AMD Linux 2008 Year in Review

[phoronix]

Phoronix: AMD Linux 2008 Year in Review

Last year when publishing our AMD Year in Review article there were numerous new features to account for, including but not limited to the new OpenGL driver, support for Compiz, and the AMD Catalyst Control Center Linux Edition. This year has been another interesting year for AMD's Linux efforts on both the open and closed fronts. We are focusing on their Catalyst driver efforts in this article, which has picked up support for CrossFire, is now capable of being overclocked with OverDrive, and AMD is now delivering same-day Linux product support. In this article we will recap some of the highlights from the Catalyst driver releases this year as well as set out on a benchmarking extravaganza.

http://www.phoronix.com/vr.php?view=13267

[sundown]

Great... but if you find some one in this forum who claims that fglrx doesn't suck -- now that will be an achievment!

[izual]

Great... but if you find some one in this forum who claims that fglrx doesn't suck -- now that will be an achievment!

agreed
Try KDE4 (4.1.3) with fglrx.
I found this on the opensuse wiki.

There seems to be an issue with the fglrx driver and KDE4, which causes windows to be painted/repainted very slow on KDE4, while this does not occur on KDE3. It doesn't matter if compositing effects are turned on or off in KDE4. Using the free radeon or radeonhd drivers seems not to cause such performance problems......


Well I can confirm this and I don't have the option to use radeon oder radonhd (Radeon4670/2600Pro) because then my video performance is terrible.

[miles]

When AMD had made the evolutionary step in ensuring equal Linux features and support, we were told by multiple AMD representatives -- even when in person at their launch event -- that some AIB partners would ship the Linux driver on their product CD and showcase Tux on the product packaging. It has been over six months and we have yet to see a single product that has either of these attributes. Hopefully this will come to fruition in 2009, as it's becoming apparent from our work and communications that more of AMD's partners are concerning themselves with Linux.

Don't quote me on that, but I remember reading on a mailing list (linux?) one of Linux main kernel developers say that graphic card vendors providing linux proprietary binary blobs would be really risky. Kernel developers don't see binary blobs kindly, and when they approached some lawyers for a check they were told the card vendor (or ATI/Nvidia) would have most chances to face huge damages in court. I expect we won't see Tux logo on the cards packaging if no Linux driver comes in the box.

Maybe when the open source drivers will prove stable and with enough feature they might provide them with the cards, but until them I wouldn't count on a Tux logo anywhere near an ATI/Nvidia graphic card. That might be different for Intel when they start selling dedicated graphic cards.

[Kano]

Well the driver "improvement" is for general use (the newly introduced features 99% of the ppl will not use anyway) is always the same:

a) feature X breaks
b) wait a certain amount of month
c) fix feature X - now listed as improvement

Of course not everything is fine with NV, but bugfix releases for latest gfx cards are much more often - ok, some problems took quite some time there too. But I can't remember a NV driver which locked the system when xv is used. Also it absolutely sucks, that features like beryl/compiz support is not working for Xorg 7.1.1, not even for Xserver 1.3. Also rendering issues for over 1 year, only community based kernel patches (NV provides em on their own) and hardlocks, especially when switching from radeon to fglrx have been never fixed. The whole driver is a complete mess - the monthly releases do not improve anything. If they would update the driver on purpose, like major bugfixes, new kernel support, better stability if needed - not when a month is over - then all would be more happy. Nothing hurts more then a driver which reached the timeout and gets released with more bugs than the predecessor.

[StringCheesian]

Don't quote me on that, but I remember reading on a mailing list (linux?) one of Linux main kernel developers say that graphic card vendors providing linux proprietary binary blobs would be really risky. Kernel developers don't see binary blobs kindly, and when they approached some lawyers for a check they were told the card vendor (or ATI/Nvidia) would have most chances to face huge damages in court. I expect we won't see Tux logo on the cards packaging if no Linux driver comes in the box.

I'm not a lawyer, but that lawyer sounds wrong to me.

AFAIK, publishing GPL code (like the Linux kernel) along with a binary blob would be copyright infringement - but only because you're copying the kernel in a way the kernel's copyright holders don't approve of. They don't approve of it being published next to a binary blob.

But publishing your own binary blob all by itself is perfectly legal. The copyright holder doesn't object, because YOU ARE the copyright holder of your own binary blob.

This is the reason Linux distros can't include the binary drivers on their install CDs, even though nVidia and Ati would be ok with it.

Again AFAIK, nVidia and Ati have every right to include their binary drivers for Linux in the box with any hardware they sell. They would just be disallowed from publishing any GPL code along with it.

That's how it works. You can't take GPL code and try to turn it closed source, and you can't mix GPL and closed source code. But if you're not mixing them at all, then it's fine.

