AMD Dropping R300-R500 Support In Catalyst Driver

[sundown]

For me it's also a big issue. I think ATI should concentrate on providing powercontrol in open source drivers especially when they dropped support for X1600 in fglrx, because nobody will switch to open source drivers when there is a possibility to burn own graphic card.

Well, I still haven't fried my mobility one with the open driver for more than 8-9 months now. Although, if it burns out, that'll be a good reason to move forward

[fermulator]

Well this entire talk has been extremely informative (less of course those people simply 'whining' about the decision and not providing any sort of constructive feedback).

My first reaction to the entire thing was negative (before actually reading all the information provided in these 32-some pages of posts). The timing is unfortunate, but ultimately I'm glad that ATI/AMD has resolved to releasing the hardware specifications/documentation last year, and that *hopefully* proper open source Radeon drivers will exist in the near future. The decision to mark cards as "legacy" is inevitable. But the means to which ATI/AMD will continue to provide support is most excellent, and should really give ATI/AMD an edge compared to nVidia if we get a rock solid open source driver out of it, something that nVidia does NOT have.

I can certainly feel for everyone who thinks "I purchased a card, the drivers should just work!" -- which is something I agree with to a certain extent, but there are limits to what a manufacturer can provide (for obvious financial reasons), ATI/AMD can't afford to support every single different Linux distro under the sun, because there's a LOT. When you bought your card, did it have a LINUX LOGO on the box? None of mine did. So expecting the hardware to work flawlessly in Linux is a little presumptuous (even if ATI/AMD have been promising for a while that Linux will work..) Indeed, Windows "just works", that's because (as I've read in this thread) the architecture doesn't change as much as the Linux kernels and X server versions change, plus of course the user base for Windows is huge compared to Linux/Mac.

I think ATI/AMD's choice here is the right one. The best way to ensure working ATI graphics cards is to focus on the open source driver, and as Mr. Bridgeman has stated, a driver that theoretically would have support "forever".

On to some more particulars...

I currently own several ATI cards:
* ATI Radon Mobility 7500 RV200
* ATI Radeon 9800 Pro R350
* ATI Radeon X800 R420
* Sapphire ATI Radeon X1600 Series R530/R535
* HIS ATI Radeon X1950 Pro Ice3 Turbo RV570

My distro of choice is Ubuntu for it's stability, and HOPEFUL appeal as a "Windows Replacement" for many people. Currently I have various PCs with 8.04 (Hardy), 8.10 (Intrepid), and most recently, 9.04 (Jaunty).

On my laptop (the mobility Radeon 7500 series), I've always used the open source driver since it's very old. Compiz and that jazz works.

For the rest however, the past 1-2 years I've had to use the fglrx driver because the open source Radeon was never up to par. Now however, with the recent events, and especially the incompatibility between proprietary fglrx and latest X server, I've given the Radeon driver another try. Low and behind, my first impressions are good!

The Test:
* Radeon X1950 RV570 Card
* Dual Monitor Output (1680x1024 each)
* Compiz Enabled

This is my main rig, of which I 'expect' dual monitor output to work, 3D graphics, compiz, and TV output. But I try to be patient ;-0

So far (for the last day or so only), the system is stable, and haven't had a single crash as of yet. glxgears is a little 'buggy' if you try to move the window around, but it runs. Compiz is enabled with dual monitor output configured with the gnome-display-properties thingy.

Really looking forward for the coming months to hopefully see some solidification on the drivers.

Couple of points @ Mr. Bridgeman if I may:

1. Catalyst Control Center: Having a "Simple" point and click way to enable and disable, clone and extend, monitors, and TV out is incredible. For me, this is a huge "loss" by leaving the fglrx driver set to the open source. Will the open source drivers ever have something similar? I know in gnome there is the "gnome-display-properties" which I can only assume uses xrandr for functionality... but when you compare the CCC to this, the gnome-display-properties GUI barely works, and certainly doesn't compare in overall functionality.

