AMD Dropping R300-R500 Support In Catalyst Driver

[bridgman]

If it's being made to work with lesser performance, then yes, users really will care and have every right to do so.

Adam

Sure, that's an easy case. What if some things are much better and some things are a bit worse ? Again, comparing today's open source driver with June's fglrx is not particularly useful; we need to compare what is likely to happen with both drivers at the same time.

If we thought that today's open source drivers could fully replace fglrx we would have made this announcement last year.

Also, if I believed that our users *really* just want fglrx in its *current* state but working on future X and kernel releases then spending time on that might be an option instead of working to improve the open source drivers, but you *know* that nobody would be happy with the current level of fglrx on 3xx-5xx for more than a month, would they ?

[adamk]

Sure, that's an easy case. What if some things are much better and some things are a bit worse ?

That was the nice thing about the situation. Users had a choice. They could decide which things were more important to them and choose the best driver for their situation

Again, comparing today's open source driver with June's fglrx is not particularly useful; we need to compare what is likely to happen with both drivers at the same time.

If we thought that today's open source drivers could fully replace fglrx we would have made this announcement last year.

Except that without some sort of official timetable or road map, this is still open source development with the credo "It's done when it's done". For all users know, it could still take another 12 or 18 months for 3D performance in the open source driver to reach the state of the fglrx driver now.

Adam

[bridgman]

That was the nice thing about the situation. Users had a choice. They could decide which things were more important to them and choose the best driver for their situation.

True, but unfortunately that option went away when we moved 3xx-5xx to a reduced support model on other OSes. The fglrx driver is our vehicle for delivering proprietary features & performance from a common, cross-OS code base to Linux users. It's not really practical to keep that code base going for a single OS; we could probably keep it running at the current level of functionality on new X and kernel versions but we can do more by putting the same effort into the open source drivers. Doing both is not really an option.

I know we're delivering quarterly updates for Windows but we're not committing to support new OS versions there either.

Except that without some sort of official timetable or road map, this is still open source development with the credo "It's done when it's done". For all users know, it could still take another 12 or 18 months for 3D performance in the open source driver to reach the state of the fglrx driver now.

But there's no official roadmap for fglrx either... and "it could still take 12-18 months for video playback and 2d acceleration in fglrx to reach the state of the open drivers today". Also possible but it's not likely. The entire PC industry is "it'll be done when it's done, but internally we know that we'd damn well better have it done by XYZ date".

You might be reading more into our estimate for ongoing performance delta between open and closed drivers than we ever intended. AFAIK the app issues you see today are generally not a function of driver *performance*, but rather a function of driver *feature* set (primarily higher level GL functions which, when absent, either cause the app to use less performant API calls or more commonly cause a fallback to dog-slow software rendering). The "decently fast hardware acceleration" mixes with "dog-slow software rendering", averaging out to "Meh".

Just to be clear, I don't think the open source drivers will ever reach 100% of fglrx performance in all situations. I also think that 3D on the open source drivers will become fast enough that most users won't care about the difference, while working better in all of the other areas. It may take until the end of the year before the last gamer says "ok, I'm happy and I can finally upgrade to that new distro I wanted to try" but I do believe that even by mid-year more users will be happy with this plan than they would be if we put the resources into fglrx instead.

[b15hop]

What I would like to see is open source drivers with good 3D, hd video playback and desktop performance. (full acceleration of xorg / sdl / opengl etc..) I don't see the point owning a graphics card to improve performance if all that power just sits there like a dormant volcano. I agree with Bridgeman that software drivers are "slow dog" compared to the hardware accelerated equivalent.

[bridgman]

Adam, I probably don't need to say this but I agree with your concerns about performance and I agree that keeping fglrx going alongside the open source drivers would offer more choice.

My only point is that if we *are* going to reduce support on the older GPUs (and we need to reduce support on older GPUs at some point if we are going to keep introducing new GPUs in what is at best a flat market) then supporting those GPUs via the open source drivers is a better use of resources.

[adamk]

My last post on the topic (at least for now)... I just hope you are right about the timing :-)

Adam

[Adarion]

Well I don't fear that much either. Of course, on short terms it's a nasty situation, yes, my X1250 isn't that old yet, I mean I biught the board with the cipset about half year ago. So I have the 6 month and then suddenly no support situation - on the proprietary driver side. And it's also not exchangeable, notebook like situation. But since AMD-ATI made all the specs and stuff accessible it is no problem. It's just a matter of time to have full support by the free drivers. And the x wiki on Radeon/~hd tells me that only few things are missing on that special chip design. Also my Radeon HD 3870 type of chip is in good progress being supported. I guess on long terms with all the specs available there is nothing to fear. That's the good thing about having full specs and free drivers.
And then you can go with latest kernels, xorg-servers and so on. And have all the fancy stuff like kernel modesetting and so on work out of the box.

