AMD Dropping R300-R500 Support In Catalyst Driver

[MostAwesomeDude]

If you wish to see change in the drivers, consider contributing. Here's what the current devs are up to:
- agd5f is working for AMD, busy fixing DDX bugs and keeping DDX and DRM sane.
- airlied is working for Red Hat trying to get GEM+KMS in Rawhide to be more stable and rewriting the classic Mesa drivers for r100/r200/r300.
- nha is a full-time student.
- MrCooper is working for TG/Vmware on Gallium interfaces.
- MostAwesomeDude (me) is a full-time student.
- marcheu is working on nouveau.
- darktama is working for Red Hat on nouveau.

I think I got everybody that was involved with the classic Mesa drivers...

~ C.

[smitty3268]

I agree with Wyatt. I originally felt this was a bad move by AMD, but I can understand where they are coming from. If the R600+ support really starts improving I could even say I agree with it, but we'll have to wait and see whether that actually happens or not. Some of the posts showing up here are really getting ridiculous. If you are really that offended by the fact that AMD is forcing you to move to the opensource drivers, then I suggest you simply move to NVidia. Deal with it, and quit talking about suing, as if you had some god given right to have the binary driver instead of the open source ones which have at least a basic level of support. God help the guy who accidentally spills some coffee on you....

What gets me is that every month there are a bunch of posters whining about how horrible the new fglrx drivers are, and how they are going to buy an nvidia card because they can't even use their AMD card anymore. Then when AMD drops support for older cards, we get a bunch of people whining about how they can't live without fglrx and that dropping support is going to make them go to nvidia.

I could swear some of the same posters are responsible for both types of posts.

Bridgman, I don't know if you have any say about this or not, but I do think that if you could support your abandoned drivers the way Nvidia does there would be much less fuss over this. Saying that you are doing the same thing they are is very misleading - 1 year from now it will be almost impossible for an average user to get 9.3 running on their new distro, while the nvidia ones will install just fine. I wouldn't think updates for the new X server and kernel versions would be very labor intensive for AMD to support, but I guess I can't say without seeing the code.

[duby229]

Bridgman, I don't know if you have any say about this or not, but I do think that if you could support your abandoned drivers the way Nvidia does there would be much less fuss over this. Saying that you are doing the same thing they are is very misleading - 1 year from now it will be almost impossible for an average user to get 9.3 running on their new distro, while the nvidia ones will install just fine. I wouldn't think updates for the new X server and kernel versions would be very labor intensive for AMD to support, but I guess I can't say without seeing the code.

That is exactly what the open source drivers resolve... Every single issue you just described is adequately resolved using the open drivers. The open drivers will work with every new kernel, and every new X server on release.. You want to look at the code have at it... This is honestly a hell of a lot better then anything nVidia has ever done. I've been very skeptical of ATi's motives in the past, but if things actually go through the way they look like they are going to, then ATi has finally won me over once and for all.

EDIT: and with the inevitable influx of new users will come new developers as well. For an open source project that is the best possible outcome.

[bridgman]

I agree with Wyatt. Bridgman, I don't know if you have any say about this or not, but I do think that if you could support your abandoned drivers the way Nvidia does there would be much less fuss over this. Saying that you are doing the same thing they are is very misleading - 1 year from now it will be almost impossible for an average user to get 9.3 running on their new distro, while the nvidia ones will install just fine. I wouldn't think updates for the new X server and kernel versions would be very labor intensive for AMD to support, but I guess I can't say without seeing the code.

For what it's worth, we're not claiming that 9.3 will keep working after an update of X or kernel versions, and we are not claiming that we are providing the same BINARY DRIVER legacy support as NVidia. What we are saying is that future distro releases will continue to be supported, but for pre-6xx GPUs that support will be via the open source drivers rather than the binary drivers.

The effort required to support a new X or kernel version really is dependent on both the specific version and the number of distro-specific patches each distro includes in their next release. In theory distros just pick up new kernels and new xorg versions, but a number of distros ship components which don't match up with kernel or x server versions. Fedora is a good example; right now a lot of the code shipping in Fedora is newer than what is available upstream.

The other non-trivial source of work is that kernel symbols and interfaces get marked as "GPL Only" on a regular basis, which means that they remain available to open source drivers but can no longer be used by binary drivers. In some ways these are the most burdensome, because they don't represent a typical API update but rather a *shrinking* of the API, ie often no replacement is provided for what was removed.

[Melf]

The radeon driver turns DRI on by default but radeonhd defaults to turning it off.

If you add :

Option "DRI"

to the Device section of xorg.conf that should give you the same performance in radeonhd.

I have tried it:
radeonhd+Option "DRI" in xorg.conf(device section)
-> cca 300 FPS in glxgears
-> xv working, bit more tearing, cpu usage for the same 720p video cca same as in radeon
radeonhd without Option "DRI" in xorg.conf(device section)
-> cca 260 FPS in glxgears with high CPU usage (software rendering ?)
-> xv not working

BTW, in the radeon driver, if I want to watch video using xv in compiz, I have to add:
Option "EXA"
to the device section in xorg.conf, otherwise the video window is just black

EDIT - agd5f just mentioned in IRC that while radeonhd and radeon both have tear-free Xv code for 6xx/7xx, radeonhd probably doesn't have tear-free for 5xx yet.

