xf86-video-ati 6.11.0 Driver Released

[phoronix]

Phoronix: xf86-video-ati 6.11.0 Driver Released

Just a week ago there was a new release candidate for the xf86-video-ati driver, but today the ATI 6.11.0 driver has been officially unveiled. Worth noting in this open-source X.Org driver update is a CRTC/output/encoder rework and Render extension repeat mode fixes...

http://www.phoronix.com/vr.php?view=NzA3NA

[ssam]

i found this bit interesting

6.12.0 will be soon to follow with accel support for r6xx/r7xx chips
http://lists.x.org/archives/xorg-driver-ati/2009-February/008479.html

how soon is soon

[d2kx]

Hopefully this one together with newest drm will not hardfreeze my system like it does with every driver version so far on my RV770Pro. RadeonHD works fine. agd5f said he fixed severals hangs if you have got both updated (radeon/drm), so I assume it will.

[FunkyRider]

3D acceleration for RV770? I'm still dreaming it would come true...

[korpenkraxar]

Any progress on power management for laptop chips?

[agd5f]

Hopefully this one together with newest drm will not hardfreeze my system like it does with every driver version so far on my RV770Pro. RadeonHD works fine. agd5f said he fixed severals hangs if you have got both updated (radeon/drm), so I assume it will.

Please try ati git master (make sure you have 42492c70c481c88911337eefb97ba9d030adf7b6).

[d2kx]

Please try ati git master (make sure you have 42492c70c481c88911337eefb97ba9d030adf7b6).

Well, it did not help. But that was without latest drm and with Xserver 1.4.2 (Debian/sid, will get uptodate again very soon). Mmh.

[Nille]

Any progress on power management for laptop chips?

I don't think so but i bet this has a high priority after 3D Support and after this comes hw video acceleration.

[bridgman]

Yep. The core power management info for 5xx is already out and devs are starting to work on various implementations; we still need to write up the corresponding info for 6xx/7xx. The 7xx mobile parts are just hitting the market now (M9x) but there are 6xx mobile GPUs out there now (M7x and M8x).

[highlandsun]

Are the chip family names intentionally different from the card/chipset family names, to confuse competitors? It's sure confusing me.

What GPU family is in my Puma laptop, with HD3450? Which rev of these drivers has the most complete support for it now?

[agd5f]

Well, it did not help. But that was without latest drm and with Xserver 1.4.2 (Debian/sid, will get uptodate again very soon). Mmh.

This should be fixed now in git master.

[bridgman]

Are the chip family names intentionally different from the card/chipset family names, to confuse competitors? It's sure confusing me.

I don't think we try to make them confusing, but when ATI and AMD joined up we kept our existing numbering/naming schemes. Now that we're talking about CPU families & products, GPU families & products *and* platforms there are an awful lot of names and it would be nice to simplify the naming a bit.

When we talk about GPUs (eg RV770, or the 7xx family) we mean the actual chip. The marketing names (eg HD3450) describe something more like a graphics subsystem - chip, clock speeds, memory type/speed. An HD3450 is an RV620 GPU chip at a certain speed with a certain memory complement.

What GPU family is in my Puma laptop, with HD3450?

The Puma platform is defined as a Griffin processor (aka Turion Ultra), plus a 780 chipset (which includes the HD3200 graphics subsystem, which in turn is roughly RV610 3D core plus UVD and display blocks from the 7xx family) plus a specific southbridge (:confused:) plus an optional (I think) HD3450 GPU (aka rv620) in a Hybrid Crossfire arrangement.

When talking about GPU families, you have two 6xx-family GPUs; one DX10 (the 610 core in the 780) and one DX10.1 (the 620 core). Both have pretty much the same processing power (2 SIMDs x 4 instructions/clock/SIMD x 5 ALUs per instruction = 40 ALUs) which is why they work well in a Hybrid Crossfire arrangement.

Which rev of these drivers has the most complete support for it now?

In general, the fglrx driver will do "complicated things" best (eg Hybrid Crossfire, general 3D features/performance, multimonitor acceleration etc..) while the open drivers will do "simple things" better (because the same devs who update the framework typically update the open drivers at the same time and actually *use* the open drivers to test out proposed changes).

The gap between the two drivers will get smaller over time, so it will be a lot easier to pick one driver for your usage patterns and not feel saddened by the things you're missing from the other driver.

[highlandsun]

Thanks, that helps clear up a bit for me.

[d2kx]

This should be fixed now in git master.

I'll try it as soon as I get home. I somehow assumed it has got something to do with > 512 GB VRAM, because that's what I also saw in related bug reports. Thank you for your work :)

[wfeltmate]

Bridgeman, on the thought that the open source drivers perhaps perform 2D operations better, could FGLRX be made to use the open source 2d code and still maintain a closed source blob of just the 3d code? For the chipsets that have 2d hardware anyway?

Allow those users the best of both worlds?

[bridgman]

The difficulty is that both 2D and 3D need to use the same memory manager, and the proprietary memory manager is one of the places we get both performance and feature advantages over the open drivers.

