No announcement yet.

TitaniumGL 3D drivers (linux version)

  • Filter
  • Time
  • Show
Clear All
new posts

  • #46
    Originally posted by Linuxhippy View Post
    This is realy interesting stuff - and extremly impressive considering it was written by a single developer.
    thankyou ^^


    • #47

      There is an update on the linux version:

      -64-bit version released.
      -fixed some stability issues
      -TitaniumGL software renderer now uses packet-based out-of-order rendering, wich allows to incrase the number of rendering threads
      -Software rendered version can now utilize up to 24 cores.
      -TitaniumGL now uses smaller memory stack
      -1.5x-2x speed-up in the software renderer when using 4 core computers
      -There is a modern and a legacy versions, compiled with different gcc-s. Legacy version is for old distributions.

      i hope this version will fix the crash problems on vesa driver too. at least it worked okay for me.
      get it while its hot:
      Last edited by Geri; 07-25-2013, 02:01 PM.


      • #48
        Impressive that you're still working on that.

        Just so people know what they can expect:

        Minetest: SIGFPE, Arithmetic exception.
        #0  0x00007ffff767375e in HaromszogRenderel_szuperthread () from /home/foo/Downloads/TitaniumGL/64bit/
        #1  0x00007ffff767b3eb in haromszogekki_szuperthread () from /home/foo/Downloads/TitaniumGL/64bit/
        #2  0x00007ffff767b724 in haromszogek0ki_szuperthread () from /home/foo/Downloads/TitaniumGL/64bit/
        #3  0x00007ffff675cdd2 in start_thread () from /usr/lib/
        #4  0x00007ffff5604cdd in clone () from /usr/lib/
        But with intel and MESA_GL_VERSION_OVERRIDE=1.4 minetest it runs.

        Performance is meh, but it's probably not far from what a software renderer can do, even when rendering rather inaccurate. Fonts are not accurate, sort of an interlacing effect. supertuxkart sort of renders mostly correctly ~1-2 fps, maniadrive renders pretty much incorrectly with heavy jumps between frames, ~1 fps.
        openarena on 640x480 has sort of playable fps. On a related note, openarena by default will start in fullscreen and disable your second monitor and exiting it will leave you with cloned screens instead of side by side like you had set up before. Guys, it's 2013. Get this right, please.

        By the way, when running minetest in gdb, there is a huge number of threads created and destroyed that scroll by. And by huge I mean huge. Maybe you can get more performance when reusing threads.


        • #49
          hi, thanks, i will check that game, becouse that crash should not happen on that place.

          reusing threads was alreday in plan, but for security reasons, not yet tryed how it perform.
          (also there is not so huge amout of threads executed, so it should not affect negatively the performance)

          also, what computer do you have? supertuxkart should be playable on a recent 4 core cpu (20 fps+)

          but it's probably not far from what a software renderer can do
          oh.. well... umm.. there is 10-40x faster rasterer algos exist than my algo

          in the last year somebody shown a raster algorithm that consumes only 4-10 clock per pixel (including texturing)

          if the really smart peoples would put they head together, to write a very fast software rasterizer, it would basically outperform on 4 cores every graphics card and igp's cheaper than 200 usd. however, when a very fast software renderer appears, nvidia and amd outbuys them immediately, and digs them very deeply (nobody would hear from them any more). for obvious reasons.
          Last edited by Geri; 07-26-2013, 09:29 AM.


          • #50
            TitaniumGL 2014, 01. released

            -Well a few days still left until 2014, whatever
            -Added mip-mapping support for the software renderer
            -5-25% generic performance gain
            -Fixed some ,,dotting'' bugs with the software renderer
            -Fixed some segmentation fault bugs (texture upload forgot to lock rendering...)
            -Fixed a memory-pool related crash on linux
            -32 bit binary now compiled again with i686 flag
            -Software renderer now supports up to 128 threads instead of 24