Forgot your password?
typodupeerror
AMD Linux

AMD Publishes Open-Source "ATI Evergreen" Driver 159

Posted by ScuttleMonkey
from the still-watching-and-waiting dept.
Several readers have written to tell us that AMD has published their code to support the Radeon HD 5000 "Evergreen" graphics cards on Linux in an open-source driver. Unfortunately the driver isn't quite as complete as some might hope. The current offering doesn't promise 2D (EXA) acceleration or 3D support. "The DDX driver supports mode-setting on the Evergreen/R800 series GPUs with VGA and DVI connectors while the DisplayPort connectivity is still not working right, according to AMD's Alex Deucher who had written most of this code. These new AMD graphics cards have been around since September while there was no open-source support at that time. In December just before Christmas there was Evergreen Shader documentation that was made publicly available and around that time it was confirmed via our forums that initial VGA mode-setting was working with Evergreen internally on unreleased code. Since then the digital connector support has been added in and this code has finally cleared AMD's legal review. The revised target was to publish this code by FOSDEM, which is this weekend so AMD did hit the target this time."
This discussion has been archived. No new comments can be posted.

AMD Publishes Open-Source "ATI Evergreen" Driver

Comments Filter:
  • Baby Steps (Score:2, Insightful)

    by mystikkman (1487801)

    Atleast they have released the specs out to OSS developers and are working towards a accelerated solution. There used to be a time when only Nvidia cards used to run at full power on Linux. Go AMD!

    • by XPeter (1429763)

      There used to be a time when only Nvidia cards used to run at full power on Linux. Go AMD!

      We still are in that time, as TFA states: "The current offering doesn't promise 2D (EXA) acceleration or 3D support." Well what IS the driver going to do, then?

      Hopefully due to the profitable quarter they just had, AMD/ATI can hire some Linux devs to get the ball rolling faster.

      • by goldaryn (834427) on Monday February 01, 2010 @07:39PM (#30989232) Homepage

        Well what IS the driver going to do, then?

        Run Nethack at 120fps?

        • Re: (Score:3, Funny)

          A 16550 doesn't really count as a graphics card anymore...
          • Re: (Score:2, Insightful)

            by babblefrog (1013127)
            You kids with your fancy-schmancy hardware. In our day, we had 8250s and we liked it.
          • by sirinek (41507)

            A 16550 doesn't really count as a graphics card anymore...

            That was the funniest comment in this entire thread. But it flew right over most of these lil whippersnappers heads.

      • Re: (Score:1, Interesting)

        by Anonymous Coward

        For the last decade (almost), it was always "free the specs and the community developers can do it". It's interesting that now it's "You've freed the specs, but now you need to hire the devs".

        The community is no closer to having community maintained drivers. If the IHVs move out of supporting the drivers, where will that leave the community? The specs are there for Intel and AMD, but there doesn't seem to be a large non-vendor set of developers out there.

        Sure there are a set of core developers (airlied,

      • Re:Baby Steps (Score:5, Informative)

        by MrHanky (141717) on Monday February 01, 2010 @08:16PM (#30989650) Homepage Journal

        The driver is functional for your regular 2d needs. Browsing the web and moving windows around is fast enough. It doesn't crash all the time (only tried it for a few hours, and no problems so far), but it does lack video overlays, so it's not quite ready for media use just yet.

      • Or they could just release te specs...

        For a while I hoped AMD would play nice with FOSS. Hell, it looks like it is their only chance of surviving, but still, their CXX people doesn't seem to agree.

    • by ak3ldama (554026)
      Badly supported 5xxx series cards merely means people will utilize cheaper older cards for Linux. I have a 4670 card with Fedora 12 and the experimental open source drivers enabled. Linux overall, and the drivers work great. Super cheap systems with ATI graphics and AMD processors with Linux is a decent proposition. Not perfect on the desktop, but taken with a few salt shakers it's ok.
    • by thsths (31372)

      > At least they have released the specs out to OSS developers

      Did they? Or did they only release the specs for 2D? And why is there no word of kernel based mode setting? It seems a bit silly do develop a new driver without it - after all it is clearly the correct solution, and everybody else is moving that way.

