Announcement

Collapse
No announcement yet.

Intel i915 Gallium3D Performance Examined

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

  • Intel i915 Gallium3D Performance Examined

    Phoronix: Intel i915 Gallium3D Performance Examined

    Last week after a modern Intel Gallium3D driver was proposed for mainline Mesa, a side discussion ended up being ignited about making the i915 Gallium3D driver the default for older generations of Intel graphics hardware. To see where the i915 Gallium3D driver is at compared to the i915 Classic Mesa DRI driver, here are some new benchmarks from aging Intel i945 hardware.

    http://www.phoronix.com/vr.php?view=18658

  • #2
    Its preformance is bad.... really bad. Has code been directly taken from the classic driver or was this a clean rewrite?

    Comment


    • #3
      To be fair, the classic driver has probably had the crap optimized out of it. The performance of the new gallium driver is very respectable considering the limited amount of work that has gone into it and I'm sure if it is made the default that equal performance will be pulled out of it as the classic driver in time.

      Comment


      • #4
        Originally posted by Ericg View Post
        To be fair, the classic driver has probably had the crap optimized out of it. The performance of the new gallium driver is very respectable considering the limited amount of work that has gone into it and I'm sure if it is made the default that equal performance will be pulled out of it as the classic driver in time.
        Of course, that would actually require that i915g driver makes it as the default driver for older Intel hardware, otherwise there is really little incentive to keep working on it. Userbase can be a very powerful motivational tool.

        At least it appears to be usable for running an OpenGL-accelerated, composited desktop which all that flashy effects like wobbly windows fluidly without slowdown or lag. I consider that to be the minimum level of performance acceptable for any driver.

        Comment


        • #5
          Where's the "examined" part of the article?

          All I see is benchmarks with similars results to what's already been shown before. I was hoping for an explanation on why performance isn't on par...

          Comment


          • #6
            "...is changing the default open-source Intel Linux graphics driver for old i915 through i945 IGPs (common to the first-generation Intel Atom netbooks)".

            Isn't it common to all atom netbooks except those with PowerVR? e.g. even those with integrated igp on-package, it's still i945 IGP, right?

            It's a pity that performance is so bad and even worse with the new driver. In the end, these benchmarks don't really show if normal desktop performance is better or worse, and that is the most important thing for these igps.

            Comment


            • #7
              Hi Michael

              (don't you just love glossy screens - the first pic on the left)

              Comment


              • #8
                I wonder if worse performance can be related to the more features (or even higher OpenGL version?) exposed. Game could use them and performance would decrease then.

                Is there any easy way of disabling some feature to make OpenGL support the same between that 2 drivers?

                Comment


                • #9
                  Someone from intel (ian or eric, i think?) posted a couple patches that brought the classic driver up to GL2.1 and a similar feature set to the gallium driver.

                  It's really too bad that all the tests here were q3 based engines. I don't think that really tells us much of anything.

                  Comment


                  • #10
                    Strange... In my gentoo tests on laptop with exactly the same configuration (old Acer Extensa 5210 with gma950, 2 gigs of RAM and cpu upgraded to T5600) gallium driver is from 10 to 30% faster, than classic.

                    Comment


                    • #11
                      Originally posted by Zajec View Post
                      I wonder if worse performance can be related to the more features (or even higher OpenGL version?) exposed. Game could use them and performance would decrease then.

                      Is there any easy way of disabling some feature to make OpenGL support the same between that 2 drivers?
                      http://www.mesa3d.org/envvars.html
                      MESA_GL_VERSION_OVERRIDE for OpenGL Version
                      MESA_EXTENSION_MAX_YEAR to limit advertised extensions by year they came out
                      MESA_EXTENSION_OVERRIDE can be used for adding or removing advertised extensions e.g. MESA_EXTENSION_OVERRIDE=-GL_ARB_get_program_binary glxinfo

                      Comment


                      • #12
                        Originally posted by Sonadow View Post
                        At least it appears to be usable for running an OpenGL-accelerated, composited desktop which all that flashy effects like wobbly windows fluidly without slowdown or lag. I consider that to be the minimum level of performance acceptable for any driver.
                        By that definition, gma500_gfx (Poulsbo) is unacceptable... which I suppose is a fair point. Although it's quite usable if you choose not to use composited desktops
                        Last edited by GreatEmerald; 04-22-2013, 05:48 PM.

                        Comment


                        • #13
                          Originally posted by GreatEmerald View Post
                          By that definition, gma500_gfx (Poulsbo) is unacceptable... which I suppose is a fair point. Although it's quite usable if you choose not to use composited desktops
                          That's because it really is unacceptable?

                          If i ever find myself with any machine which uses PowerVR graphics I can assure you Linux is not going to go on it, period. I rather slap Windows on it (I have a Microsoft subscription which I pay out of my own pocket so my licenses are free) and enjoy a better user experience.

                          In fact, my personal rule with netbooks is that Linux does not belong there unless there is a very good reason to do so, mainly because of battery life and heat. I know many people claim that they experience better battery uptime with Linux as opposed to Windows, but that is not what I am seeing in my personal machines.

                          For notebooks Im a-ok with Linux because they are not meant to last the entire day on a single charge of juice and I generally lug my charger around when I have to bring it out with me for the day.

                          Comment


                          • #14
                            this is so fucking stupid....


                            why does this guy keep running the exact same tests on some gay old quake3 games that no one plays?


                            you do realize that just running gtkperf would be a million times better for this set and give us much more info right?

                            Comment


                            • #15
                              Originally posted by Pallidus View Post
                              this is so fucking stupid....


                              why does this guy keep running the exact same tests on some gay old quake3 games that no one plays?


                              you do realize that just running gtkperf would be a million times better for this set and give us much more info right?

                              That wouldn't prmote his testing suite, which he sells to game developers.

                              Comment

                              Working...
                              X