AMD Catalyst 7.12 Linux Driver -- The Baby's In Surgery

[pedepy]

bridgman, has anyone at AMD ever considered released weekly/bi-weekly snapshots of the proprietary driver, for the general public to try out ?

Maybe your bosses don't realize this, but most of the OSS community is more than willing to try out beta quality software, for the sole purpose of helping out. It's not quite the same mentality that you see with the rest of Windows users, where 99% of them just expect everything to just 'work'.

We know you guys are working on a new code base, and I'm sure if you ask around, you'll see most of us won't mind putting out precious little boxes to the test of whatever you might wanna try out. Perhaps you would have already ironed out that widescreen resolution bug, or the memory leak, weeks ago, had you taken that approach.

Oh and if you need further incentives for your boss: its FREE ! ... no added costs whatsover..

Just a thought..

[benwick]

8.01 - Waiting for Godot?

Sounds about right... and then 8.02 - Nausea

[bridgman]

bridgman, has anyone at AMD ever considered released weekly/bi-weekly snapshots of the proprietary driver, for the general public to try out ?

That is essentially how the current beta program works, and it has been quite useful. We can't really call it "free" though, since part of the implicit deal with the beta testers is that we *will* have devs look at the issues immediately and respond with questions or feedback so we can try to fix the problem and decide if it should be a release blocker. I don't think we can ramp up the community testing unless we also ramp up the resources available to investigate issues as they are reported. Without immediate investigation there is no real way to discriminate between a "stop the presses" issue and an install- or configuration-related issue which usually can be worked around and would not be a release blocker.

One of the big challenges with the Linux world is that the number of different configurations and distros is quite a bit broader than in the Windows world, so the noise level is quite a bit higher. The beta testers do a lot of work to help narrow down the problems and that keeps the "noise" sufficiently low that we can do our part -- not sure how many additional testers would be willing or able to make the same effort. Having said all that, we are definitely looking into ways to improve your experience with the drivers, and making better use of both community testing and the knowledge exchange on forums like Phoronix will probably be part of the solution. Another part of the solution will be aligning our internal testing to better match our user base, and this is where your input can help a lot.

Anyways, we'll try to cover this more when we do the Q&A session and get input from the Linux driver team. I can comment on general stuff but for the specifics we'll want to get the driver experts involved. The last time I worked on XWindows code was around the time the world was moving from X10 to X11

[smlbstcbr]

That is essentially how the current beta program works, and it has been quite useful. We can't really call it "free" though, since part of the implicit deal with the beta testers is that we *will* have devs look at the issues immediately and respond with questions or feedback so we can try to fix the problem and decide if it should be a release blocker. I don't think we can ramp up the community testing unless we also ramp up the resources available to investigate issues as they are reported. Without immediate investigation there is no real way to discriminate between a "stop the presses" issue and an install- or configuration-related issue which usually can be worked around and would not be a release blocker.

One of the big challenges with the Linux world is that the number of different configurations and distros is quite a bit broader than in the Windows world, so the noise level is quite a bit higher. The beta testers do a lot of work to help narrow down the problems and that keeps the "noise" sufficiently low that we can do our part -- not sure how many additional testers would be willing or able to make the same effort. Having said all that, we are definitely looking into ways to improve your experience with the drivers, and making better use of both community testing and the knowledge exchange on forums like Phoronix will probably be part of the solution. Another part of the solution will be aligning our internal testing to better match our user base, and this is where your input can help a lot.

Anyways, we'll try to cover this more when we do the Q&A session and get input from the Linux driver team. I can comment on general stuff but for the specifics we'll want to get the driver experts involved. The last time I worked on XWindows code was around the time the world was moving from X10 to X11

Good to read that! Most of the forumers were waiting for such an answer.

[givemesugarr]

In short, there's not a lot about talking to the codenamed drivers, but there IS some discussion all the same- and all I did was plug in "ati orca" in Google to get to what I handed you just now.

i did find out that links, but they only say that orca is a rewrite of opengl code for windows and linux. from the links you've pointed that the only thing that is know about orca. i thought it was some new framework but it seems it's just a codename for the rewrite.

Something about the name should have tipped you off.

Accelerated INDIRECT GLX...

