ATI's New Drivers: Did The Paradise Come?

[phoronix]

Phoronix: ATI's New Drivers: Did The Paradise Come?

It's been nearly seven months since I wrote my first article about ATI drivers and how they literally crippled my computer and my daily life. Last month, ATI had finally released their latest drivers which contains the new OpenGL component and AIGLX support. Was this release worth the amount of hype and did it solve everyone's problems? For me, simply, no. After seven months of waiting, updating, and struggling, things are not good for me, in every possible feature of the ATI Linux driver I use.

http://www.phoronix.com/vr.php?view=11424

[rolz]

so it isn't just me ...

Well i hope they stabilize it soon ...

but i am also sick and tired of the word soon; I know it isn't easy. I am a developer so i can respect that. but still ... I feel like i might have waited long enough.

[Mithrandir]

Maybe it is time to stop waiting for better drivers and start waiting for 3d specifications even though that would mean waiting one more year until stable 3d driver that would work out of box in all distributions.

[Ole-Martin Broz]

well i can agree with everything except powerusage for my hd 2900 XT, it doesnt seem to run at 100% all the time, but this is not the case on the x850 XT PE for sure, it goes on 100% all the time.

but i dunno hows the 2900 XT goes, if its on 100% or not .

performance is godlike though, my aiglx issues are small, but still annoying, my videoplayback issue is a very annoying issue, but they have certanly done something, and the 8.43.xxx may bring some hope.

[Malikith]

well i can agree with everything except powerusage for my hd 2900 XT, it doesnt seem to run at 100% all the time, but this is not the case on the x850 XT PE for sure, it goes on 100% all the time.

but i dunno hows the 2900 XT goes, if its on 100% or not .

performance is godlike though, my aiglx issues are small, but still annoying, my videoplayback issue is a very annoying issue, but they have certanly done something, and the 8.43.xxx may bring some hope.

I hope you're right, 8.43 should be mostly a big bugfix release. Or at least we hope so. About your 100% load thing, I think you're right, because with my x1600XT, if I touch the heatsink while running Linux it feels hotter than normal. But I tried it in Windows for a test and it was nearly ice cold. The card is a pretty good overclocker too. It can go from 590/690 to 700/760 or in other words, 700/1520.

Granted, I can bring it a little higher but I don't like pushing it to the absolute limits. It still doesn't get as hot in Windows as it does in Linux. I wish the x1600XT had a temperature sensor on it and wish they had a temperature reader in the CCC for Linux.

[Svartalf]

Two Words describe this situation with the new codebase drivers...

Train...
Wreck...

[chair]

Same here. Runs hot under Linux and video playback sucks. 3D runs well enough, but who cares if I can't watch video.

I'm using my old Nvidia 6200 in Linux. I throw in the 2600HD when I boot into Windows to play games, which isn't too often.

[aznabaal]

The memory leaks are amazing. I just valgrinded glxgears
and there are bunch of them in fglrx_dri.so

Don't ATI developers use basic tools?

[Svartalf]

The memory leaks are amazing. I just valgrinded glxgears
and there are bunch of them in fglrx_dri.so

Don't ATI developers use basic tools?

Keep in mind, most of the dev team there are WINDOWS developers. It probably hasn't occurred to them that they've got better tools at their disposal than BoundsChecker. :D

[yoshi314]

ATI's New Drivers: Did The Paradise Come?uhhhh....no. :] quite the opposite :]

[container]

It's funny that although I've always used fglrx, I haven't really ever noticed any video playback issues, certainly not "I can't watch video". I tend to use mplayer, maybe that's why. And -vo gl.
I wonder if that memory leaking stuff is inherent to the driver, I should try to recreate it and find out. Since installing the newer drivers, some games and other accelerated Gl programs have frozen my Ubuntu box. It might be either that or the heating Issue.

[mdseymour]

I've had nothing but bad experiences with ATI since the mid-90s with Windows and Linux -- mostly bad driver support (TV cards, linux drivers) and poor display quality compared to Matrox and to a lesser extent nVidia. When I could, I have avoided their video cards since then. I'm still wondering why people use ATI, especially with Linux.