      > and are working towards a accelerated solution.

      Not good enough. At least 2D acceleration (Xv etc) is essential, and with a modern desktop you want at least some 3D functions, too.

      > There use

      • AMD / ATI has some very nice closed source drivers. Been using them for a goodly time now, and they've vastly improved over the years. Now my dual monitors work, with full 3d acceleration, and comparable framerates in Doom 3 for me.

        However, this is about the open source drivers that were just released in their absolutely initial form. If you'd have liked them to withhold their opensource project from outside eyes and outside hands until completion, so that there was no chance for community input, then
  • by Darkness404 (1287218) on Monday February 01, 2010 @07:37PM (#30989192)

    The current offering doesn't promise 2D (EXA) acceleration or 3D support.

    So if it doesn't offer 2D acceleration or 3D support... what does it do? Framebuffer mode? Seriously why would ATI even release a driver in this pathetic of state, at least when I can buy an nVidia card for the same amount and have 100% of features work just fine.

    • by Jurily (900488) <jurily@NETBSDgmail.com minus bsd> on Monday February 01, 2010 @07:42PM (#30989254)

      why would ATI even release a driver in this pathetic of state

      To gather developer attention. At this stage, it's not about the features, it's about the mindshare.

      • Re: (Score:2, Insightful)

        by Jorl17 (1716772)
        More than that, it's about manipulation of the media. It's about image. It's about making stupid people like some of us believe that this actually is a great leap forward. Even if this ever goes somewhere, it will always have started with this objective and this purpose.
        • It's not a "great leap" - nor anybody involved claims that it is.

          But this is nevertheless important statement from AMD/ATI that they are still on the OSS track. What is IMO rather important and significant.

          P.S. Plus it is extremely important to folks who run systems which are not supported by nVidia. Unless you run x86 or x64, nVidia wouldn't even acknowledge your existence.

      • Re: (Score:3, Interesting)

        by Darkness404 (1287218)
        It doesn't matter. Drivers should be -standard- unless I'm screwing with something seriously experimental, drivers should be expected. The 5000 series has been out since last year, so drivers should be standard on the day they ship.
        • What do you mean by "standard"? If you mean "standard across devices of similar type", then different companies have different expectation for what the OS is responsible for and what the device is responsible for.

          If you mean "standard for a particular hardware", the driver still has to interface with the OS, and OSs change over time. Drivers need to be recompiled and modified over time to allow them to continue functioning.

        • Why should drivers be standard for Linux? Considering that it represents a pretty minuscule percentage of their market it seems to me they are going above and beyond by supporting it at all. On top of that they have "standard" drivers for previous cards.

          Seems to me they are doing the right thing.

    • by electrosoccertux (874415) on Monday February 01, 2010 @07:45PM (#30989284)

      FTFS:

      The DDX driver supports mode-setting on the Evergreen/R800 series GPUs with VGA and DVI connectors

    • by Sycraft-fu (314770) on Monday February 01, 2010 @07:45PM (#30989290)

      Well they have a working proprietary driver. This is just the OSS one. ATi is attempting to make nice with the "OSS only" crowd. The problem is they can't just open up their normal driver, it contains licensed, patented code from other companies they can't hand out. As such you get a very different, stripped down, driver for the OSS community to work on. How useful this is is something up for debate.

      nVidia's approach is that they only want to release the proprietary, fully working driver. They aren't interested in releasing a semi-broken driver just to be OSS. As such you only get their binary download.

      Now in either case, the nVidia drivers are superior quality wise. They've always been good at drivers on Windows, and it translates over to Linux it seems. However ATi does have an open option that nVidia doesn't. For some people, this is important as they won't run closed code at all, even if it means a better experience.

      • by Draek (916851) on Monday February 01, 2010 @08:15PM (#30989648)

        Well they have a working proprietary driver.