Direct mode means that you're submitting the GLX rendering requests directly to a dispatch engine from the local app- little to no abstraction involved. You need Direct mode to get peak speeds with indirect mode, which accelerates the traditional pathway for GLX, allowing remote rendering and things like Compiz.

well, i never thought about the acronym and since i don't usually use remote rendering, for the moment, i wasn't really interested. anyway, if i were to chose from stuff from ati i'd really have chosen a stable a performant driver, regardless aiglx support even if it can do remote rendering.
in my opinion the surprise that bridgman had when he found out that aiglx was the firt request for ati users was justified. the state of fglrx before 8.41 was really awful; the driver was working but the performance was really bad (i had about 100fps with my x200 at max processor speed) and i was really planning to switch my boards to nvidia. after 8.41 i bought a new hd2600 and now i'm quite happy with it, but i'd really have wanted bugfixes to long known bugs and performance increase without stability cost than aiglx.

[givemesugarr]

Well well. How sweet. However, I, like many users here, really do not care for performance, at least not now. I do care about stability.
You seem to be a lucky user. Read a bit more in this forum to see that there are many many users with real problems. And many of those users are not newbies...

maybe i've got a lighter use for the driver and that's the reason i don't experience much pain in the axx with fglrx. i've read arouund and there are users that have problems that i don't experiece (the wrong bus id is one of those). i'm not saying that the users don't have the problems, but maybe the problems they have is not entirely due to the driver.

I'm sorry but that is total crap. If the driver is not stable then do not release it to the entire community. Instead release it to your beta/test group and/or explicitly label it as alpha/beta quality, which ATI has not done (read the release notes...)
Like you I'm a developer and the single most damaging thing to do for your reputation is to release unstable code, as you very well know.

when in my company the deadline gets near and the code is not finished they assign a lot of devs from other stuff to the one near the deadline. with amd i undestood that this is not possible, and since you have to release something that works but it's not really ok, you have to fix out as much possibile for the average users than fixing the single problems. then the problem in normal code release is that if your company doesn't deliver a good product the client goes to another company, which is not possible for the fglrx driver since other companies don't have the right to work on it. so as you see this developing process is different than software developing processes.

Beside, many users are seeing regressions that should not be seen. WTF happened that non-vesa resolutions can't be used any longer??? This is a major f&*k-up.
As a professional developer this gives me serious doubts about the development process they are using. This should not be possible! And this is only one of the examples! There are even regressions that were fixed in the 'old' codebase and re-appeared in the 'new' codebase. Are they actually using new code? Or is it simply that the don't have their process under control?

from what bidgman told us about the orca codebase and about how the chips are working and from what i seemed to understand, i think that amd/ati wants to remove all the old codebase in favor of the new orca and make 2d accel via 3d emulation instead of hw 2d. this is not simple since the old 2d codebase needs to be changed and because old chipsets have 2d hw acceleration.
from what i've understood they're substituting the old code with new orca one, which is not bad for scalability and which will help future interoperability for windblows and linux based on opengl and also the performace between the 2 drivers should be really similar. the problem is that the new orca needs to work out very well or else the driver won't work

I think you are seeing it all a bit too rosy. Try to _really_ look at what they accomplished. Aside from a very crappy implementation of the OpenGL extension that is needed for Compiz there is really very very little that they have done. In fact, many things do not work anymore, like suspend (which is critical for laptop users).
Ok ok, they have improved the performance a lot. But at what cost? By breaking many things that did work? I would rather have stability than performance. And if you're adding features, make them work! Do not experiment with your users (like M$, Apple, Cisco, etc.) That is really really damaging for your reputation, and reputation is worth soooo much, especially in the community.

well, you cannot negate that the driver had some really good performance increase, which was a real bless. the problem with broken features in my opinion is all due to the code substitution with the new one. this is true for the old features. for the new features i think that they should be introduced when they're ready and work well, but if the community asks for aiglx and you have aiglx that somehow works then you'll push it out even if it's not complete and is only at 60-70% of its development. else you'll continue to loose clients. in the last months lots of potential nvidia clients were dissuaded to go with nvidia due to some good market strategies as the release of the specs and of the publicity around the new features in fglrx. the problem is that these clients will go with nvidia if they're not satisfied in less than 3-4 months.

Not completely true or false. I did switch my desktop machine (I have a spare X1900XT, wanna buy it?). But my laptop I cannot switch, like many users.

I do want to support ATI because of their recent initiatives towards the community but they are making it very hard for me. My main machine is my laptop and it sucks bigtime that suspend doesn't even work on it because of the ATI driver. This is a daily frustration, which is why 'enough is enough' and I switched it to vesa.

laptops with less than 2 years are most likely to not be changed, but one as mine that has 3 years ago tech inside is starting to be a little slow and will need a change in the future months.

Maybe. Maybe not. Like many others I'm very frustrated with the quality of the releases of their software. Maybe this makes it a bit more clear. Being politically correct many times doesn't drive the message home. They need to feel in their guts that they have a quality issue. Understanding it is not enough, they need to FEEL it.

