Announcement

Collapse
No announcement yet.

Intel Comes Up With An Alternative To Bringing Back UMS

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Intel Comes Up With An Alternative To Bringing Back UMS

    Phoronix: Intel Comes Up With An Alternative To Bringing Back UMS

    The Intel Linux driver has been challenged by stability problems and other issues for owners of i8xx hardware since they rolled out kernel mode-setting and Graphics Execution Manager (GEM) support with UXA 2D acceleration more than a year ago. There were initial problems for other Intel users as well when switching to this overhauled driver stack -- to the point that it killed the netbook experience -- but those problems were quickly worked away. But for those using Intel's oldest supported hardware under Linux, the problems to this day remain. To circumvent this issue there's been the approach to add back user-space mode-setting to the Intel driver with EXA 2D acceleration to simply avoid these problems rather than correct the actual issues with KMS/GEM/UXA, but now another alternative has emerged...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    OMG
    Next time I will buy an ati powered laptop instead...
    ## VGA ##
    AMD: X1950XTX, HD3870, HD5870
    Intel: GMA45, HD3000 (Core i5 2500K)

    Comment


    • #3
      LMFAO!

      "Duh... we're too dumb to do it the right way, and nobody wants the old way, so here you go.... nothing at all. Enjoy!"

      Comment


      • #4
        Having dealt with one of these chipsets under Windows and being shocked at how shitty it was, I can understand that it would be difficult to make it work properly. Based on what's in this article, I can infer that it would be unreasonably difficult to make these pieces of crap work properly with DRI2 at all. That said, I still have a hard time believing that they can't spare at least 2D acceleration.

        Here's my alternative: fork an older UMS driver that worked, make it work reasonably well with modern X server releases, call it "intel-legacy", and don't worry about supporting i8xx in the main "intel" driver. Aim to maintain the level of quality that existed before the KMS switch. Don't worry about making OpenGL play nice with Compiz. Don't worry about supporting OpenGL 2.1. Just make sure that 2D and 3D are acceptably fast. I think that will keep those unfortunate enough to have this hardware sufficiently happy until said hardware dies.

        Comment


        • #5
          sigh...

          I have one of these "old" chips, I know it's not gaming hardware, but that with my 1.6GHz Pentium M still work just fine - I can surf most websites, play videos, do most office functions, and even play current cool indy 2D games...

          So I'd be saddened to be abandoned to a frame-buffer solution and even bit-rotting in some not quite finished rewrite of the Intel driver (if we went that way, I wonder if the old i810 driver could be revived as at least I had TV-out)

          I've been reasonably happy with Intel 2.11 KMS on .34, so I assumed we were pretty close (3D even runs for quite a while under a clean profile before it dies). I guess others aren't so fortunate

          (I really don't know what the difference is between my main profile/homedir and my test account that allows me to play games for a few hours rather than 2 minutes... but it feeds my hope that we aren't that far off making my 855GME happy enough)

          Comment


          • #6
            Ancient hardware is already supported perfectly well on ancient distributions. Why spend this much huffing and puffing trying to make ancient hardware work on modern OSes?

            Fact is, system requirements are constantly going up. This is true of Mac OS X and Windows, but now Linux has to face this fact. You can't boot up Ubuntu 10.04.1 on the same hardware you could boot up Ubuntu 6.06 on. The kernel is constantly being worked on to gear performance towards computers with more than 2-way SMPness, and NUMA (i.e. Core i3/i5/i7/i9). These performance improvements are definitely helpful for people running newer systems, but they add bloat and slow down ancient systems. In reality, it's difficult to have it both ways, because supporting both the old code and the new in a single codebase is, as Intel is quickly finding out, difficult to impossible.

            So I propose that anyone still using an i8xx chipset ought to stick with something like Ubuntu 8.04 LTS, or CentOS 5.x. You will still get very good security updates for a few more years with either distro, and the graphics stack in these distros still uses concepts that the i8xx can cope with, like UMS, OpenGL 1.3, and the good ol' XAA.