Besides, the binary driver coming in the box is no different from letting people download it. It's still a form of publishing that falls under copyright law. If one is illegal then they both are. If that laywer is correct then nVidia and Ati both have to remove their binary drivers from their websites.

[Yfrwlf]

...linux proprietary binary blobs...

Kernel developers don't see binary blobs kindly...

To nitpick...the current video drivers are very proprietary, even the open source ones because they are drivers to be used by only one vendor. Also there's nothing wrong with binaries, you mean the problem is that they are closed source, of course open source is a feature, but at least Linux users can use their cards in the meantime until that feature becomes more common.

I expect we won't see Tux logo on the cards packaging if no Linux driver comes in the box.

That's silly, and I dunno if you're right about that, but regardless, to me it would be common sense to include the Linux driver if you put on the box that you support Linux. Demanding that users go online and download and install their drivers isn't the norm at least right now.

Maybe when the open source drivers will prove stable and with enough feature they might provide them with the cards, but until them I wouldn't count on a Tux logo anywhere near an ATI/Nvidia graphic card. That might be different for Intel when they start selling dedicated graphic cards.

Again I don't think you're right about that, but IANAL and don't know the ifs, ands, and buts with trademarks. IMO, you shouldn't need someone's permission to talk about them as I feel it violates free speech. If I want to say that my product "supports Linux" I should be able to, but obviously outright lies to the customer could come with a backlash. Either way, I don't think that effects whether or not the drivers you include are open or closed source, or even requires that you include any drivers at all in the box.

[bulletxt]

ok whatever. anyways, while doing a review , you must compare the product you review to its rival, in this case NVIDIA. Reading this article seems that NVIDIA has only VDP(blabla) more than AMD: SILLY!

come on, do a correct review and simply state: if you are a customer and want your card to work in 99% of cases then DON'T BUY AMD if you are using Linux(x86 and x86_64) because its drivers has more bugs than internet explorer ever did in 15 years. Instead buy NVIDIA.

This isn't a review, it's a friendly speech about AMD and its progress. But AMD's enemy is years ahead and we all know that. (if in year 2008 my operating system has chances to hang watching a video then...there isn't much to say about this).

It's not because I want to talk bad about AMD, but if you do a review how can't you not compare it to it's main rival? maybe because you know that would just hurt AMD? (actually totally kill).

Why don't you do an NVIDIA review starting from the first year they published drivers for linux? year 2001. yea.

[jeffro-tull]

Sorry, I must have missed the "evolution" in the driver. Yeah, overclocking. Yeah, Crossfire, whatever. Things a good chunk of people don't do.

Most of the benchmarks, the performance across the year was comparable amongst all releases. A few of them, the latest releases wouldn't run. Yep, that's evolution.

Every month when a new fglrx release comes out, soon after there's at least one thread on here of "new catalyst broke fearure X" or "new catalyst won't install" or "new Catalyst freezes when I look at it funny". You might have great luck with your particular card, your particular system, and your particular distro. Someone else could have one seemingly insignificant difference in their setup and the driver will bork on them. I've never heard of a more finicky driver.

[bulletxt]

Sorry, I must have missed the "evolution" in the driver. Yeah, overclocking. Yeah, Crossfire, whatever. Things a good chunk of people don't do.

Most of the benchmarks, the performance across the year was comparable amongst all releases. A few of them, the latest releases wouldn't run. Yep, that's evolution.

Every month when a new fglrx release comes out, soon after there's at least one thread on here of "new catalyst broke fearure X" or "new catalyst won't install" or "new Catalyst freezes when I look at it funny". You might have great luck with your particular card, your particular system, and your particular distro. Someone else could have one seemingly insignificant difference in their setup and the driver will bork on them. I've never heard of a more finicky driver.

exactly.... there are more threads on fglrx drivers not working than anything else! not to mention how many howtos and wikis to actually install the driver!! why isn't this mentioned in the review? why doesn't this review actually dedicate 1 page about the infinite bugs the driver has? why why why? who knows.. I suggest to put an AMD flag as logo for website. at least if a user enters this website he knows it's a pro-AMD. and I'm not inventing anything, I just read and extract clear conclusions...

and all this is said from an ATI customer. (yes, I love dreaming).

[mityukov]

Agree. There's a huuuge lack in good-working "must-haves"..

1). 2D-accel(should be good with GTK, Qt3, Qt4, whatever)/Video/OpenGL(of declared version) --> it's the FIRST priority.

2). Wine --> 2nd.

3). Overdrive/Cross-Fire/what else?? --> these are the latest things to spend developer-hours on!!...