I think you should take off your pink glasses and truly look around. Then re-evaluate your convictions.

I have done it. I was an ATI fanboy too. Still am a bit but their driver forces me to make different choices.

As for your reaction, your not quite as politically correct as you would like to believe yourself, but that might be your pink glasses. I do understand that all these reactions see like whining when you do not have problems, but try to place yourself in our shoes, we do have serious issues, and many of those issues frustrate the hell out of us...

I do hope that the driver improves fast but I have much higher hopes for the documentation drops that are coming. Once the community has the 3D documentation I think a stable functional driver will be coming fast in 2008. I have high hopes for both the RadeonHD and the regular Radeon drivers.

the problem with the documentation is that pieces of evil boards like mine won't come out fast and before they're out i'd already have switched the notebook for a newer r600 powered one. that's the only solution for this old hw. i know that fglrx is not a good driver, but it works and works better than the opensource one on my boards (i still haven't tried the new xf86-video-ati to see how it has improved on the r500). but if you compare it to the one we had a year ago you'll see that it has quite some improvements.

[lockheed02]

I'm not sure if I am doing something wrong but in my /var/log/Xorg.o.log I am still getting an unrecognized chipset error with my ATI 3870.

I got the same, but in Catalyst Control Center it shows "Radeon X1600 series"

When I was at Fedora 7 x86_64 there was a little workaround (http://www.phoronix.com/forums/showthread.php?t=5983). But I don't mind it actually so I didn't investigate it any further.

PS: I use a Radeon X1600Pro in Fedora 8 (32bit)

[LavosPhoenix]

Yeah, I could also care less about having an OpenGL accelerated desktop ATM. What I want is the ability to play widescreen videos, like 1280x720, just as smoothly as I could on Windows. That means no dropped frames, horrible tearing, and fullscreen playback using OpenGL, so the video doesn't look like ass. Game playback is also secondary. How is this too much to ask for? I bought your product, and I expect it work. I also expect to be able to use the full resolution of my monitor, 1680x1050, and not some godawful 4:3 res. Yes, I realize you haven't been working on the drivers for as long as the nVidia team, but at least you could get your priorities straight.

1. SMOOTH video playback using OpenGL without AIGLX.
2. Use of the maximum resolutions supported by a display device.
3. Decent or at least operational Game playback, while NOT breaking #1.
4. AIGLX, while not breaking #1 and/or #3.

Seriously, this is what users what, not shiny eye-candy. Users are not raccoons.

[Berniyh]

Actually I just want a driver, that *JUST WORKS*.

I think I didn't have a single driver in the last 12 months, that didn't at least have one issue, which is sad.
I don't care so much about performance and/or AIGLX, but I care about other things, that fglrx seems to have, but the open source driver (radeon, not radeonhd) doesn't seem to support (yet?).
Like powerplay.

Actually 8.42.3 and 8.433 seemed to go in the right direction, but 8.443.1 is a bit disappointing.
It really would be cool if once, somewhere in the future there would be a driver release, that doesn't bring new bugs, but resolves some (I'm hoping, but not requesting a bug free release, that doesn't have any bugs, just one, that doesn't introduce new bugs). That would be a major step forward.
Currently it seems to be like every driver release fixes a few bugs (normally more than there are listed on the release page), but at the same time brings some new bugs which seem to annoy everyone.

Maybe AMD should think about releasing "Beta" drivers (and in my eyes, 8.443.1 is a beta driver) as beta drivers.
I think I and others could live with that, if some of the months there is no "stable" driver, but a "beta" driver, if there are some bugs, like the resolution issue, that could not be solved out in time.
Or, what would be a better idea in my opinion, change the release cycle.
That One release/month cycle doesn't really help anyone.
Drivers should be released when they are ready, not when there is a pin in the calender.

Just some ideas.

[jackkerouac]

Well, I took the plunge and installed 7.12 in Ubuntu Gutsy with my X1650 Pro card and lost the Xserver. It would boot up and give me a blank screen. Then nothing.

So I went back, reinstalled 8.40.4, copied my old xorg.conf back and still the blank screen.

After much hemming and hawing and swearing, I had to uninstall Compiz, completely remove 8.40.4, 7.12 and anything else resembling fglrx, REINSTALL 8.40.4 and the reinstall Compiz.

Now everything is back to where it was before I installed 7.12.

Sigh. Serves me right for trusting AMD/ATI. I swear, if running an ATI card under Linux has taught me ONE THING, it's do NOT RUN AN ATI CARD UNDER LINUX.

Nvidia, here I come!