Announcement

Collapse
No announcement yet.

Gallium3D's LLVMpipe Driver Is Now Much Faster

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

  • #11
    That is truly impressive. It's too bad it didn't come sooner, since the cases where we've needed this are quickly dissolving. Still, it's awesome since it means we can largely dispense with 2D-only interfaces. I think 3D accelerated hardware has become increasingly common and cheap due to Android's prevalence in recent years. So if you have a computer that needs LLVMpipe as a driver, you probably aren't planning to do anything all that intensive to begin with.

    In essence, the new crappy computer is the SoC, so you might as well replace your old Pentium with 256 MB of RAM with a Raspberry Pi. With Wayland & Mir allowing you to run multitudes more options on that hardware, the whole argument for using Linux on old PCs will seem much less important. Until then, however, let's see if we can get KDE running on a Raspi with this new driver.

    Comment


    • #12
      I'm guessing it was the "allow software drivers to do direct copies instead of emulating blits" commit. Just wondering why it wasn't noticed earlier.

      Comment


      • #13
        Originally posted by Honton View Post
        Better yet. GNOME. The better software drivers are more important for GNOME since they provide a uniform user interface. These performance gains pretty much obliterates all the "software drivers are useless" hearsay. Great news.
        As a GNOME user and community member, I agree. I just wonder if the crazy stuff like blurs will ever be possible on this, since these games seem to be running well enough to watch. Also, it is intriguing to consider the concept of CPU-aided rendering. Like, you have a decent GPU, but your CPU is insanely overpowered, so you might as well use some of that processing power to push the graphics harder when your GPU is reaching its limit. Kind of like how we've been using GPUs for computational tasks, lately.

        Comment


        • #14
          I was amazed to see how much of a gain in performance this had, but then I noticed the frame rates were still pretty low.


          It does get me to wonder though - perhaps the frame rate would double if the refresh rate were set to 30Hz. I get the impression that LLVM isn't limited so much by complex instructions but rather amount of instructions. While dropping the refresh rate can have little to no impact on a standard GPU, it might make all the difference for a CPU.

          I also wonder if the AMD FX series would run better, since they use shorter pipelines and higher frequencies.

          Comment


          • #15
            It'd be nice if Michael would add Minecraft to the benchmarks. It'd be cool to see FPS above 50 on some hardware somewhere.

            Comment


            • #16
              Originally posted by scionicspectre View Post
              As a GNOME user and community member, I agree. I just wonder if the crazy stuff like blurs will ever be possible on this, since these games seem to be running well enough to watch. Also, it is intriguing to consider the concept of CPU-aided rendering. Like, you have a decent GPU, but your CPU is insanely overpowered, so you might as well use some of that processing power to push the graphics harder when your GPU is reaching its limit. Kind of like how we've been using GPUs for computational tasks, lately.
              The problem is...how do you synchronize that rendering? Also CPU's are good at integer math, GPU's are better at floating point so there's an unequal balancing act going on there. Its an interesting idea, and i'd be curious about an even theoretical implementation, but im not sure how good it would really be.

              Comment


              • #17
                Originally posted by curaga View Post
                I'm guessing it was the "allow software drivers to do direct copies instead of emulating blits" commit. Just wondering why it wasn't noticed earlier.
                I don't think that should be it.

                AFAIK, that feature was enabled in Gallium only recently, and the commit to disable it on software drivers happened almost immediately after. So it shouldn't have changed llvmpipe at all.

                Comment


                • #18
                  This one? "llvmpipe: implement blit"

                  If you mean that one, it was added in 9.1-rc1. So the question is then, if this caused a regression for 9.1 and if 9.0 had faster llvmpipe speeds. Michael, 9.0 please

                  Comment


                  • #19
                    Hmm, yea, if it was a regression, then the driver was just broken before, and this technically isn't a big speed boost after all.

                    Oh, and hey, I just noticed that there's a Razer DeathAdder in the article picture. I use one of those as well

                    Comment


                    • #20
                      Where can we expect to see next Mesa release? It is possible to see it for Fedora 19 in July?

                      Comment

                      Working...
                      X