2. Full 3D Rendering and Acceleration: I've heard you say that "this is in the pipe", and will hopefully achieve 90% of the fglrx capabilities. Yay! What does this mean exactly though? When the Radeon driver reaches 90%, ATI/AMD will no longer be paying you guys to finish it off, and the open source community will have to finish it off? ... or ... is it more of a 'not worth the effort' last 10% ... ? Curiuos ;-0

3. Wine: Ultimately, I'm a gamer (even if I don't play games as often anymore, heh). I do not expect the latest and greatest games to run in Wine on Linux. But I was HOPING that we could at least run older and 'recent' games to run in WINE. Examples (Starcraft, Warcraft III, UT2K4, Team Fortress 2, etc.). In my experience, these worked with the proprietary fglrx drivers for a while before 9.x Catalysts came out, but then WINE had some updates and nothing worked since. Posts 207 and 205 show some decent remarks on this. Can we expect WINE functionality from all these cards marked as "legacy"? ... or will it just never work? Am I doomed to dual boot to Windows for gaming forever? ;-0

@ The Community In General (and Michael)
* Testing: You posted in post 93 that there would be something "nice and easy" for users to assist in testing. Any pointers on how we can contribute?


Thanks again everyone, and especially thank you to Bridgeman. It's really impressive to see an active ATI/AMD rep on these forums helping to answer people's questions.

[Michael]

@ The Community In General (and Michael)
* Testing: You posted in post 93 that there would be something "nice and easy" for users to assist in testing. Any pointers on how we can contribute?

http://www.pts-linux-live.com/

To get involved, subscribe to the mailing list to find out more information as its developed - http://phoronix-test-suite.com/mailm...test-suite.com

[bridgman]

1. Catalyst Control Center: Having a "Simple" point and click way to enable and disable, clone and extend, monitors, and TV out is incredible. For me, this is a huge "loss" by leaving the fglrx driver set to the open source. Will the open source drivers ever have something similar? I know in gnome there is the "gnome-display-properties" which I can only assume uses xrandr for functionality... but when you compare the CCC to this, the gnome-display-properties GUI barely works, and certainly doesn't compare in overall functionality.

Not sure here. I think the community preference would be to see the standard desktop tools extended rather than having vendor-specific solutions, but one of the issues may be how much is feasible with a standrad tool.

My assumption is that anything which can be done through RandR could be done by a generic configuration tool, and I *think* that pretty much everything you mentioned is covered by RandR. If that is the case, then I guess we just need to get the G team and the K team competing on this

2. Full 3D Rendering and Acceleration: I've heard you say that "this is in the pipe", and will hopefully achieve 90% of the fglrx capabilities. Yay! What does this mean exactly though? When the Radeon driver reaches 90%, ATI/AMD will no longer be paying you guys to finish it off, and the open source community will have to finish it off? ... or ... is it more of a 'not worth the effort' last 10% ... ? Curiuos ;-0

The various % numbers floating around are related to performance, not capabilities. I think most of the devs expect that the open drivers will get pretty close to the capabilities of proprietary drivers in terms of OGL extensions reasonably soon, while the *performance* of the open source drivers will probably lag behind the proprietary drivers when running with complex workloads.

The issue is simply that proprietary drivers can use the same code across 100% of the market (and spend a corresponding amount on development) while the open source drivers (or any Linux-specific driver, whether open or closed) only addresses the Linux market share and can only afford a proportionally sized development effort.

The good news is that 2% of the development effort can get you a *lot* more than 2% of the performance... our guess is that you'll see maybe 90% on simple workloads and 60-ish % on complex workloads once the transition to Gallium3D is completed. If LLVM works as hoped (which I personally doubt) or if another effective shader compiler solution is implemented then the performance on complex workloads might get to 80% or higher.

These are all guesses though - the key point though is that once performance gets above a certain level most users won't notice or care, and that level is probably closer to 50% than you might think. Modern GPUs are getting faster and cheaper at an alarming rate.

I should mention that we are just one part of the development effort, and that most of the performance-related improvements are coming from other members of the development community, whether from developers working for major distros or from independent devs. Volunteer developers still make major contributions to the entire Linux/X/DRI system. We are focusing mostly on adding support for new GPU generations so the other devs can spend more time improving the overall driver framework.

You *can* thank Alex for the EXA and Textured Video code though Um... and power management, and initial tear-free support, and...

