Announcement

Collapse
No announcement yet.

Linux Might Wipe Out The Notorious Intel Poulsbo/Moorestown 2D Acceleration

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

  • Linux Might Wipe Out The Notorious Intel Poulsbo/Moorestown 2D Acceleration

    Phoronix: Linux Might Wipe Out The Notorious Intel Poulsbo/Moorestown 2D Acceleration

    Longtime Linux users still likely cringe when hearing "Poulsbo" as Intel's first-generation Atom processors that featured "GMA 500" graphics that were based on Imagination Technologies PowerVR SGX IP. The Linux driver support was just awful and now as we prepare for 2021 the Intel Linux kernel driver might just drop its 2D acceleration support for Poulsbo and the short-lived Moorestown plaform...

    http://www.phoronix.com/scan.php?pag...-Drop-2D-Accel

  • #2
    Originally posted by phoronix View Post
    Hopefully no one is actively using Poulsbo anymore come 2021. In any case, deleting the 2D acceleration code does relieve the GMA500 kernel driver by some 300+ lines of code.
    We can fairly much bet someone somewhere is actively using it. That "GMA 500" 2d accelerated use to be only slightly faster than the CPU software rendering and then someone found how to optimise the CPU software rendering and since that point its always been slower. The horrible part is the "GMA 500" the per operation setup to use that 2d accelerated end up taking more cpu instructions that if you do well optimised cpu software rendering to perform the same task.

    This ticks all the boxes of horrible.
    1) GMA 500 2d acceleration is slower than CPU software rendering.
    2) GMA 500 2d acceleration uses more power than CPU software rendering it.(you want at least a power saving if it not faster)
    3) GMA 500 2d acceleration end up using more ram than CPU software rendering it.
    4) GMA 500 2d acceleration end up using more CPU instructions than software rendering it so CPU tied up for longer so adversely effecting general performance.
    5) GMA 500 documentation is with a vendor who is not cooperative so unable to find out if problems 1,2,3 and 4 are a case of us doing it wrong.

    I don't think you could get a list of more wrong for so called accelerator its more of a brake pad than accelerator in this case.

    So for the users still using Poulsbo deleting these 300+ lines is a benefit to them. Even if the Imagination Technologies does answer on how it should be done if that improves things that would be we are doing it wrong it that results in rewrite those 300+ lines anyhow with in the process them being deleted and replaced.

    Comment


    • #3
      What is current situation with Intel GMA 3600 (PowerVR SGX 545) by the way?

      Comment


      • #4
        Typo:

        Originally posted by phoronix View Post
        Intel Linux kernel driver might just drop its 2D acceleration support for Poulsbo and the short-lived Moorestown plaform.
        I actually feel that Poulsbo never worked on Linux after all...

        Comment


        • #5
          Glorious Dell Mini 9 in the photo, still works like a charm 👌

          Comment


          • #6
            Originally posted by RussianNeuroMancer View Post
            What is current situation with Intel GMA 3600 (PowerVR SGX 545) by the way?
            https://wiki.archlinux.org/index.php/Intel_GMA_3600
            Fairly much forgot any form of graphical acceleration you will be software rendering. Yes horrible software rendering on that cpu in that combination is more power effective and less cpu time consuming than using the PowerVR bit.

            Expect that you may need to add a kernel driver or run a custom kernel depending on the distribution with a extra bit of user-space installed that you get output at all.
            https://www.kernel.org/doc/html/latest/fb/uvesafb.html

            Not all distributions build there kernels with uvesafb enabled.

            Horrible this is the best Intel GMA 3600 stuff has ever worked with Linux.

            When it comes to running modernish applications the older Pineview atoms 45nm with working intel graphics runs rings around the newer Cedarview atoms 32nm (Intel GMA 3600).

            Yes those 2011-2012 Cedarview(Intel GMA 3600) laptops are fairly much junk for Linux usage. People will want to charge you more for them than a Pineview device but Cedarview atoms (Intel GMA 3600).going to perform worse if your plan it to use it with Linux.




            Comment


            • #7
              Originally posted by oiaohm View Post
              Even if the Imagination Technologies does answer on how it should be done if that improves things that would be we are doing it wrong it that results in rewrite those 300+ lines anyhow with in the process them being deleted and replaced.
              Iirc, the 2D/modesetting/DRM part is not from Imagination Technologies but from Intel. There were even suggestions to have the GMA500 support code as part of the i915 driver due to large amounts of code duplication, but the i915 maintainers refused: https://lkml.org/lkml/2011/2/22/279

              Originally posted by RussianNeuroMancer View Post
              What is current situation with Intel GMA 3600 (PowerVR SGX 545) by the way?
              The hardware support of the gma500 driver is

              GMA500 (Poulsbo)
              GMA600 (Moorestown)
              GMA3600 (Cedar Trail, experimental)

              Newer Intel Atoms with PowerVR graphics like Clover Trail are not supported at all.

              Comment


              • #8
                Originally posted by oiaohm View Post
                This ticks all the boxes of horrible.
                1) GMA 500 2d acceleration is slower than CPU software rendering.
                2) [...]
                Sadly this isn't that uncommon. Most Creative Labs audio hardware after ~2002 is likewise consistently slower than software rendering on both Windows and Linux. I switched to the open source OpenAL-Soft (thanks Chris!) for my projects and never looked back. Don't forget that Creative even destroyed Aureal through Intel-style illicit means, because they couldn't compete with the impeccable audio quality and performance of Aureal's A3D (which I still regard as unmatched, even today).

                Comment


                • #9
                  Those old Atoms where terribly slow back them, now you can barely open a modern webpage without a ad blocker.

                  I still have a old netbook from Asus, a EEEPC Seashell with a Atom N270 (DiamondVille?). The thing is very derelict, but I keep it for sentimental reasons, it was the first laptop I owned. Today the better use I have for it is moving it around to do maintenance in networks outside my workplace, because it is easier to carry than my bricky old Thinkpad.

                  Comment


                  • #10
                    There actually was a Linux driver for GMA 500 that even supported 3D and video playback hardware acceleration. But there is a big BUT, it was a proprietary one that run only few specific OS releases of the era (like MeGo - remember that thing?), and still was problematic to get going.
                    Fortunately there is an effort to make the propietary EMGD driver work on newer OSes, here it is: https://github.com/EMGD-Community/intel-binaries-linux This means there actually is little reason to keep that 2D driver in the kernel.
                    There were many really cool machines with that GPU, my favourite was my Nokia Booklet 3G, a passively cooled mini laptop by Nokia that even on Linux achieved more than 6 hours of battery life under heavy use. It was very well build with its aluminium chassis, awesome keyboard and touchpad. And it featured a 3G modem which made it so quick to use anywhere I carried it everywhere for good few days. I really wish it was well supported in Linux, but we have to let go of old and obscure hardware :/

                    Comment

                    Working...
                    X