            I think that trying to pull the current intel code back to the point where it can support the i8xx is going to hinder work on current- and future-gen chipsets, so this is pretty much the best way. OTOH, forward-porting an old graphics stack to the latest kernel and X server might not be impossible, but if it's done I think it should be in its own separate repository. And you'd need a corresponding mesa-legacy and libdrm-legacy for this, because you can't just forward port the DDX, you have to forward port any components it depended on.

            I saw this coming the moment I heard Chris was working on bringing back UMS. I knew they would throw it out before it was merged to master.

            I have to say, though, that the shadowfb with KMS can be very fast. I've used shadowfb on several cards, and it is marvellous at 2D. You don't need 2d hardware accel most of the time because, let's face it, the CPU is a good blitter for simple GTK and web browsing; this is exactly what was used in Windows and Mac for many many years before beefy video cards were the norm. But even the oldest systems having i8xx would have at least Pentium 4 era CPU, which is plenty fast with standard desktop apps.

            And be reasonable, people. If you have such ancient hardware, don't expect to run Google Earth and Unigine and Heroes of Newerth. If you do expect it, you are being unreasonable. Buy new hardware if you want a real 3d experience, or use an old distro if you want to watch a 3d slideshow running Mesa 6.x on your old hardware. Sometimes you just have to upgrade, too bad.

            Comment


            • #7
              Originally posted by allquixotic View Post
              Ancient hardware is already supported perfectly well on ancient distributions. Why spend this much huffing and puffing trying to make ancient hardware work on modern OSes?
              Well partly is a feature by design of Linux distributions... which is that to get an upgraded web browser, office suite, or even fix a driver you are expected to replace your entire stack. Yes you can get PPA's for some things, but running an older distribution (as a user) doesn't feel built in by design - if a user doesn't want to get into compiling code to make their new Wacom tablet work they will need to upgrade their distro.

              Fact is, system requirements are constantly going up.
              In part, this is less of a fact than it used to be, it in part is what enabled netbooks to succeed... I can reap many of the benefits of current versions of Linux and Windows 7 on this same hardware. Sure, I don't get Aero [due to 8xx hardware limitations] but I certainly get a more secure system and regardless I could have still left it on Windows XP and there is still hardware and software written that will run on it. It's not that I can't do this with an older distribution of Linux, it's just one has to jump through more hoops.

              The kernel is constantly being worked on to gear performance towards computers with more than 2-way SMPness, and NUMA (i.e. Core i3/i5/i7/i9). These performance improvements are definitely helpful for people running newer systems, but they add bloat and slow down ancient systems.
              New kernels seem to run just fine on my system. The most significant risk is ensuring it gets tested to prevent regressions (and I suspect there are plenty of Pentium M's and other pieces of my chipset out there still that this isn't too high right now... things that don't work never did)

              In reality, it's difficult to have it both ways, because supporting both the old code and the new in a single codebase is, as Intel is quickly finding out, difficult to impossible.
              This isn't exactly the case here - Intel chose to rewrite their graphics stack a while ago so it's all 'new' code so the 'older' driver (with UMS) never was quite stable on any hardware and the abandoning of UMS by Intel was to eliminate as many bugs as possible in one swoop and allow them to focus their energies on the remaining bugs. So going back to buggy code doesn't make hardware more stable.

              If anything, the code originally coded for the old hardware was found in the i810 driver, which Intel abandoned as part of re-writing their driver. Since it wasn't updated with the new features in X/DRI2 (I believe) it was abandoned by distributions.


              So I propose that anyone still using an i8xx chipset ought to stick with something like Ubuntu 8.04 LTS, or CentOS 5.x. You will still get very good security updates for a few more years with either distro, and the graphics stack in these distros still uses concepts that the i8xx can cope with, like UMS, OpenGL 1.3, and the good ol' XAA.
              Except that I believe that was still the 'rewrite' driver and was not necessarily stable or quick at that point, plus around 8.04 the pulse-audio/flash issues were at their peak. I'm not inclined to re-install and fight through layers of fixes (not sure which 8.04.x fixed) to get it all going again. I wasn't until 10.04 that Pulse-Audio was running well enough that I could run the new Skype... there are significant limitations to saying 'run an old distro'

              OTOH, forward-porting an old graphics stack to the latest kernel and X server might not be impossible, but if it's done I think it should be in its own separate repository. And you'd need a corresponding mesa-legacy and libdrm-legacy for this, because you can't just forward port the DDX, you have to forward port any components it depended on.
              I don't know why the old driver wasn't carried forward... and I don't know what parts need to be carried forward versus the driver being updated to be compatible with newer pieces. If it's easy to install, and stuff runs, the end user won't care.

