ATI FAQ, maybe

[leef]

I keep reading the same questions and answers, so I thought maybe a FAQ might be useful. I'd like to keep bridgman working on releasing docs rather than his formidable writing skills in the forums (although it's great reading). Sadly, I lack those skills, so please no one be offended. There wasn't much interest before, but, just in case, here's my first draft. Remember, if this sounds harsh, it's only the first draft. Should this be in the forums? On AMD's site?

If this is helpful, maybe we can improve it and Michael can give it a home:

ATI FAQ

1. Why does fglrx "suck" (fail to meet my needs adequately)?
A. It's aimed at a narrow market segment doing specific tasks with FireGL cards. That it kind of works with Radeon cards in games is useful (for RE and for those games which happen to work), but academic, unless you like blobs.

1a. I do like blobs.
A. Buy the FireGL card and complain to AMD.

2. When will [feature] be fixed?
A. As soon as someone fixes it. The linux side is open, but lacks some video infrastructure. Download the source and add your feature.

2a. But I'm not a programmer. When will [feature] be fixed?
A. When someone fixes it. Nagging the people who are working might help. Sending them money might help more. Sending good bug reports is useful, too - Download the git or svn version and test.

2b. But ATI hasn't released the documentation for [feature] yet. When will that happen?
A. When ATI gets around to writing the documentation and their legal staff gets around to allowing release. So far, AMD has promised to release documents and has actually released quite a few.

2c. When will the cards work "out of the box"?
When Xorg/Xserver/Mesa/DRI/DRM (driver) are each updated for your "out of box" features and your "distro" has incorporated them. This could take years depending on whether you like compiling your own (quicker) or relying on the "stable" variant of a "distro" (slower).

3. Which documents aren't available?
A. R2xx, R6xx 3D, R7xx, Power management, UVD, UVD2, and IDCT. Don't expect UVD. Don't expect all of UVD2. Rage 128 and R1xx aren't available either. RS4xx is AWOL./1/

3a. So, basically, the latest snazziest cards aren't supported?
A. Sorry. Soon. Maybe fglrx works for your application? Seriously, good people are working on this. /current status/

3b. And High Definition video (1080/AVCHD) isn't supported?
A. Not yet. Probably in fglrx before OSS.

4. Where can I get info on status?
A. Try X.org. Hint: http://www.x.org/wiki/RadeonFeature. Read the commit logs for the various X and driver projects. Hint: http://www.radeonhd.org/. Check Phoronix. Michael usually posts a news item for more interesting developments.

4A. I need more detailed information on a specific feature.
A. Google. Search the Phoronix forums. The variety of cards, drivers, and features is overwhelming. Here's a short list of features. Please turn it into a table and add status by card and driver:

Mode Setting
Not yet for 4670, 45xx variants. VGA maybe, no DVI. /article/

Multihead - Radeon uses XRandR - see XRandR1.2 for limitations
clone desktop
big desktop
multiple X servers
multiple video cards
Xv across screens/Xservers/Cards

Power - some basic function available to AtomBios cards, maybe. - need docs
automatic load based depowering of unused sections.
automatic load based underclocking
manual clocking
shut down card without unplugging

Status
Temp,voltage,clock sensor monitoring - need docs

Overlays
Xv - See X.org
GL - yes (but R5XX and below)
Redirected direct rendering - in GL, rest needs X improvements

Video Decode
h.264 - not on linux yet
mpeg2 - boring, unless you have low power machine or requirement
tearing - needs X changes (fallback to GL or fullscreen or no compiz)/2/

OpenGL
version - OSS is at 1.3 - need X changes (memory management) for higher functions

Stability - my R2xx is quite stable.

GPGPU - Need GLSL. (For cards with shaders and docs, you can use the shaders directly. Per AMD, CAL requires fglrx and R6xx or better.)

5. Linux gaming?
A. The games want better OpenGL than 1.3, so you'll need the blob, but the blob isn't designed for games. And no one is going to spend time on OpenGL 'til X is updated with a memory manager. And, yes, good people are working on that too.

6. What about the competition?
A. Nvidia is blob. They're ahead of the game because they also wrote some of their own X routines into the blob. See Nouveau for reverse engineered open mostly 2D driver. Intel is fairly open, but their 3D hardware hasn't the oomph. Nobody's drivers have all features.

