Announcement

Collapse
No announcement yet.

How can that be?

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

  • How can that be?

    $ export vblank_mode=0
    $ glxgears

    mobility x1900 yields 1400 frames
    HD5850(profile high) only 1200

    distribution is debian testing, with kernel 3.8 (liquorix)

    Also, I noticed current r600 driver on 5850 is EXTREMELY inefficient when it comes to textures.

    For example, in urban terror 4.2, map Cascade, when player looks at fontain that is essentially moving (multilayered?) textures the fps goes down to 20.
    But if player rises head straight up, it goes to 80 fps.
    On simple maps without this, there is no such issue.

  • #2
    glxgears is mainly a cpu benchmark, and not a good one even there.

    Comment


    • #3
      Yep. The glxgears demo does so little drawing per frame that you're mostly measuring per-draw and per-frame driver overhead, so there's not much correlation between glxgears numbers and real world numbers these days.

      Driver optimizations can even make the correlation negative -- the same extra driver code which allows the chip to run more efficiently under heavy load slows things down under trivial load.
      Test signature

      Comment


      • #4
        glxgears is definitely not a benchmark. It is only useful as a way of showing that your gpu has direct rendering working. Thats it.

        Comment


        • #5
          Thanks for replies!

          So, apparently mobile x1900 on lastest kernel 3.8.4 can achieve 2300 fps in glxgears due to HyperZ!
          And easily runs 32 (thirty two) glxgears instances distributed over screen without fps drop!

          This way glxgears acts as awesome "furmark"

          So, glxgears is not a benchmark, its a burnout tool xD

          Comment


          • #6
            you could just run furmark. it runs just fine.

            Comment

            Working...
            X