3. Wine: Ultimately, I'm a gamer (even if I don't play games as often anymore, heh). I do not expect the latest and greatest games to run in Wine on Linux. But I was HOPING that we could at least run older and 'recent' games to run in WINE. Examples (Starcraft, Warcraft III, UT2K4, Team Fortress 2, etc.). In my experience, these worked with the proprietary fglrx drivers for a while before 9.x Catalysts came out, but then WINE had some updates and nothing worked since. Posts 207 and 205 show some decent remarks on this. Can we expect WINE functionality from all these cards marked as "legacy"? ... or will it just never work? Am I doomed to dual boot to Windows for gaming forever? ;-0.

There are two main factors here. One is the degree of OpenGL extension support - many 3D games running on Wine require OGL 2.x functionality from the underlying GL driver. The other is that Wine was primarily developed on NVidia hardware and probably still has some dependencies on the specifics of their driver implementation (eg memory management).

The open source drivers are moving fairly quickly towards GL 2.x support and will probably get "close enough" once Gallium3D becomes broadly used. I'm confident the Wine devs will take care of the rest. Both will take time, of course.

[monraaf]

1. Catalyst Control Center: Having a "Simple" point and click way to enable and disable, clone and extend, monitors, and TV out is incredible. For me, this is a huge "loss" by leaving the fglrx driver set to the open source. Will the open source drivers ever have something similar? I know in gnome there is the "gnome-display-properties" which I can only assume uses xrandr for functionality... but when you compare the CCC to this, the gnome-display-properties GUI barely works, and certainly doesn't compare in overall functionality.

I have to disagree here. Amdcccle is an ugly (probably Qt) tool, with AFAIK only some extra options for OpenGL settings, but I could not even find an option to rotate the display, and changing screen resolutions and back is buggy.

The gnome-display-properties program is simple to use and works beautiful here with the open source drivers.

[etnlWings]

Well this entire talk has been extremely informative (less of course those people simply 'whining' about the decision and not providing any sort of constructive feedback).

Since when is, "whining", about dropped support not constructive? Customer feedback is only useful when it's positive?

I can certainly feel for everyone who thinks "I purchased a card, the drivers should just work!" -- which is something I agree with to a certain extent, but there are limits to what a manufacturer can provide (for obvious financial reasons), ATI/AMD can't afford to support every single different Linux distro under the sun, because there's a LOT.

This is a strawman and non-sequitur.

When you bought your card, did it have a LINUX LOGO on the box? None of mine did. So expecting the hardware to work flawlessly in Linux is a little presumptuous (even if ATI/AMD have been promising for a while that Linux will work..)

You just undermined your own point, unless you're seriously suggesting 'a sticker on the box' (or lack there of) vetoes statements from the manufacturer.

expecting the hardware to work flawlessly in Linux

Another strawman. I'd imagine people on these boards would know better than to expect flawless hardware support on Windows, let alone *nix.

[fermulator]

@bridgman:

Not sure here. I think the community preference would be to see the standard desktop tools extended rather than having vendor-specific solutions, but one of the issues may be how much is feasible with a standrad tool.

My assumption is that anything which can be done through RandR could be done by a generic configuration tool, and I *think* that pretty much everything you mentioned is covered by RandR. If that is the case, then I guess we just need to get the G team and the K team competing on this

Interesting points there. Indeed, I suppose the general consensus would be to improve the standard tool. I've no idea the feasibility though of having "custom" options available to the various different brands of graphics cards (ATI/nVidia/Intel/etc.)

more talk on this below with monraaf.

The various % numbers floating around are related to performance, not capabilities. I think most of the devs expect that the open drivers will get pretty close to the capabilities of proprietary drivers in terms of OGL extensions reasonably soon, while the *performance* of the open source drivers will probably lag behind the proprietary drivers when running with complex workloads.

The issue is simply that proprietary drivers can use the same code across 100% of the market (and spend a corresponding amount on development) while the open source drivers (or any Linux-specific driver, whether open or closed) only addresses the Linux market share and can only afford a proportionally sized development effort.

The good news is that 2% of the development effort can get you a *lot* more than 2% of the performance... our guess is that you'll see maybe 90% on simple workloads and 60-ish % on complex workloads once the transition to Gallium3D is completed. If LLVM works as hoped (which I personally doubt) or if another effective shader compiler solution is implemented then the performance on complex workloads might get to 80% or higher.

These are all guesses though - the key point though is that once performance gets above a certain level most users won't notice or care, and that level is probably closer to 50% than you might think. Modern GPUs are getting faster and cheaper at an alarming rate.