I don't think there will be a permanent 2D performance offset between the open and closed drivers, it's more that when the framework changes (eg recent improvements to EXA in the X server) the open drivers will make use of those changes more quickly than fglrx.

[d2kx]

This should be fixed now in git master.

Yes, it works now with latest -master. Thanks :)

[msebast]

The Puma platform is defined as a Griffin processor (aka Turion Ultra), plus a 780 chipset (which includes the HD3200 graphics subsystem, which in turn is roughly RV610 3D core plus UVD and display blocks from the 7xx family) plus a specific southbridge (:confused:) plus an optional (I think) HD3450 GPU (aka rv620) in a Hybrid Crossfire arrangement.

I appreciate the effort but this sort of answer is more mind-numbing then helpful. I'm the sort of person who ought to be able to understand. (I'm an electrical engineer and write semiconductor test software for a living.) But the confusing product names, chipset names, driver versions, mesa versions, xorg versions, etc. is just way too much. If this is too complicated for me (and I like complicated things) then the average person must be hopelessly lost.

A year ago I bought a board with a 780G chipset to run mythtv and play occasional games. I was excited by the news of upcoming documentation releases and AMDs/ATIs commitment to working with the community to make good open source drivers a reality. I'd been burned by brittle binary drivers from Nvidia and was hoping for something better.

I've been using Mythbuntu 8.4 and fglrx with mixed results since then. By turning off compiz and putting the right magic in xorg.conf I was able to get things usable. But there are a variety of problems. mythfrontend crashes when starting playback, some games crash, and video playback has tearing. Binary video drivers are brittle and buggy. This is true (to varying degrees) on both Windows and Linux and for both ATI and Nvidia. I suspect the problems with binary drivers are not solvable.

I've been following the news on Phoronix and I understand the many huge software changes that are happening in xorg, mesa, and the open drivers. But I still can't figure out when or if I will ever get a stable system. The Phoronix news reports talk about acceleration on R600 or R700 but I'm never sure if that applies to my chipset.

I'm still not sure if I should switch to Ubuntu 8.10, wait for 9.4, or hold out for 9.10. Which of those will have working, open, drivers? I want something that won't crash mythfrontend. I want the video tearing to stop. It would be nice if games with modest 3D needs would work.

I don't really need technical answers to my questions. Those answers won't really solve your problems or mine. I'll still be confused, the binary drivers will still be unstable, and AMD/ATI will still have a marketing problem with confusing product names.

Please give this feedback to the marketing folks and engineering managers:

The 780G chipset could make an excellent mythtv system. Everything about my system is great. All the software is stable and works smoothly EXCEPT the video drivers.

When people visit my house they always like my system. I've had several ask me to setup something for them. But I always refuse for ONE reason; I know the video drivers are too unstable.

Bridgeman, you do an admirable job representing your company in these forums. I've been reading your posts for the last year and I'm impressed with how well you control expectations, respond professionally, etc. Are you able to do the same thing in reverse? Can you be a voice for our needs within AMD/ATI?

Please see here: http://laforge.gnumonks.org/weblog/2009/02/04/

Although your company is more responsive then many, I think that blog post still applies. I suspect AMD is missing many marketing opportunities due to the lack of good open source video drivers. I know there are people at AMD who understand that. (That's why things changed last year and you are on this forum.) But AMD is still missing out, because we still don't have good open drivers in the distros. And most managers at AMD are about 4 or 5 steps removed from the end customers who ultimately drive demand for your products. That order for 100,000 780G chips from some assembly house in China depends on guys like me building cool PVRs for their friends.

Regards,
Michael Sebastian

[bridgman]

I appreciate the effort but this sort of answer is more mind-numbing then helpful.

Understood -- that answer wasn't particularly useful to most people, I was just trying to give a specific (and detailed) answer to a specific (and detailed) question. Hopefully most of my answers are easier to understand :D

I'm the sort of person who ought to be able to understand. (I'm an electrical engineer and write semiconductor test software for a living.) But the confusing product names, chipset names, driver versions, mesa versions, xorg versions, etc. is just way too much. If this is too complicated for me (and I like complicated things) then the average person must be hopelessly lost.

I agree, and I don't see it changing for another 6 months or more. Home Theater has a lot of the same issues (and since you're doing Home Theater with Linux, you get a double shot of fun)

It's nice to think that everything will be rosy once the development community has caught up and we have nice ready-to-go drivers for all shipping GPUs, but new GPUs will continue to arrive and distros tend not to like updating kernels frequently just to pick up driver support for new hardware. We need to make sure that new drm packages can come down automatically in the same way that userland drivers and apps can be updated today.

A year ago I bought a board with a 780G chipset to run mythtv and play occasional games. I was excited by the news of upcoming documentation releases and AMDs/ATIs commitment to working with the community to make good open source drivers a reality. I'd been burned by brittle binary drivers from Nvidia and was hoping for something better.