[Svartalf]

Because they seem to have a hammer-lock on the laptop market (I've got a couple laptops with their stuff in it... One that's supported with Open Source drivers and one that's not- sadly, the one that's supported with fglrx is actually slower than the one that's not- but I suspect that's more a function of the lack of vertex shader support in that chip than the drivers themselves.).

That, and many people are coming in from the Windows world and they're stuck with this stuff and don't have budget to support buying an NVidia. Couple this with a desire to support them now that they're slowly opening up all the info in spades- well, you've got the picture.

Me, I'm not waiting for the driver to get done. I want their tech specs. If they can staff up and get their act together, fine- their driver will be faster because they've been up the food chain in 3D graphics longer than we have. However, until they get it stable, it's just not going to be an option for most people- period.

[tsuru]

With 8.42.3, I lost the Render to texture extension (GL_render_texture I think) compared to previous drivers. Along with the logo artifacting at the bottom right. The are the two most obvious and annoying...

[DirtyHairy]

Thanks for the comment on the memory leak; I was simply disbelieving my eyes. I first noticed it in quake3: it leaks nearly one GB in ~45 minutes! I also can confirm the leak with glxgears; again, it is of the order of MB/s ! Quite an impressive example of bad coding... I really hope this gets fixed in the next revision.

[chefkoch]

I'm not seeing any memory leak. Nothing as extreme as reported here at least.

I just ran glxgears for a couple of minutes and its memory consumption was constant.

Maybe it depends on the hardware? I'm using a 9550.

[DarkFoss]

I'm not seeing any mem leak here..I'm running an x800pro agp though.
glxgears mem usage remains a steady 1.2% fgl_glxgears is a constant 1.4% both run for 15 min and observed with top... I've run UT2K4 under compiz-fusion for over an hour no mem leaks there either ..compiz-fusion had been up for over a week and I still had plenty of free ram..

[igknighted]

The second problem is with AIGLX support. After this months driver release, folks around the globe were dancing happily because the latest drivers were successful in initializing Compiz, Beryl, and friends. I'm an eye-candy lover but I believe that it should be subtle and should be used to increase the productivity of the end user. In my opinion's light, I find KDE's internal composition is sufficient and neat so, I enabled them but, unfortunately, if I understand it correctly, while AIGLX is implemented and working, it's lacking composition support and hence, all of the KDE desktop was rendered by my CPU with the expense of extreme CPU usage and slowness. As a result, I turned it off and my eye-candy quest has ended even before starting, again...

IIRC, aren't KDE3's "compositing effects" software driven? I didn't think they were capable of using AIGLX or a true compositing WM properly (and therefor getting the video hardware to do the work).

Also, there are indeed incompatibilities between Compiz/Beryl and ATI's drivers. But you can't put the blame squarely on one or the other. C-F does some squirrelly things, and ATI shouldn't add squirrelly code to cover for those.

That all said, this has been a major code rewrite, and with it comes a lot of bugs. Lets see if ATI cleans it up, because the rewrite has major promise in benchmarks and is anything but a trainwreck so far... but it could end up there if they don't start fixing bugs.

[Silent Storm]

Hi all,

I'm Hakan, the writer of the article and errm so, here are some replies...

1- @DarkFoss && @chefkoch: It seems like memory leak is "generation" dependent since driver design for such devices require several code sections which are special for some hardware.

2- @igknighted: No. XComposite & KDE Translucency is HW accelerateable and KDE uses it. nVidia cards accelerate it very well. I use the same OS (I cloned my home PC to office) @ office with an 7600GT and Use Pardus with KDE on geForce Go 7200 on my laptop. Both use the effects without any CPU usage. Ah, FYI, KDE has software composite and XComposite is SW emulated if no accelerating HW is found.

Edit: Added more info in "2" and fized some typos.