Yep, i think there is definitely more tearing in xv than in radeon. Thou, even in the radeon driver there is tearing sometimes when things get ultra hectic

[duby229]

The other non-trivial source of work is that kernel symbols and interfaces get marked as "GPL Only" on a regular basis, which means that they remain available to open source drivers but can no longer be used by binary drivers. In some ways these are the most burdensome, because they don't represent a typical API update but rather a *shrinking* of the API, ie often no replacement is provided for what was removed.

Genius...

We need to figure out an adequate method to accelerate that mechanism...

[Louise]

I think someone should start a donation website, where upset users can put money on AMD and nVidia OSS driver developers to speed up the process

But I doubt that even $10000 would change much. What is a GPU programmer paid a month? $10000?

And AMD's got 3 full time? Wow. My hat off to AMD!

I think AMD and nVidia are doing the right thing. I even impressed that they will continue to support legacy hardware.

And as Open Source? Can we really be this lucky?

[Louise]

I wonder how many downloads nVidia have for their legacy Linux drivers.

And what does that make each download cost?

[MostAwesomeDude]

I think someone should start a donation website, where upset users can put money on AMD and nVidia OSS driver developers to speed up the process

But I doubt that even $10000 would change much. What is a GPU programmer paid a month? $10000?

And AMD's got 3 full time? Wow. My hat off to AMD!

I think AMD and nVidia are doing the right thing. I even impressed that they will continue to support legacy hardware.

And as Open Source? Can we really be this lucky?

I'm currently paid much less than $10,000 a month by Oregon State University to evangelize Open Source software in the community. So nine months out of the year X.org work must take a backseat to other work.

That said, bounties for X.org work have been discussed before, and there is a Vacation of Code program in place for getting money to developers who have specific targets to work on. I'm sure this will get brought up at XDS in September.

[gfxdrone]

For about 4 weeks I've been predicting a major move from AMD/ATI re their driver development for Linux systems. Not because of any technical reasons, but for the financial reasons one early poster in this thread mentioned.

AMD is losing market share to both Intel and nVidia last I checked, and that's in a declining market. Not a slowly declining market either, a radically collapsing one. They just dumped their chip fab plant. There is the question of the UAE investor's voices in this. These are the real issues here, obviously, though I haven't really followed AMD specifically that much recently in terms of where they are going financially.

One thing I learned long ago was that engineers who work for companies rarely have much sense of why such decisions really are made, but given this economic reality, and the rapidly collapsing global economy, which is not trivial and is not a joke, AMD has to take action, and this is an obvious cost cutting measure while still doing what they can do in terms of supporting end users. I've read threads here before where bridgeman gave some excellent and very accurate estimates for actual real world Linux market share, so rather than repeat those debates, I'd just point out that the numbers he gave were and are very accurate, and fit exactly with the real data I've seen, so that really has to put what resources AMD can allocate to the Linux driver groups into a context that can't just be ignored as something you don't want to look at or think about, these are real numbers, and AMD knows what they are, as do most other major entities that need such information. So I'm actually kind of impressed by the support AMD is committing to even now, it's pretty decent in my opinion given these economic realities.

I wanted to be mad about this, and I kind of am annoyed because I have to buy / upgrade my current ATI testing card, but that's really the only reason I'm annoyed, this was totally predictable, and I'm glad to see that things are moving more or less as expected in terms of what decisions the AMD corporation is taking in terms of trying to keep things going.

Must be a tough situation bridgeman, good luck, I think in practical terms, this is about as good a solution as you could hope to do in such a serious economic downturn, AMD isn't a charity or non-profit corporation, so you have to allocate your resources as well as you can, and this is a good way to do it.

Obviously, I'd never consider or recommend a high end ATI card based on this support level, but I would never recommend that anyway, so that's not changed, but the radeon/radeonhd stuff is fine for most average users.

One thing I found notably absent in this long thread was actual technical comparisons between the current radeon/radeonhd and fglrx compared to Windows performance, that kind of answers all the real questions in terms of what you really got for your money. I don't have this real data, I'd be interested to see the comparisons between XP/Vista/Linux for various cards and drivers just to see what percent of the full performance each system is going to get, and now is getting.

But free drivers are nice for the long term, and it's about as good as you can hope to achieve at this point, nobody was planning on a global full scale economic collapse, and it's not going to get much better in the near future (and if you think it is, maybe I can try to sell you some stocks and a house or two...), so AMD has to take action, and they have. Good choice I think, too bad for the end users who were expecting full support but I never believed that in the first place so that's what it is, as bridgeman said, they never promised it.

To the posters who said NVidia has dropped support for their cards, what are you talking about, they are running on 2.6.29 and xorg 1.6 except for the very oldest series I believe, and even that might get a new legacy release, I don't know that status. That goes back a long way now, we're talking seriously old hardware that still is running on the latest kernels and xorg versions, though maybe the very oldest stuff will be dropped, but that's very old now, the 400 series is still running, I think earlier is fine too.

And no, it's absolutely correct to point out that should nVidia drop support, all users would essentially be screwed, since there is no free xorg nvidia driver worth discussing, no 3d driver that really works yet, and no specs for the cards. But they have supported, and continue to support, their cards very well with non free drivers, I talk to almost nobody, except apparently a few people in this thread, who regrets buying nVidia, nobody. I know people have issues, that aren't fixed sometimes, but I can speak for a large group of people, and they don't in general have that many issues, and most have none that I am aware of. So that's what it is.

But free drivers are an inherently good and desirable thing, so I really think AMD is doing the best they can do in this situation, mad as it might make some people, for very good reason, but you have to see the bigger picture sometimes, and if it really makes you mad, get nVidia, but if you want 3d free xorg drivers, AMD is an excellent option now.