I should mention that we are just one part of the development effort, and that most of the performance-related improvements are coming from other members of the development community, whether from developers working for major distros or from independent devs. Volunteer developers still make major contributions to the entire Linux/X/DRI system. We are focusing mostly on adding support for new GPU generations so the other devs can spend more time improving the overall driver framework.

You *can* thank Alex for the EXA and Textured Video code though Um... and power management, and initial tear-free support, and...

Thanks for the plethora of info here, this was very informative and helpful. I assume that a "complex workload" would qualify as "gaming"? ;-)

It sounds like though, that if performance might not ever match a proprietary driver's performance, then what kind of "bottom line" are we looking at? I don't really know the status for nVidia (which is a default comparison for us I suppose), but if performance doesn't come close, will we ever be able to call "Linux & ATI" a comparable gaming platform (compared to, say "Windows & ATI"/'Windows & nVidia", or even "*nix & nVidia")? (This question is asked with the assumption that we ignore the obvious performance loss in WINE, and consider only "native" Linux games, e.g. UT2K4)

There are two main factors here. One is the degree of OpenGL extension support - many 3D games running on Wine require OGL 2.x functionality from the underlying GL driver. The other is that Wine was primarily developed on NVidia hardware and probably still has some dependencies on the specifics of their driver implementation (eg memory management).

The open source drivers are moving fairly quickly towards GL 2.x support and will probably get "close enough" once Gallium3D becomes broadly used. I'm confident the Wine devs will take care of the rest. Both will take time, of course.

Well this is good to hear too. At least once the OGL extensions are implemented, we'll be able to report bugs to the WINE devs.

Is there any idea of ETA for all the Gallium3D action?

Thanks!

----------------------------------------------------------------------

@monraaf:

I have to disagree here. Amdcccle is an ugly (probably Qt) tool, with AFAIK only some extra options for OpenGL settings, but I could not even find an option to rotate the display, and changing screen resolutions and back is buggy.

The gnome-display-properties program is simple to use and works beautiful here with the open source drivers.

I suppose it isn't the prettiest thing, but had you tried the latest version of amdcccle with 9.3 Catalysts? I found that it's functionality had improved significantly. Maybe I preferred it simply due to "Windows CCC" experience, and it was what I was used to as far as 'how to change video settings'.

The gnome-display-properties is pretty simple, but maybe that's the problem for me. All it can do is basic stuff. There's no advanced config options for your graphics card, AFAIK it's unable to configure TV out either?

As Bridgman said, not sure if it's feasible to have more 'advanced' options in there that are particular to the Radeon(HD) series since it's meant to be more 'generic'...

I created a thread on Ubuntu forums to hopefully start addressing this question: http://ubuntuforums.org/showthread.php?t=1151881

----------------------------------------------------------------------

@etnlWings:
.. not sure why you're attacking me here, since I thought my post was mostly useful by providing my own comments on the topic, but also asking relevant follow-up questions ... but I'll attempt to reply.

Since when is, "whining", about dropped support not constructive? Customer feedback is only useful when it's positive?

I made this statement simply because it seemed like a lot of people replied and "whined" without doing much research on the matter. It seemed apparent to me that much of this sort of "feedback" was given solely on the first post/topic, and there was no care given to reading the entire discussion (which is something I spent several hours reading). If one actually reads the entire discussion, it becomes much easier to understand (as Bridgman has already stated) that there are two separate issues being discussed simultaneously. Most importantly to this point, that ATI/AMD has not dropped support, but merely shifted it from a proprietary fglrx driver, to the open source Radeon driver.

This is a strawman and non-sequitur.

My statement there was by no means a strawman fallacy, perhaps though it was slightly non-sequitur ;-0. I was merely trying to indicate that I sympathize (feel) for all those that "just want it to work" because I'm in the same boat, like most others. Indeed, it would be nicer if stuff just worked, nicer for everyone. Again though, if we read this entire discussion, 'market shares', 'man power', 5-7 years driver support, and all of that has been touched upon, which explains the decision. (My apologies for the bold though, it came out as harsh instead of as an emphasis.)

You just undermined your own point, unless you're seriously suggesting 'a sticker on the box' (or lack there of) vetoes statements from the manufacturer.