And - as I see they want to implement full OpenGL 3.0 specifications in the free drivers for the newer cards - that would mean, when it's finished that all OGL stuff is being rendered in hardware (as far as developers think it makes sense to put OGL-function X better to GPU-hardware when it will perform there better than on CPU). Might take some time until all implementation is finished though.
I don't know if any GPU hardware vendor really has OpenGL3 support yet fully implemented in hardware and drivers.

What I'd appreciate is seeing AMD-ATI employ folks like Luc Verhaegen or similar to develop full or half-time on the free drivers. I don't know about the current situation - maybe bridgman could tell us how many developers are working which time on the ati-radeon/~hd drivers. (I tried to ask the AMD folks on recent Chemnitzer Linux Tage (popular German Linux weekend, student friendly prices (e.g. entrance 4$, lunch about 4$), many lectures on various different topics and exhibition part for commercial stuff as well as free projects like distributions, OOo etc.) but they were from the Dresden AMD research center which is dealing much more with CPU related stuff, GCC, virtualization and so on so they didn't know that much about the graphics and chipset side.

[pedepy]

but as long as your laptop works well do you really care ?

you know, I couldn've put it better myself, mr bridgman. Problem is, my laptop does not work well, at all.....

I've just tried to use dual head for the first time since my last upgrade, and for some reason, amdcccle doesn't know what to do with my external hdtv anymore. Not to mention the awful flicker that threatens to break my 1500$ television and the fact that dri somehow 'crashes' and 'recovers' while amdcccle launches.. .... That's when KDE launches at all. Every now and then, I just get a black screen. Or the desktop will just jitter and flicker for minutes on end. When it does work, resizing windows makes my laptop feel like im running a Windows 3.1 port on a 1973 lamp based monster. Again, that's when I'm able to resize at all.

About 'dri2', I guess I was just talking about the whole 'redirected of direct rendered content' thing, which seems to be starting to be making progress in the latest releases, but progress which I will never have a chance to see come to fruition on my laptop, since you are dropping support. I guess I waited 18 months for a decent driver for nothing.

I understand that the linux graphics stack has alot of other components besides fglrx and from which most probably origin a few of the issues I'm experiencing. But it is also for this precise reason that I fear 9.3 won't be able to last me very long, nor that the open drivers will be able to keep up with their closed sourced counterparts, ever.

Oh and the open drivers are even worse on my configuration, just so you know. I paid for this god damn ati chip, I should be able to use it to it's full extent.

I'm not one of those Stallman apostles that believe every damn last line of code should be made public. If you guys need to keep some stuff secret, then so be it. There should be a stability & performance gain from that, however, and this is experienced by no one so far.


That said, thanks for the concise response ... more of this would be welcome from ati in general these days.

[crumja]

You guys are overreacting. I'm perfectly happy staying with 9.3 fglrx on debian for the next 2-3 years until the OSS driver matures in terms of 3d performance. Heck, RHEL and SLED are on much older kernels and xservers.

Think about it this way. What's the point of upgrading to a newer software stack if it doesn't work properly? No OSS driver - Intel, ATI, nouveau, etc. works properly with the latest Gallium/DRI2 code. The folks with Intel chips are staying behind on the 2.3 version of the driver and older xservers because of performance regressions that makes OSS ATI look speedy.

And what of nvidia? Their driver rewrites the whole X stack such that you're not going to see benefits from everything introduced in X over the past 6 months. The main driver might in a few months, but the legacy drivers certainly won't ever support those features. Plus, you get all the headaches of fglrx in suspend/resume as well.

[zeldaman55069]

I agree with crumja that everyone is overreacting! The fact of the matter is that the even if you don't stick with fglrx 9.3, the open source Radeon driver is at a fairly usable place. I've been running it on fedora for 3 years now, every once in a while trying fglrx but having that fail or not meet expectations. The fact is that this doesn't change much unless you are using the computer for gaming or CAD.

Even if we have to use Radeon, it's better than what it would be like on windows... using the provided drivers for Microsoft are just terrible... but as I fould out the Windows Catalyst driver will have a legacy branch. The linux side won't, but instead AMD/ATI has decided to help "provide ongoing support via the open source
drivers than by spending the same effort porting the current fglrx functionality to new OS and kernel versions. "

So stop complaining and be happy that ATI is opening documentation and providing help with the OSS drivers.