at individual users but at corporations that need fast OpenGL support for workstation use.

Those corporations don't download the newest distro every six months. They use a stable old release. And when they purchase new hardware, they'd like that to get all of the attention. It's better ROI for AMD's driver budget in a lot of ways.

I've always been much happier with the quality of the open source drivers. Of course the performance isnt as high, but if that's the concern you can profile the problem and fix it yourself. It's not like you're being left in the cold like the NVIDIA users were when 96 stopped working with the latest xorg and 173 required SSE. At least the OSS ati drivers have more mature 3d.

[Saist]

@saist: RadeonHD also uses atom bios now for some cards

I know. I said that.

What some people have also noted is that the X.org ATi driver has seen faster development than the RadeonHD driver, mostly due to Novell's approach being based on a complete reverse-engineering of the hardware and registers, rather than by using AMD tools such as AtomBIOS. Last year though, the RadeonHD developers pretty much gave in and started to use AtomBIOS : http://www.phoronix.com/scan.php?pag...atombios&num=1

[sreyan]

I know. I said that.

yes you did apparently i can't read.

[yotambien]

I've always been much happier with the quality of the open source drivers. Of course the performance isnt as high, but if that's the concern you can profile the problem and fix it yourself. It's not like you're being left in the cold [...]

Really? What button do I press to fix them?

At least the OSS ati drivers have more mature 3d.

What's 'more mature'? And 'more mature' than what?

[bridgman]

It still seems progress is slow though. Do the devs working on then do so exclusively, or is it just one of the projects they're working on?

All three of our devs are full time on open source drivers, although Cooper's time is being split between new GPU support and ongoing support of Linux server OEMs with older GPUs (typically RV100-class parts).

The bigger issue is that most of the "thousand lines of code" work has been done and now we're spending more time in the "million lines of code" sections of the driver stack.

The other issue is that most of the community developer effort has shifted from fixing and tweaking the current drivers to implementing the next generation driver stack (KMS, GEM, DRI2, Gallium3D etc..). There's a ton of work being done but from an end-user perspective all you see is that the level of bug fixing seems to have dropped off.

The reality is that most of the remaining bugs aren't going to be fixable with tweaks and minor changes, and the dev focus is on the core issues that will provide more stable and performant drivers going forward.

[Saist]

All three of our devs are full time on open source drivers, although Cooper's time is being split between new GPU support and ongoing support of Linux server OEMs with older GPUs (typically RV100-class parts).

The bigger issue is that most of the "thousand lines of code" work has been done and now we're spending more time in the "million lines of code" sections of the driver stack.

The other issue is that most of the community developer effort has shifted from fixing and tweaking the current drivers to implementing the next generation driver stack (KMS, GEM, DRI2, Gallium3D etc..). There's a ton of work being done but from an end-user perspective all you see is that the level of bug fixing seems to have dropped off.

The reality is that most of the remaining bugs aren't going to be fixable with tweaks and minor changes, and the dev focus is on the core issues that will provide more stable and performant drivers going forward.

Okay. I think I'm taking this seriously out of context... but that really read something to the effect of: If Novell doesn't follow through on RadeonHD, our own guys are still writing the other driver.

I guess from my point of view, the loss of Novell as a backer really doesn't seem that important. As I understood the AMD strategy, the Novell developers just got an early lead on the same documentation everybody else would get at http://x.org/docs/AMD

[tball]

@Bridgman
Everytime I read something you post, I like what I hear. Really enjoy an Ati employed, posting in this forum.

You probably won't answer this, but I'll try anyway. Can you give a hint, how long the progress is with working 3d drivers for > R500?

[bridgman]

Okay. I think I'm taking this seriously out of context... but that really read something to the effect of: If Novell doesn't follow through on RadeonHD, our own guys are still writing the other driver.

Other driver... you mean -ati aka radeon ? We have been adding new GPU support to both drivers and I expect that will continue.

The bigger issue here, I think, is that most of the hard work on the DDX drivers (radeon and radeonhd) has been done other than bug fixing and ongoing addition of new GPU support. Most of the development efforts now are on the 3D side, in the drm and mesa drivers, and a lot of effort is going into moving functionality *out* of the DDX drivers.

There was an interesting milestone last night, when MostAwesomeDude got X running over the xorg state tracker, in turn running over Gallium3D (ie Gallium3D doing the basic 2D acceleration), running over KMS, GEM and DRI2 on an RV410. I don't get surprised often but that sure surprised me.

You probably won't answer this, but I'll try anyway. Can you give a hint, how long the progress is with working 3d drivers for > R500?

We're still hoping to have rough parity with 5xx by the end of April. We have glxgears running in house -- it still has some problems but I think we know the cause for each of the problems. Richard and Cooper are working on texture support now.