Announcement

Collapse
No announcement yet.

Intel Just Released A Crazy Fast Acceleration Architecture

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

  • #11
    Originally posted by rohcQaH View Post
    This is a patch to the xorg driver. It shouldn't have much impact on 3D rendering, which is done by mesa.
    Right, in this case, ut2004 on an Atom netbook, we are not swap bound and there should be no performance difference between the two ddx.

    Originally posted by rohcQaH View Post
    It would be interesting to know whether this new acceleration architecture could be used by ati or nouveau. If it can be used on older intel chips, it's not SB-specific. But are there other differences in GPU architecture that make it useless or incompatible for GPUs of other vendors?
    There are a lot of similarities with EXA, which given the heritage is not unsurprising, and there are a few changes in the generic layers (most importantly to the interfaces into the backend) that would be useful for any shaderful architecture. But most of the performance improvements come either from improving GPU management, which is Intel specific, or by improving the chipset specific code, and from pushing down the interface to the backends - trying to reduce the midlayer. Here incorporating the generic improvements into glamor or st/xorg would be a better path for improving EXA. But the most of the work is in writing good backends, hence why everybody favours first creating good GL drivers and then leveraging those from a generic ddx. [However, there are a few issues with that approach and some of the more limited hardware; X advertises far more than the hardware can deliver and applications commonly try to take advantage of that. Just imagine a company that manufactures a GPU that can drive a 4096x4096 framebuffer, but only render to 2048x2048 of it... The result is that the driver must do a complicated little dance to hide those deficiencies, keep the GPU busy and so avoid the expensive fallback to the CPU.]

    My real motivation for the design will hopefully become manifest when a couple more features land upstream...

    Comment


    • #12
      Originally posted by rohcQaH View Post
      - SNA is fundamentally incompatible with ati and nvidia GPUs
      - SNA is fundamentally incompatible with ati and nvidia GPUs, but some of the research and work on SNA could benefit other drivers
      Those sound the most likely given the kind of crap Intel usually does. "We're too special to help develop EXA, we're going to fork it and duplicate most of it to make it run a little faster with our chipsets then just expect X to support both"

      Edit: Isn't Intel (older chipsets anyway) also most of the reason why the old userspace mode setting code can't be removed? They can't ever seem to fix their drivers for older chipsets to support KMS, but they fork EXA, twice. Go figure.
      Last edited by DaemonFC; 06-04-2011, 12:40 PM.

      Comment


      • #13
        Beyond the performance improvements, what I'm really liking is the addition of a test suite by this commit.

        Comment


        • #14
          Originally posted by mattst88 View Post
          Beyond the performance improvements, what I'm really liking is the addition of a test suite by this commit.
          I forgot to comment on that. What's the point of benchmarking something this weak? My LibreOffice framerate is better than yours?

          Comment


          • #15
            Originally posted by DaemonFC View Post
            Those sound the most likely given the kind of crap Intel usually does. "We're too special to help develop EXA, we're going to fork it and duplicate most of it to make it run a little faster with our chipsets then just expect X to support both"

            Edit: Isn't Intel (older chipsets anyway) also most of the reason why the old userspace mode setting code can't be removed? They can't ever seem to fix their drivers for older chipsets to support KMS, but they fork EXA, twice. Go figure.
            Agreed totaly. This said is the main reason I would never buy any kind of Intel graphics hardware. At least in the near future.

            Comment


            • #16
              Originally posted by DaemonFC View Post
              I forgot to comment on that. What's the point of benchmarking something this weak? My LibreOffice framerate is better than yours?
              Huh? What do you mean?

              Comment


              • #17
                Originally posted by rohcQaH View Post
                This is a patch to the xorg driver. It shouldn't have much impact on 3D rendering, which is done by mesa.
                Er, wait. Why then are there World of Padman benchmarks, which also show a huge performance improvement? Am I misinterpreting something?

                Comment


                • #18
                  Very nice!

                  This supports my impression that Chris Wilson is a real win for the intel team.

                  Comment


                  • #19
                    Originally posted by pejakm View Post
                    Agreed totaly. This said is the main reason I would never buy any kind of Intel graphics hardware. At least in the near future.
                    That's a little too radical IMHO

                    I think the problem here (which you and parent poster obviously missed) is that the hardware is too different to fit within one API. It might work but is far from optimal. Making big API changes isn't very popular since it can make other peoples code non-functional (or introduce tons of legacy crap). Letting Intel have their own solution for now makes it possible for them to actually be able to try new things. SNA is obviously the way to go for Intel hardware, so let evolution have it's way.

                    I'm pretty sure that when things slow down (or hell freezes over) it will all be put in a stable API that fits everyone. In the mean time, we'll just have to live with the fact that things change.

                    Comment


                    • #20
                      Originally posted by patrik View Post
                      That's a little too radical IMHO
                      Huh? We're talking about Intel, a massive company with near unlimited resources and comparatively only a small handful of GPU designs to maintain can't handle the move to KMS, compared to AMD who where in a bad way financially till only just recently is handling the switch very well in their OSS drivers even though they have a much larger backlog of GPU designs to maintain.

                      Maybe it's just me, but something just doesn't add up.

                      Comment

                      Working...
                      X