        For varying definitions of "working". As an ATI user I must say, the propietary driver is the single worst piece of software I've ever had the displeasure to run on my Linux system, and the only thing besides faulty RAM and a dying HDD to ever cause Linux (yes, the kernel, not just X) lock up on me. It sucks so badly that ArchLinux even removed it [archlinux.org] from their repositories, prefering to not give it as even an option rather than deal with the support nightmares it causes.

        The Open Source driver on the other hand is excellent, stable and completely hassle-free (something I can't quite say of NVidia's propietary driver, though it wasn't nearly as bad as ATI's), and even supports 3D acceleration on older chipsets. My guess is that it won't be long until 3D is also supported on the HD5x00 series as well, development is quite fast on it.

        • Last ATi card I had was a Radeion 8500. The Windows drivers came in two flavours, the Microsoft Certified ones that didn't support a recent version of DirectX, and couldn't actually run anything that used the GPU in a nontrivial way, or the ones from ATi which ran software reasonably well until you hit a bug in the driver and the kernel crashed. Sometimes the Creative Labs SoundBlaster Live drivers would crash the machine before the ATi ones had a chance to (same situation there; you could use the MS Cert

        • Re: (Score:2, Troll)

          by drinkypoo (153816)

          For varying definitions of "working". As an ATI user I must say, the propietary driver is the single worst piece of software I've ever had the displeasure to run on my Linux system, and the only thing besides faulty RAM and a dying HDD to ever cause Linux (yes, the kernel, not just X) lock up on me.

          ATI drivers have been causing me crashes in various operating systems since they brought out the Mach32. Their subflavors of the Mach64 caused me crashes in Windows NT 3.51, and Solaris x86 2.5.1. The catalyst drivers were amazingly foul and had a footprint I would never have believed had I not installed many, many drivers trying to get their driver to stop cratering Windows XP. ATI has never been able to develop a worthy driver, and their donation of OSS drivers is worse than useless, as they can use it as

          • While some of my recent negative moderation may have been justified, I've never been down-modded so much, so quickly. Am I really being attacked by Scientologists? Pretty amazing. This post, however, is NOT a troll; I believe all that I have said. I will stand by it until I die.

        • by Curtman (556920) *
          I agree. I'm sticking with my HD3200 and free drivers for the forseeable future. The video playback is beautiful compared to what it looked like with nvidia. The proprietary driver didn't seem to work at all for me, but I had two free drivers to choose from, "radeon", and "radeonhd". Both work great, and I don't even need an xorg.conf unless I want to choose radeonhd over the radeon driver.
        • by BitZtream (692029)

          had the displeasure to run on my Linux system

          The windows drivers are shit too.

          I supported ATI for a good while, but the quality of their drivers is so horrible that I refuse to have anything to do with them anymore. I'd rather use a crappy intel integrated chip with stable drivers than the crap ATI pedals.

          At least the OSS driver may be more stable since others can contribute fixes.

          I'd bet the OSS driver is more useful to ATI for bug fixes than anything else, and wouldn't be entirely surprised if it became

      • by Schraegstrichpunkt (931443) on Tuesday February 02, 2010 @12:14AM (#30991470) Homepage
        I'll run closed code, but it's bloody well not going to be something as crucial as my video drivers. I've done it before, and I'll never do it again on my main computer.

        How many years did it take for nVidia to add DRI support to their driver? Xinerama support? Not-corrupting-the-virtual-console-when-running-more-than-one-instance-of-X support? Do they support XRandR 1.3 yet? (That last question isn't rhetorical---I've stopped following the status of nVidia's proprietary drivers.)

        The last time I used them, the nVidia drivers exhibited a severe case of Not-Invented-Here syndrome, and they weren't particularly stable.

        I really don't know where all these people come from who say "nVidia's drivers just work". I suspect it's just a lack of experience with *actual* stable drivers. The best X driver experience I've had is with free drivers for hardware that's a few generations old. Super stable and everything *really* just works.

        • by Kjella (173770)

          I really don't know where all these people come from who say "nVidia's drivers just work".