All my point here is that upon purchase, there was no indication "on the box" that this product works in *nix. It shows a "Windows" logo, so we should certainly expect 100% Windows compatibility, but without more information, we can only *hope* for *nix compatibility. This is of course assuming that prior to purchasing an ATI product, the consumer didn't do any research and thus not finding such "promised statements" from ATI saying something to the effect of "we're working on Linux drivers, soon, soon, they'll be ready soon".

Indeed though, ATI has been promising such a thing for years, and we all know how horrible it is to continue to be put on the back burner with *nix drivers ... I'm just glad that open source drivers are getting attention from ATI GPU devs. We're all hoping that it will be a positive outcome. *crosses fingers*

Another strawman. I'd imagine people on these boards would know better than to expect flawless hardware support on Windows, let alone *nix.

Ah, here's an actual strawman, you're right there, my bad. Most people here probably don't expect flawless support on *nix, but we all hope for some support. Still though, people DO expect flawless hardware support on Windows (baring generally tolerated hiccups here and there). In the general sense, and in my own past experience with Windows (for the past 10 years we'll say), and in almost all different hardware configurations, my graphics hardware has worked flawlessly in Windows. But hey, maybe my experience is unique and there have been (maybe still are) problems with GPU's + Windows?

----------------------------------------------------------------------

Anyway, I digress, especially since this post is huge. ;-)

[nanonyme]

Then again, someone could say that the people with older "abandoned" cards are getting the best out of everything since KMS + memory manager work in the open drivers is getting very far for them which'll mean working open DRI2. (also fglrx *again* managed to reach a situation where it's dragging behind in kernel versions which isn't really surprising considering how rapidly the kernel changes - guess who has no 3D atm?) I'd personally be happy for myself if the support situation was the other way around.

[bridgman]

I assume that a "complex workload" would qualify as "gaming"? ;-)

Not really. I'm thinking of the most demanding Windows games really, where shader utilization and ability to operate under extreme memory pressure are key to performance. I would not class most of the native Linux games as a complex workload in that sense.

If the rendering operations are primarily limited by fill rate (which is still pretty common) then the open source drivers will probably get very close to the proprietary drivers in performance. Note that fill-rate limited apps also often perform just as well on older cards as they do on newer cards. I think it will all work out.

It sounds like though, that if performance might not ever match a proprietary driver's performance, then what kind of "bottom line" are we looking at? I don't really know the status for nVidia (which is a default comparison for us I suppose), but if performance doesn't come close, will we ever be able to call "Linux & ATI" a comparable gaming platform (compared to, say "Windows & ATI"/'Windows & nVidia", or even "*nix & nVidia")? (This question is asked with the assumption that we ignore the obvious performance loss in WINE, and consider only "native" Linux games, e.g. UT2K4)

Not sure I understand the question. If getting every last bit of gaming performance is top priority then I imagine the user would be running a relatively recent graphics card and the proprietary drivers. We expect the open source drivers to satisfy most users, but not all.

It's really 6xx and above where the shader compiler and other components in the proprietary drivers are key to maximum performance, since that is where the VLIW shader core was introduced (sometimes referred to as superscalar). If you have a sufficiently smart compiler the 6xx-and-up shader core allows you to pack a number of simple operations into a single instruction and significantly improve shader performance. The older chips basically have vector processors with a single opcode per instruction (not quite, but close enough) while 6xx and up allow 5 independent opcodes per instruction.

[etnlWings]

.. not sure why you're attacking me here

I wasn't. It was really only those small parts of a much larger post that I took objection to. I don't abide by people purporting objectivity, while taking little fallacious pot-shots at one side of the debate. Although perhaps I've overestimated you, and your inflections are merely misleading.

Most importantly to this point, that ATI/AMD has not dropped support, but merely shifted it from a proprietary fglrx driver, to the open source Radeon driver.

This would be spin. Might be a valid argument if all the models recently dropped currently had a decent level of support in the open driver. To follow your lead: reading this entire thread, it's easy to see that for many people, this is a serious backwards step, which has left them with little practical alternative other than to upgrade their hardware (which isn't possible for all) or not upgrade their OS, until such time as someone does provide solid support. It takes a fair stretch for full hardware documentation to count as, "support", all by itself. Although this is resting on the assumption that no one from ATI had been providing commits to the open drivers, in order to bring them up-to-spec, in anticipation of this decision. I may be completely wrong.