:-/


Still using Radeon.. No reason to change LESS-PERFORMANCE FOSS driver to STILL BUGGY IN THE MOST NEEDED AREAS proprietary one %)

[bugmenot]

When I read the title I expected a summary of the things that have changed on the amd side on the linux platform. But this review is only about the closed source driver, so I'm kinda disappointed.

[Yfrwlf]

I took it that they were planning on talking about the progress of the open source driver in a different article, but we'll see.

Regardless, yes, this was an article focusing on AMD only and their progress on Linux. That's a good thing, there's no denying that, but obviously if you want to actually recommend a card for consumers to buy, you should review a wide range of brands and cards. This wasn't that article. That article would be called AMD vs. Nvidia vs. Intel, or something. This article was AMD Linux 2008 Year in Review. AMD. AMD. So, if you want a good comparison, which most of those who browse this site would like to see, it hasn't happened in a while, so I agree:

Phoronix, many of your readers I think would like to see an all out comparison of those three mentioned companies and their latest drivers on a good sampling of their cards and how they actually perform under the most common Linux tasks (common games, desktop performance, video playback).

More more more, give us give us, whine whine whine. ^^

No, but seriously, it was a good review on AMD though, just responding to some of the comments here. Funny/sad about the latest drivers not working with that Wine benchmark though...I'm sure there are several Wine games which can be benchmarked in both OGL and DX tho.

[Michael]

When I read the title I expected a summary of the things that have changed on the amd side on the linux platform. But this review is only about the closed source driver, so I'm kinda disappointed.

Something on the open-source status is already written and is just waiting for a special date.

[Yfrwlf]

I know, to help with buying hardware, not that you need it, but you could have an upgrade-a-thon, and tell your readers to buy a specific system setup for benchmarking but getting different video cards, and then everyone can submit their scores, and it'll be a massive community bench marky...event...thing. :P

[miles]

I know, to help with buying hardware, not that you need it, but you could have an upgrade-a-thon, and tell your readers to buy a specific system setup for benchmarking but getting different video cards, and then everyone can submit their scores, and it'll be a massive community bench marky...event...thing. :P

I think it's possible without such a hassle, the only problem remaining (easy way to compare multiple setups with the web interface) will be solved when the next big version of Phoronix global arrives.

You don't really need people to buy new hw - there will be enough gamers with similar CPU (Core 2, and if you're really a benchmark freak you can also ask people to downclock their processor), deactivate some RAM so everybody has 2Go (but realistically there's no need, games don't show differences over 2Go).

I think the only difficulty is agreeing on standard resolutions, a set of tests and the games/gx bench.

I'd suggest:
- 1024x600 (so netbooks can do it)
- 1440x900 (most gamers)
- 1920x1200 (Full HD monitors, close to 1920x1080 for those playing on TV)

And maybe 25XXxXXXX for those with a big monitor.

Everybody could run the first two test, a relevant number could run the third. Just eliminate those that have a Core2 < 2Ghz or less than 2Go of RAM, and the rest just don't overclock or downclock.

[Kano]

I would say there are more 1680x1050 displays out now than 1440x900.

[miles]

I would say there are more 1680x1050 displays out now than 1440x900.

I agree, but since gamers have been choosing 19' for years since that was the uper resolutions were considered to taxing, I though most gamers were still using that. 1680x1050 might also be too big a gap coming from netbook resolution. I wouldn't mind running tests in 4 resolutions instead of 3 though.

All aside, my 7900GS is still ok in 1920x1200 for all Linux games (except maybe the ET:QW demo), so I'm still not sure we really need benchmarks. What would be really informative is a table with supported features and - possible, but far more complex - list of games playable through Wine with each brand. For benchs, I'd expect anything starting from the HD46XX/9600GT to run everything up to FullHD with no hiccups.

[denali]

I'd like to own some AMD cards, but the primary use of my home computer is Wine and World of Warcraft. From what I've read, using an AMD card with that software combination is simply an exercise in frustration. Unless thats changed with 8.12... (I somehow doubt it has, tho)

[gururise]

When my GeForce 8500GT died, I bought a Radeon 4670 based upon reviews on Anandtech. Boy, was I dissapointed at the quality of fglrx. Things have improved slightly with Catalyst 8.12; however, the number one complaint is that I cannot play videos w/o flickering when running Compiz. I have to disable Compiz to get smooth video playback! C'mon, Nvidia has this problem solved in their driver, why can't ATI do it? Do we really have to wait for DRI2 for a fix?