7. Why RadeonHD and Radeon?
A. No paraphrasing is possible to answer this question. Get your favorite caffienated beverage and check out:
The short beginning: http://www.phoronix.com/forums/showthread.php?t=7032
The relentless: http://www.phoronix.com/forums/showthread.php?t=11458
The topical: http://www.phoronix.com/forums/showthread.php?t=13390
And possibly: http://www.phoronix.com/forums/showthread.php?t=11664

/1/ R6xx and R7xx example code was released December 2008. See post 19 for references to Rage and R200 documentation. See 2008-10-17 Radeon IRC for RS4xx.
/2/ See from post 24 on down for Tearing status and explanation.

[TechMage89]

Allow me to make a slight correction. The information needed for MC (uses the 3d engine, actually) is already available, but no one has shown an interest in doing it. It looks like the way of the future for video decode may be a generic implementation on top of Gallium (there's already a partially working one for mpeg2), so it will need a Gallium driver first. If someone wants to implement MC, however, the door is still open (for <=r500 at least.)

Also a note that development on r600 and r700 has been going on behind the closed doors of NDA while AMD works through the legal hassles to get the docs released. Hopefully there will be the good beginnings of a driver done by the time the docs are public.

[duby229]

Also a note that development on r600 and r700 has been going on behind the closed doors of NDA while AMD works through the legal hassles to get the docs released. Hopefully there will be the good beginnings of a driver done by the time the docs are public.

Which isnt really open now is it? AMD needs to adopt an open source strategy that actually allows open development.

[bridgman]

We have a strategy that allows open development -- we're just doing a bit of NDA development first

This approach is pretty common -- but what you normally get told is "vendor XYZ releases info, driver written in only 2 weeks, amazing"

[leef]

Allow me to make a slight correction. The information needed for MC (uses the 3d engine, actually) is already available, but no one has shown an interest in doing it. It looks like the way of the future for video decode may be a generic implementation on top of Gallium (there's already a partially working one for mpeg2), so it will need a Gallium driver first. If someone wants to implement MC, however, the door is still open (for <=r500 at least.)

Also a note that development on r600 and r700 has been going on behind the closed doors of NDA while AMD works through the legal hassles to get the docs released. Hopefully there will be the good beginnings of a driver done by the time the docs are public.

Thanks for the reminder. I'll update the list. Here's the reference:

http://www.phoronix.com/forums/showpost.php?p=39296&postcount=83

Were you refering to the Nouveau GSOC project?

You are also quite correct about the semi open R600/700 development. At least the resulting code is open. I was trying to leave the philosophical questions alone. Included links to some of the better discourses. A section on history might be interesting. We could list the accomplishments (fastest open source 3d in 2007, the 2001 R200; fastest in 2008, the 2006 R580+) and missteps, but that's really a whole different, if deserving, project. Certainly the Wikipedia has good hardware history and next to nothing on driver development.

I was kind of hoping the last 3D docs would get released, Xorg would get updated, and the open driver overlap would get sorted out soon, so we could move on to new issues.

[duby229]

We have a strategy that allows open development -- we're just doing a bit of NDA development first

This approach is pretty common -- but what you normally get told is "vendor XYZ releases info, driver written in only 2 weeks, amazing"

Actually more often then not headlines read "Preliminary beta driver released despite total lack of documentation" I'll admit that over the past year AMD has been a little better then that, but we still dont have anything useful for that last two generations. And the the generation before that is working, but far from stable.

In other words support for RV700 and R600 is still completely missing And even after all this time support for R500 is buggy and incomplete. Support for R300 and R400 is more stable, but still incomplete. R200 and R100 is both complete and stable, but too slow to be useful in todays day and age.

So you've got the gall to have couple of people developing code behind closed doors while calling it open.... It doesnt make any sense at all... None...Why not allow them to do it in public where the entire community can be useful? And whiler your at it, why not take the next step, and drop the totally worthless fglrx driver and reallocate those poor folks developing the linux portion of that to the open source team? And hell if your going to take the time to do it right, tell Novell to go screw themselves, and hire there devs to work on your team. Any way you look at it, Novell is using YOUR money to pay there devs. And right now they are getting nowhere. Why not get rid of Novells anti-linux agenda and pay them to actually accomplish something?

Why not take the time to do it right? You'll use fewer resources, and have one driver that works across the board. Thats the goal isnt it? And if it's not shouldnt it be?

[bridgman]

Aw nuts, I just finished replying to your email and you changed the whole thing

[duby229]

email? I did a pretty big edit on my post above, but I didnt change anything, I just added the last line and the paragraph above it.

[bridgman]

Whoops, make that "post". I'm really tired, time to go home I think...

[duby229]

Whoops, make that "post". I'm really tired, time to go home I think...

hehehe. I've been there too. Take a break. Friday night, it's time to drink some beers... Or scotch if you have the taste for it.