          I guess people like me - that use distros with driver versions we know work, if nVidia doesn't support the latest kernel I don't upgrade. There are things I haven't been able to get working, but I've never had a crash I could trace back to the closed source drivers. Unlike Catalyst, which I had one really, really bad experience with from back before AMD bought ATI. To put it this way, if nVidia's support was bronze then ATI finished last with a broken leg. Open source is stable yes, but Intel always sucked

        • by HonIsCool (720634)
          nVidia doesn't have DRI support. Why? Because DRI is totally broken and they were forced to come up with their own framework instead. Check http://linuxhaters.blogspot.com/2008/06/nitty-gritty-shit-on-open-source.html [blogspot.com]
    • by zyklone (8959) on Monday February 01, 2010 @07:46PM (#30989304) Homepage

      Please tell me where nVidia has an open source driver with 100% features working?
      This is about the AMD open source driver, not about the AMD closed source drivers, which supports Evergreen just fine.

      • by Yaa 101 (664725) on Monday February 01, 2010 @07:50PM (#30989364) Journal

        Please tell me where AMD has any good Linux driver? Their closed source driver is such a piece of crap.

        • Re: (Score:3, Insightful)

          by zippthorne (748122)

          Good point. Let's take this opportunity of AMD doing something for the open source community to bitch about all the stuff they didn't do yet. Way to play right into nVidia's hands, smart guy.

          • by Techman83 (949264)
            Any direction towards providing working open source drivers to the community is a good one. But they've been promising lots of things for a long time and failing to deliver. My notebook has both intel and ati graphics, currently Intel's oss drivers are fantastic. Who'd a thunk it, slowest piece of crap in the windows world, performs quite well in the F/oss world (admittedly no gaming, but it handles compiz dual screen quite nicely).
          • Bottom line: nVidia closed source drivers are better than AMD closed source drivers, and AMD open source drivers are feature-incomplete.

            That about right? Or am I "playing right into nVidia's hands" like the GP? If so, nVidia's hands is where I want to be, because their drivers work.
          • by Yaa 101 (664725)

            It's called critique, and it's supposed to make people who are on the receiving end of that think better.

    • by WeatherGod (1726770) on Monday February 01, 2010 @08:23PM (#30989716)

      It gives a basis for the community to work around. The entire issue with NVidia has been that developers have been asking for at least some sort of documentation so that they don't have to reverse-engineer everything. Companies have said that they don't want to support Linux or handle bugs, and we reply "you don't have to!". With documentation and a core set of code to work around, AMD has done exactly what we have asked for. Now, it is up to us to take that code and build it up to be a full-fledge graphic driver.

      AMD/ATI has nothing but my fullest appreciation for what they have done.

    • by ppanon (16583) on Monday February 01, 2010 @08:43PM (#30989924) Homepage Journal

      Your post is either erroneous or misleading - ATI has closed source Catalyst drivers [lwn.net] that support Evergreen cards with 2D and 3D acceleration. What the Evergreen cards haven't had up until now is the open source driver support. However, the open source driver support for NVidia cards is much worse because the developers are having to reverse engineer functionality from NVidia's closed source drivers because NVidia hasn't made any open hardware specs available. When it comes to open source driver support for 2D and 3D acceleration, NVidia lags far behind AMD/ATI and Intel. As a post in the above link indicates, in the long run the shared open source code base eventually will be a significant competitive advantage for Intel and AMD and a disadvantage for NVidia.

      I have switched over to the open source AMD R600 drivers because, even though the 3D support is not yet quite as good as the closed source drivers, it should catch up and it's already good enough for what I do. In the meantime I won't have to worry about waiting a number of months for the closed source drivers to become available when a new distro/kernel release requires new binary blobs from the vendor. That also means that my graphics hardware investment is protected and not dependent on the continued support of the hardware vendor if I want to continue to upgrade the O/S in the future.

    • by LWATCDR (28044)

      Not exactly. You can not buy an nVidia card for the same amount and have 100% of the features work just fine using FOSS drivers.
      However you can get closed drivers for both ATI and Vidia.
      Just goes to prove that the myth that if you just release the specs great FOSS drivers will get written quickly and by the community.