I've been using Mythbuntu 8.4 and fglrx with mixed results since then. By turning off compiz and putting the right magic in xorg.conf I was able to get things usable. But there are a variety of problems. mythfrontend crashes when starting playback, some games crash, and video playback has tearing. Binary video drivers are brittle and buggy. This is true (to varying degrees) on both Windows and Linux and for both ATI and Nvidia. I suspect the problems with binary drivers are not solvable.

The 9.2 Catalyst driver release has some improvements for myth; wouldn't hurt to give it a try. The release notes mention corruption rather than crashes though...

The open drivers have experimental tear-free code which is starting to get pretty solid; we may be able to do something similar in fglrx, not sure yet but stay tuned. You probably should be thinking about trying the open drivers fairly soon anyways based on your current needs.

I've been following the news on Phoronix and I understand the many huge software changes that are happening in xorg, mesa, and the open drivers. But I still can't figure out when or if I will ever get a stable system. The Phoronix news reports talk about acceleration on R600 or R700 but I'm never sure if that applies to my chipset.

Yes, it applies to your chipset. IGP parts are always a bit hard to classify, since the 780 is basically 7xx display and video hardware combined with 6xx 3D hardware. Anyways, the 6xx/7xx HW accel code is living in development branches right now but should become available to distro packagers quite soon.

I'm still not sure if I should switch to Ubuntu 8.10, wait for 9.4, or hold out for 9.10. Which of those will have working, open, drivers? I want something that won't crash mythfrontend. I want the video tearing to stop. It would be nice if games with modest 3D needs would work.

8.10 has full open drivers for 5xx and RS690, plus "modesetting" drivers for newer chips including 780, but I don't think the drivers in 8.10 are worth an upgrade for you.

I am hoping to get 2D and video acceleration into 9.4, but don't think we will be able to get 3D acceleration ready in time since 9.4 is already rapidly approaching freeze. Hopefully we can at least get the required kernel bits in so that a new 3D driver can be easily downloaded after 9.4 is released. 9.10 will have everything but I don't think you'll have to wait that long.

[QUOTE=msebast;63562]I don't really need technical answers to my questions. Those answers won't really solve your problems or mine. I'll still be confused, the binary drivers will still be unstable, and AMD/ATI will still have a marketing problem with confusing product names.

We do have a wide range of interests here though; some people really ARE looking for technical answers and anything less will just annoy them. I do try very hard to match the answer to the question; ask me a simple question and I can usually give you a simple answer, although there are a few (seemingly) simple questions which *require* complicated answers ;)

The 780G chipset could make an excellent mythtv system. Everything about my system is great. All the software is stable and works smoothly EXCEPT the video drivers. When people visit my house they always like my system. I've had several ask me to setup something for them. But I always refuse for ONE reason; I know the video drivers are too unstable.

Yep; our Linux focus until a year or so ago was entirely on commercial workstation products; now we are also trying to improve the consumer experience. You should continue to see improvements each month for a while longer.

The open source drivers are probably also just coming up to meeting your needs, and for the short term they may be a better fit for you based on earlier comments.

Bridgeman, you do an admirable job representing your company in these forums. I've been reading your posts for the last year and I'm impressed with how well you control expectations, respond professionally, etc. Are you able to do the same thing in reverse? Can you be a voice for our needs within AMD/ATI?

I already do. I just can't tell you about it :D

Please see here: http://laforge.gnumonks.org/weblog/2009/02/04/

Although your company is more responsive then many, I think that blog post still applies. I suspect AMD is missing many marketing opportunities due to the lack of good open source video drivers. I know there are people at AMD who understand that. (That's why things changed last year and you are on this forum.) But AMD is still missing out, because we still don't have good open drivers in the distros.

Harald is also trying to stay professional and not name names, but my guess was that he was *not* talking about us. Harald is dealing with the same challenges at Via and I think he understands what both we and Intel have been doing. Only guessing though; I haven't had a chance to meet him yet.

Getting working drivers into distros takes time; typically about a year from the initial availability of documentation. We have tried to reduce that time by working in NDA repositories in parallel with document preparation, so for 6xx/7xx I think we're already more than half way through that year.

I do have to disagree with a couple of Harald's points, at least from a GPU perspective (in fairness, he was talking about all hardware not just graphics, and I think it's safe to say that GPUs have the most complex programming model of any hardware out there) :

As a chip maker, you first and foremost concern should be to sell as many units as possible. You don't care what kind of software your customers use. You don't care where they get their software from, or what development methodology they use.

This sounds nice (and probably works for most hardware) but modern GPUs are sufficiently complex that *very* few developers can be effective without support. The open source community is pretty good at "making do" but customer software development can often result in support costs which outweigh any money made from chip sales.

So if you can take any step to encourage more alternatives and more competition/innovation on top of your chip, you can only gain market share, but not lose it. And if you don't gain market share, well, you didn't have to make any investment. So no matter what you do, you can hardly loose anything.

Again, I fully agree with the upside part of the statement but from a GPU perspective Harald is significantly understating the cost and risks of providing documentation and code to enable and support open source driver development.