[igknighted]

@igknighted: No. XComposite & KDE Translucency is HW accelerateable and KDE uses it. nVidia cards accelerate it very well. I use the same OS (I cloned my home PC to office) @ office with an 7600GT and Use Pardus with KDE on geForce Go 7200 on my laptop. Both use the effects without any CPU usage. Ah, FYI, KDE has software composite and XComposite is SW emulated if no accelerating HW is found.

Ahh, thanks. I guess I'll have to give those effects another shot...

[Tillin9]

I wholly agree that AMD/ATI needs to get its act together. I complained every driver revision about artifacts in ivview, an open source OpenGL program used by many researchers on Linux/ Unix workstations. I'm 100% sure if the developers ran it, then looked at the code they could figure out what OpenGL feature is causing all the trouble rather easily.

Not to mention 8.42 and higher have WORSE artifacting in this program than previous drivers. Also, with the new codebase I get a torn watermark in the corner. While performance may be better, I'd rather have old bugs squashed. Hint: with a stable and reliable driver, I can always just buy a faster card if performance is not good enough for my needs.

I'm probably not the only one who wants this, but if anyone from AMD/ATI is reading, please, pretty please release the R300/R400 3D specs to the DRI team. The Open Source driver for R300 cards is okay and with specs I'm sure they could get it perfect. I actually got Compiz working with the DRI driver, fglrx I just got a big white screen. Yes performance in many areas is slower and its not quite feature complete yet, but a lot of the really nasty fglrx bugs aren't there, and 95% of 3D works.

[Thetargos]

I know what you mean, I also work in the scientific 3D visualization as well, and extensively use Open Inventor based applications in my field. That's why we've had gone with nVidia cards for our lab computers which most of them run Linux. I really like ATI's hardware (seems more efficient [when working properly] than nVidia's), but without proper drivers, they're useless to us.

[Spydr4590]

I'm having a heating issue on the 1900xtx. I'm using the latest Linux ATI driver. The fan doesn't increase in speed and the card over heats. Works fine in windows. Sitting in linux overheats my card, but I hard lock on games in linux also..

[Alistair]

I'm not seeing any mem leak here..I'm running an x800pro agp though.
glxgears mem usage remains a steady 1.2% fgl_glxgears is a constant 1.4% both run for 15 min and observed with top... I've run UT2K4 under compiz-fusion for over an hour no mem leaks there either ..compiz-fusion had been up for over a week and I still had plenty of free ram..


DarkFoss
can you report a) kernel version b) install method used c) distro d) previous used fglrx version in the memleak thread?
thanks

[Svartalf]

I wholly agree that AMD/ATI needs to get its act together. I complained every driver revision about artifacts in ivview, an open source OpenGL program used by many researchers on Linux/ Unix workstations. I'm 100% sure if the developers ran it, then looked at the code they could figure out what OpenGL feature is causing all the trouble rather easily.


This implies that they've staffed the group appropriately and they have the manpower to do that check and fix things... :D

Having said this, they're missing functionalities (as in API edges that ought to be supported in a modern OpenGL 2.X compliant driver...) in the Vista driver that they did ship- and they're just as missing in the Linux side. While the Windows version is relatively stable, it's missing all the same API edges we're beginning to find on the Linux driver- do YOU think they're staffed up enough on the Linux side? :D


Not to mention 8.42 and higher have WORSE artifacting in this program than previous drivers. Also, with the new codebase I get a torn watermark in the corner. While performance may be better, I'd rather have old bugs squashed. Hint: with a stable and reliable driver, I can always just buy a faster card if performance is not good enough for my needs.


Yeah, that's kind of my take on things. It's my gut level feeling that the Vista/XP driver might be a house of cards like the Linux driver seems to be- it's just that the cards aren't toppling on them yet.