    • >So if it doesn't offer 2D acceleration or 3D support... what does it do?
      unaccelerated 2d

      It is just like any other code push to upstream. You get a feature working on your local branch, you push it up. Then go on to the next feature. And this is not a standalone driver, it is just new code for the driver that currently handles r600 and r700.

  • Nightmare (Score:4, Interesting)

    by Anonymous Coward on Monday February 01, 2010 @07:46PM (#30989302)
    There is some amount of ridiculousness here - perhaps I don't understand something but explain to me if you will, how this works - Every time a new GPU is released, a shit load of new driver code is required to just get it working. And then there a a truck load more of code required to get 2D acceleration working. And then the same for 3D. How come the GPU vendors do not have a freaking portion of their hardware always work the same way, with same driver code - it just does mode setting and sets up the GPU for decent level of 2D acceleration. The you write a per GPU, dynamically loadable module that will deal with that particular family of GPU. I mean there is not a whole lot you can do with modesetting and 2D - no one cares of 2D accel anymore - it should just work the same way with same driver code for all series of GPUs for a particular vendor. NVidia has to drop support for older chips, fork the driver and have it only support newer chips because of bloat that it becomes having to support different families of GPUs each requiring lots of code.
    • Re:Nightmare (Score:5, Informative)

      by MostAwesomeDude (980382) on Monday February 01, 2010 @11:32PM (#30991252) Homepage

      Actually, we're fortunate to be able to reuse most of our code on lots of GPUs -- there are some bits that apply from r100 to r700. The "fun" is in the sheer complexity of the hardware, the inability of the hardware to cope with incorrect programming, and the lack of documentation, manpower, and testing available to assist us.

      Oh, and hardware bugs. You don't wanna know how many there are. Really. Try getting an RS480, or RC410, to do 3D. It ain't fun.

      • by RAMMS+EIN (578166)

        You wrote you need more manpower and testing. How hard is it for, say, someone proficient in C to get into video driver development? Do you have any pointers to materials that would help one get started?

    • by Kjella (173770)

      How come the GPU vendors do not have a freaking portion of their hardware always work the same way, with same driver code - it just does mode setting and sets up the GPU for decent level of 2D acceleration.

      Because it doesn't have a 2D engine anymore, it's all treated like a special case of 3D where z is always 0. That is just one example, there's a lot of rewiring going on inside the pipeline. Try for example reading this page [anandtech.com] on how nVidia is changing their fixed function pipeline and see if you manage not to get a headache. The drivers have to work the new way even just to achieve the old ways, it's not like CPUs where you slightly extend the x86 interface.

    • by squizzar (1031726)

      "no one cares of 2D accel anymore" - yes they do. Try running your desktop (whatever OS) with the 'standard' drivers. Yes that's right, there is a default way of running most graphics cards, probably VESA or something. How else do you think you can install your OS without the correct drivers from the get-go. Likewise, the bios screens etc. all operate through a standard interface (VGA adapter?).

      2D Acceleration is probably more important that 3D, it just doesn't get noticed as much. Try dragging/resizi

      • That's not 2D acceleration. The big win in desktop performance is from accelerating compositing. Every app draws to a private display buffer and these are then composited by the GPU. This is basically performing something like a scaled multiply operation, but on every single pixel. It's very slow on the CPU, but the GPU can typically do something like 128 pixels simultaneously, while the CPU can do 4 if you're lucky. If your windowing system doesn't do compositing then every window has to redraw when i

  • The 4800 series of cards have excellent acceleration support with the radeon driver these days (the latest source releases). It's not quite fast enough to play Nexuiz at high settings but compositing runs great and the desktop is stable! I honestly can't remember a time when the proprietary driver wasn't locking up or corrupting the screen, so this makes me pretty happy as an AMD owner.