              I have to say, though, that the shadowfb with KMS can be very fast. I've used shadowfb on several cards, and it is marvellous at 2D. You don't need 2d hardware accel most of the time because, let's face it, the CPU is a good blitter for simple GTK and web browsing; this is exactly what was used in Windows and Mac for many many years before beefy video cards were the norm.
              There has been 2d acceleration for quite a while, perhaps not as nice as Windows7, but that's more relying on architectural changes to hardware to eliminate much duplicated memory/copying... and building 2d on modern 3d hardware. Even 3d has worked on this hardware for a while no?

              And be reasonable, people. If you have such ancient hardware, don't expect to run Google Earth and Unigine and Heroes of Newerth. If you do expect it, you are being unreasonable.
              Anyone who bought Intel 3d hardware was never expecting this... and this has been said so many times this is getting tiring.

              Comment


              • #8
                To be honest, the root of it (that is causing so much wind ;-) ) is the rolling model of Linux, the changing APIs, the incomplete software, is in need of evolution/refinement.

                I'd be happy running my old distribution to use old hardware (if there ever was this mythical stable driver for Intel graphics) if I could buy new hardware, and easily install a driver to work on it (ie, not compiling and upgrading numerous packages). I'd stick to the same distribution if I wasn't being forced to upgrade to get a stable sound system, or working Web Browser, or recent copy of Flash. PPAs help but they don't go far enough and aren't standard enough.

                I suspect part of it is just some refinements to the LTS model (can 10.04.1 give me updated end-user software on my stable core system?) but part of it is just an extension to the same issues video game makers are complaining about -- the lack of a stable/predictable target that vendors (even hardware vendors) can target.

                Comment


                • #9
                  Well partly is a feature by design of Linux distributions... which is that to get an upgraded web browser, office suite, or even fix a driver you are expected to replace your entire stack. Yes you can get PPA's for some things, but running an older distribution (as a user) doesn't feel built in by design - if a user doesn't want to get into compiling code to make their new Wacom tablet work they will need to upgrade their distro.

                  In this case, no your wrong.

                  The Pre-GMA 8xx video devies sucked ass. I mean they were really really really really really bad. They never worked right and never were that stable in ANY operating system. They sucked in Windows, they sucked in Linux, they sucked in everything.

                  OS X will never support them _at_all_. Windows Vista and Windows 7 will never support them properly and never did. Only XP works marginally better then Linux and even then it still is wretched.


                  The fact that the driver authors are bothering to go back and try to update the support for these old and crappy devices _at_all_ to be compatible with the sweeping modernization efforts that have happened to Linux and X Winodws in the past decade goes to show that they actually do care about their users and backwards compatibility with hardware.

                  I wouldn't. I owned 8xx devices and even worked with them professionally in embedded devices.... and I would not even give that much of a shit about it.

                  If they actually get 8xx devices working with xrandr and all that, even with out hardware acceleration, this is still better then 8xx users ever had.


                  OMG
                  Next time I will buy an ati powered laptop instead...
                  Your a idiot. Try going back to using pre-R200 ATI laptop devices in Linux (or any OS) and then maybe you'll have a clue why this is such a absurd statement.

                  Comment


                  • #10
                    Originally posted by drag View Post
                    Your a idiot. Try going back to using pre-R200 ATI laptop devices in Linux (or any OS) and then maybe you'll have a clue why this is such a absurd statement.
                    It's not because of i8xx, they just did too much wrong decisions concerning the development of the intel driver (like dropping ums too early).

                    With the latest bits I still cannot use my gma45 properly.
                    ## VGA ##
                    AMD: X1950XTX, HD3870, HD5870
                    Intel: GMA45, HD3000 (Core i5 2500K)

                    Comment

                    Working...
                    X