I'm probably not the only one who wants this, but if anyone from AMD/ATI is reading, please, pretty please release the R300/R400 3D specs to the DRI team. The Open Source driver for R300 cards is okay and with specs I'm sure they could get it perfect. I actually got Compiz working with the DRI driver, fglrx I just got a big white screen. Yes performance in many areas is slower and its not quite feature complete yet, but a lot of the really nasty fglrx bugs aren't there, and 95% of 3D works.

Heh... I wish they would just do that, then. It'd be a win for them and I would plug their stuff again. Even if we can't drive it QUITE as fast as they can or NVidia could with theirs, it'll be a vast improvement as it won't be this thorn in the side like the current drivers are. We'd probably have had a fix either done or on the way for some of these problems had it been an open source driver.

[yoshi314]

[quote]We'd probably have had a fix either done or on the way for some of these problems had it been an open source driver.[/qupote]
the binary nonsense will get us nowhere in linux. it causes more problems (esp with software compatibility) than it solves.

i can clearly see it after a couple of months using opensource drivers only. that's reason enough for me to never go back to fglrx.

[DarkFoss]

DarkFoss
can you report a) kernel version b) install method used c) distro d) previous used fglrx version in the memleak thread?
thanks

Done :) Let me know if there is anything else I can add.

[Alexander Heß]

I can't see any memleaking on my system either. I ran glxgears for like 8 hours and the memory usage didn't change at all.

I'm running Ubuntu 7.10 with kernel 2.6.22.

Driver has been installed as described on wiki.cchtml.com and thinkwiki.org. Only difference is, that I compiled a hacked fglrx.ko as described by Kano and ambro814.

Hardware is a FireGL V5250 in an IBM Thinkpad T60p.

[bridgman]

DarkFoss, what hardware are you running on ? This doesn't *seem* to be hardware specific since the same hardware is showing leaks for some folks but not others, but not ruling it out yet. This does seem to be more config, distro or kernel specific as Alistair suggested.

Thanks,
JB

[DarkFoss]

Hi bridgman,
Thank you for responding on these boards :)

My hardware is an AMD 3700,MSI K8T Neo2-F v1 board flashed to bios 9.3
Soundblaster Audigy 2 zs, 2 g of OZC (http://www.ocztechnology.com/products/memory/ocz_el_ddr_pc_3200_platinum) timings set to spec. x800pro agp at 8x fast writes enabled in the bios.
*edit agp aperture is set to 256 other lower setting don't seem to work as well*

[macmus]

the driver works on HD2600 with AIXGL.

But the behaviour i can only describe as that, nothing more ..

I am runing Gnome with compiz now and that is what i am experiencing atm:

1. Flickering of glxgears during rotation of cube and it does stay in place (does not rotate during rotation process)
2. Freaky flickering of movies.
3. Enourmous CPU consumption during compiz effect (minimalizing, scaling, cube rotating)
4. FPS drop down dramaticly during some of destop effect.
5. Internet browsers are choking incredibly.
6. Sometimes Gnome/KDE does start with black screen instead of login sreeen
7. Random X crashes during video playback.

For me it is just a joke not a driver...

[pedepy]

ive ran 8.42.3 a two seperate occasions on my mobility X1600 thing and this is what ive found:

benchmarks are pretty much the same in either aiglx, xgl, compiz, no compiz, glxgears, from either 8.40 or 8.42. However visible performance is choppier in compiz under 8.42's AIGLX. But the worse problem is no doubt video playback, which is just unwatchable.

It felt nice to run compiz without XGL for a change though; very nice ! :)

I think theyre on the right track.. Memory leaks and all that stuff are something to be worried about because it seems to stem from a lack of interest on the part of the developers, but aside from that you cant argue that its come a long way since two months.

Look at it that way: How many people wouldve believed you back in August had you said you'd be running an AIGLX enabled fglrx driver before year's end?..

I hope 8.43 is nothing more than bugfixes, if only to stop the crazy, bipolar, yoyo swing of high hopes and bitter disapoinments that all this AIGLX, fglrx 8.41 driver created in the first place. Theyre just drivers people and this one's only a month old!..

[macmus]

gief new driver so we can start whining on new bugs :)