    Hopefully it's only a matter of time until the 5000 series is supported -- the proprietary driver just isn't an option if
    • Re: (Score:1, Interesting)

      by Anonymous Coward
      Dynamic power management still sucks for laptop users though. I have a laptop with a HD3650 GPU and it runs terribly hot with the default Fedora 12 and Ubuntu 10.04 Alpha drivers. If I use the "ForceLowPowerMode" it runs cooler, but the radeon driver will corrupt the cursor and radeonhd will slow down. With the Windows 7 driver it never runs hot.
  • by MostAwesomeDude (980382) on Monday February 01, 2010 @08:05PM (#30989534) Homepage

    Some reading between the lines is needed.

    Any r600 acceleration code *should* work with only minor tweaks on Evergreen (r800). The biggest changes are supposedly in GPGPU-land; r800 supports a lot more shader instructions than r700 or r600.

    I don't have one of these yet, but I'm sure Cooper and Richard, the AMD 3D devs, are furiously coding away to make stuff run.

  • by markdavis (642305) on Monday February 01, 2010 @08:24PM (#30989720)

    >and this code has finally cleared AMD's legal review.

    Has nothing in it that we feel might be secret or licensed....
    Doesn't do 3D CHECK
    Doesn't do 2D CHECK

    Passes our legal review- let people enjoy it now!!!!

  • 'The driver doesn't offer 2D or 3D acceleration' -- o.O

  • Thats why I stay with Intel graphics, they may not be the fastest graphics card in the world
    but their Linux drivers never made any problems. Be it mode switching or dual-head, everything *just works*.

    And that by using existing Xorg standards (XRandR, etc.), e.g. no nightmare "nvidia-settings" program as in the case for "the other" manufacturer.

    And yes, for most models the drivers are open source, IIRC.

  • by 7-Vodka (195504) on Tuesday February 02, 2010 @12:00AM (#30991418) Journal

    I've had a R600 based radeon HD3870 on linux for a while. I used to go with the fglrx proprietary drivers, but I've recently switched to the radeon (ddx) driver.

    I have to say I'm extremely happy with the Open driver. Now is it Free Software? I'm not sure, I mean *most* of the driver seems to be, but you still need to load microcode firmware.

    As far as the quality of the driver though, it seems very bug free and the kernel mode setting is awesome. Switching from vt-1 to an Xorg session for example is instantaneous. Mesa seems to need some work on the 3d side, but you can play quakelive on it and run the kde version of compiz.

    It's really great that ATI has both released documentation and paid developers to work on getting these drivers up and running, they should consider both sponsoring some 3d work on the mesa side and also figuring out a way around the microcode situation.

    • Microcode firmware is control software for the hardware. It's like a BIOS but stored in RAM on the hardware expansion card itself. The other method for this would have been to load the microcode as permanent flash, rather than in the driver, and not bother publishing updated versions or just require you to bios-upgrade your graphics card. Basically, they've moved that data from one place to another; the "way around the microcode situation" is to push it out of sight, literally, just to get people like y
      • by 7-Vodka (195504)
        Or the microcode itself could be distributed under the GPL.

        Is there some reason you ignored this ideal solution?

  • by Exter-C (310390) on Tuesday February 02, 2010 @02:42AM (#30992190) Homepage

    Its great to see some hardware companies coming out with open source drivers for their technology. Even if the driver is so far incomplete its at least a good starting point which will hopefully be improved on. I feel that by providing this sort of information AMD may have a repreive which will help it have a fighting chance in the future. Its such a shame that Via have not been doing more with their graphics drivers in Linux. I really wish that openchrome had more support given that so many cheap nettop/netbook style systems have via chipsets (at least in asia).

  • Now if only they'd release an open-source Windows driver....
    • by BitZtream (692029)

      Port it. All the required tools are free, I'm pretty sure the Windows DDK is free as well. I get it from an MSDN subscription so I haven't actually tried to find it without one, but I seem to remember it being there if you look for it.

      You have the source for a Linux driver, shouldn't be that hard to port it for someone who understands the WDDM model, or hell the 2000/XP model since the this driver isn't accelerated anyway.

Parts that positively cannot be assembled in improper order will be.

Working...