Announcement

Collapse
No announcement yet.

Intel Bay Trail Graphics On Linux Are Slower Than Windows

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

  • #11
    Originally posted by bridgman View Post
    Cool...

    "Linux users demand closed source driver"

    (runs away)
    Well that was just me, of course.

    Though it does beg the question as to why open drivers can't perform as well if not better than the closed ones. But I assume that's because there is a lot of benchmark "skirting" and other shortcuts going on in the closed drivers.

    Comment


    • #12
      Originally posted by johnc View Post
      Though it does beg the question as to why open drivers can't perform as well if not better than the closed ones. But I assume that's because there is a lot of benchmark "skirting" and other shortcuts going on in the closed drivers.
      I don't think that kind of thing makes as much difference as the sheer # of developers working on the code. Closed source code is not inherently faster than open source code, it's just that if you're leveraging "free code" from work done on other OSes that code also comes with all the *restrictions* from those OSes, which practically means the resulting drivers have to stay closed source.
      Test signature

      Comment


      • #13
        Originally posted by bridgman View Post
        I don't think that kind of thing makes as much difference as the sheer # of developers working on the code. Closed source code is not inherently faster than open source code, it's just that if you're leveraging "free code" from work done on other OSes that code also comes with all the *restrictions* from those OSes, which practically means the resulting drivers have to stay closed source.
        But is Intel really taking their Windows drivers and shoe-horning them onto Android?

        I guess it's possible.

        Comment


        • #14
          Please compare to the Jetson TK1

          I don't have a full data set, but in Xonotic at least, my Jetson TK1 puts out nearly double the performance compared to the windows driver on the higher settings:

          level tk1 Intel L Intel W
          Low 83 49 58
          Med 75 ... ...
          Normal 71 ... ...
          High 42 19 23
          Ultra 29 13 15
          Ultimate 19 8 10
          When NV upgrades the Jetson TK1 driver from beta so that it is more stable (it crashes a few times a day ... Appears to be a memory leak), it will be a great daily driver.

          I read a blog post maybe a year back explaining that mobile processors will take many years to come close to desktop CPUs. I guess the author was wrong.

          In any event, Michael if you could post a guide on using pts on arm, I (and probably a few others) would love to help lighten your load and do a fuller comparison to the tk1.

          And dude, working 18 hours a day is way too much. You gotta figure some way to lightning your load! Strive for 6 - its much healthier!
          Last edited by deppman; 02 June 2014, 12:33 PM.

          Comment


          • #15
            Originally posted by deppman View Post
            In any event, Michael if you could post a guide on using pts on arm, I (and probably a few others) would love to help lighten your load and do a fuller comparison to the tk1.
            I'd be interested in such a thing as I intend to get the Doom 3 benches running on the TK1, if they aren't already...

            ...and if nvidia ever ships my kit...

            Comment


            • #16
              Originally posted by johnc View Post
              Well that was just me, of course.

              Though it does beg the question as to why open drivers can't perform as well if not better than the closed ones. But I assume that's because there is a lot of benchmark "skirting" and other shortcuts going on in the closed drivers.
              Considering the open source drivers, and gallium/mesa for that matter, don't support opengl 4.x, it shouldn't be much surprise that there are performance optimizations to be had too. I'm not sure about intel, but at least for the radeon drivers vs catalyst, one of the reasons the open source drivers perform worse is because catalyst has built-in micro-optimizations that are application specific. The open source drivers are, to my knowledge, designed to be generic. So suppose every single feature for the open source drivers matched with everything the card is capable of, the drivers MAY still perform worse. This is one of the reasons closed source drivers, including from nvidia, are so bloated.

              Comment


              • #17
                Originally posted by deppman View Post
                I don't have a full data set, but in Xonotic at least, my Jetson TK1 puts out nearly double the performance compared to the windows driver on the higher settings:

                level tk1 Intel L Intel W
                Low 83 49 58
                Med 75 ... ...
                Normal 71 ... ...
                High 42 19 23
                Ultra 29 13 15
                Ultimate 19 8 10
                Those are inproper comparation i think, you have arm 32bit procesor but compare results with 64bit x86, so proper comparation for that tegra k1 32bit will be to compare with 32bit x86 of course. Yep, those results from intel should be even lower in that case .

                Comment


                • #18
                  Originally posted by johnc View Post
                  Thanks for the information as that explains the discrepancy.

                  Then my next question becomes... why won't they provide a high-performing proprietary driver on Linux like AMD and nvidia do? Why should we accept less than optimal performance from the hardware we purchased?
                  This Intel driver does not support the Xorg driver model. It is based entirely on the SurfaceFlinger/StageFright driver model. You can use it simply by copying the relevant kernel binary, .so and .xml files from the Android device. Packing the kernel binary image is difficult to do right and getting GRUB2 UEFI to load it is even more difficult, but definitely possible. I have used it with Android IA's 4.2.2 builds for getting working H264 accelerated decoding on my Windows tablet running Android. I didn't bother to test anything much beyond that yet.

                  You might be able to get Mir running since that code is supposed to work with Android's driver model, but I know nothing about Mir beyond that, and haven't ever tried it.

                  Comment


                  • #19
                    Originally posted by deppman View Post
                    I don't have a full data set, but in Xonotic at least, my Jetson TK1 puts out nearly double the performance compared to the windows driver on the higher settings:

                    level tk1 Intel L Intel W
                    Low 83 49 58
                    Med 75 ... ...
                    Normal 71 ... ...
                    High 42 19 23
                    Ultra 29 13 15
                    Ultimate 19 8 10
                    What is the test bench for the Intel results exactly? Which CPU package, board and RAM, and at what power level?

                    The NVidia board pulls nearly 11W, far far too high for a tablet or phone: http://developer.download.nvidia.com...ef_May2014.pdf

                    Comment


                    • #20
                      I appreciate the comparison, but...

                      Originally posted by dungeon View Post
                      Those are inproper comparation i think, you have arm 32bit procesor but compare results with 64bit x86, so proper comparation for that tegra k1 32bit will be to compare with 32bit x86 of course. Yep, those results from intel should be even lower in that case .
                      I appreciate the comparison. The point is the TK1 looks likely to smoke the NUK even when used at its full potential. Now when the Denver cores appear, that should be real interesting

                      Comment

                      Working...
                      X