[NeoBrain]

I'd like to own some AMD cards, but the primary use of my home computer is Wine and World of Warcraft. From what I've read, using an AMD card with that software combination is simply an exercise in frustration. Unless thats changed with 8.12... (I somehow doubt it has, tho)

Running Warcraft 3 in Wine works fine for me (though I have to workaround a _wine_ bug that prevents fglrx from using 3d acceleration, see Wine bugzilla 13335), so I guess WoW will play nice with fglrx, too.
However, your mileage may vary, fglrx behaves different on each computer ;D

[sundown]

C'mon, Nvidia has this problem solved in their driver, why can't ATI do it? Do we really have to wait for DRI2 for a fix?

Coz they need a year to decide whether they want to use DRI2 or not and then another 6 months to write the code and another 6 months to implement it.

Seriously, recent word is they will not use DRI2. :D... so expect your video to work fine in Q4 of 2010.

[bulletxt]

yea I already said some days ago that I'll be old before I'll see my ATI GPU working as it should. I'll have a lot of shrinkles on my face that day.

[val-gaav]

C'mon, Nvidia has this problem solved in their driver, why can't ATI do it? Do we really have to wait for DRI2 for a fix?

Maybe nvidia should write their own X server then ? I mean they solved DRI issue so why not contribute that to Xserver ? Why not contribute something beside the limited nv driver ?

[NeoBrain]

Maybe nvidia should write their own X server then ? I mean they solved DRI issue so why not contribute that to Xserver ? Why not contribute something beside the limited nv driver ?

I guess the xorg developers wouldn't even accept that code as it's probably quite ugly and very nv-specific (i.e. optimized for the binary blob and not usuable for radeon/intel/etc).

[mityukov]

I guess the xorg developers wouldn't even accept that code as it's probably quite ugly and very nv-specific (i.e. optimized for the binary blob and not usuable for radeon/intel/etc).

What will be the benefit for NVidia? At the moment, NVidia is a good choice for those ones, who are desperately looking for something compatible with *nix (especially Compiz-lovers).


By the way, I like the Intel's and AMD's position in this question: there's a set of good standards/architectures, in X world. You can write your closed blob, which hacks/overrides these standards, or, you can write driver that comply to them, and therefore, give X guys a reason to evolve "base". And even help them in doing so, like Intel does, for example.

[Yfrwlf]

give X guys a reason to evolve "base". And even help them in doing so, like Intel does, for example.

That'd certainly be best, help keep as much as you can out of the drivers and in X instead, so the drivers are merely only the parts needed for direct communication with the hardware. That's part of the goal of Gallium 3D is to simplify the drivers much more by doing this. Definitely a smart programming move if they can pull it off which I guess they have/will. Hell, it'd be nice to just make generic video card drivers like with USB standards or the firewire standard, have a universal standard which simply grows over time. It's just a matter of using the same language is all, completely possible, and it'd really help every OS out there.

(Maybe that's why it's not wanted, maybe it's partly due to MS's monopoly and them trying to compete against other OSes by making drivers be more exclusive to their OS, I dunno.) ^^

[Muesli]

1). 2D-accel(should be good with GTK, Qt3, Qt4, whatever)/Video/OpenGL(of declared version) --> it's the FIRST priority.

I agree. The state of 2D acceleration in fglrx is ridiculous. I use konsole with a visual alarm (black/white flash instead of loudspeaker beep). With fglrx and its slow text output, this is no flash anymore but some slow filling effect. Restoring a text window takes about a third of a second while with other drivers this is instantaneous.


2). Wine --> 2nd.

Yes. With fglrx you can forget about anything more recent than World of Warcraft - see Wine FAQ and the postings linked there.

Another bug: For me, Xine reproducably crashes the whole machine with some videos (HD 3xxx, fglrx 8.11). With nvidia I last saw this happen in 2003 or something.

...and to top things, as you said already, instead of fixing these long-standing bugs, ATI seems to work on new features. Really, who needs some let-programs-run-on-GPU feat if text consoles feel sluggish and video playback occasionally makes the system crash? Not that I would ever expect these new features to work properly if even the basics don't work...

Michael, PLEASE, in next articles about fglrx, mention the horrible basic state the driver is in. I definitely dislike NVIDIA, and I like the fact that ATI released specifications a lot. Based on articles like the one you wrote I purchased a recent ATI card and was quite horrified by the state of things when trying it first-hand. The card has been returned, and I will probably stay away from ATI graphics for quite some time now even though I'm an open source advocate.

Old ATI cards (X1xxx etc.) with OSS drivers may be cool (if you don't want to play recent games in Wine, that is).

Recent ATI cards (HD3xxx/4xxx) with the fglrx driver definitely don't cut it. They suck both in 2D (with no OSS alternatives ready yet) and 3D/Wine. And you don't want fglrx if you want your system to run stable.

Don't pretend fglrx is mature enough for daily use.