Announcement

Collapse
No announcement yet.

GTX bottleneck in RTX mode and minimal performance gains in CUDA

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

  • GTX bottleneck in RTX mode and minimal performance gains in CUDA

    Just installed an EVGA RTX 2080 Super that is paired up with my MSI GTX 970, and tried to do some tests to see the performance. I immediately I noticed that the 970 severely bottlenecks the 2080 when in RTX engine mode. For example:

    RTX
    2080 + 970 = 1 min 12 sec
    2080 = 49 sec

    And for comparative purposes:

    CUDA
    2080 = 1 min 16 sec
    970 + 2080 = 1 min 10 sec
    CPU + 970 + 2080 = 1 min 5 sec

    Now I'm just going to assume that in RTX mode the 970 doesn't play nice for incompatibility reasons and it's a software issue. Which I can understand. But what I don't understand is how minimal the gains there are in CUDA mode when both cards are enabled. Is this to be expected on certain scenes? The 970 just old and outdated? Or something else going on. FYI I have all of the latest drivers and using 3ds max 2020 SP3 and V-Ray 4.30.

  • #2
    How fast is the single 970 in Cuda mode? So we could see what the 970 can add too a mixed hardware rendering.
    www.simulacrum.de ... visualization for designer and architects

    Comment


    • #3
      Originally posted by Micha View Post
      How fast is the single 970 in Cuda mode? So we could see what the 970 can add too a mixed hardware rendering.
      That was dumb of me for not adding that data point in there. So I redid the test on the same scene but different camera angles, resolution, larger region render, so the render times will be completely different, anyways the results were interesting.

      CUDA

      970 = 5:58 min
      2080S = 2:30 min
      970 + 2080S = 2:02 min
      970 + 2080S + 4930k@4.2ghz = 2:06 min

      RTX

      970 = 8:25 min
      2080S = 2:19 min
      970 + 2080S = 2:10min

      As you can see the gains are pretty minimal, even though with the specs of the 970 (even low as they are) you would think it at least show 15-30% improvement in the CUDA time at a minimum? I'm not so much concerned about the RTX performance simply because of the hardware incompatibilities as I mentioned before.

      I should probably also list my specs if that helps

      Intel 4930k o/c @ 4.2ghz
      ASUS Sabertooth X79
      Corsair Vengeance 32GB (4x8GB) DDR3 1600 MHz
      EVGA RTX 2080 Super
      MSI GTX 970
      3ds Max 2020 SP3
      V-Ray Next 4.30.00
      Latest drivers on everything FYI
      Last edited by S_K_I; 24-03-2020, 11:54 PM.

      Comment


      • #4
        Looks like the 970 has 40% of the power of the 2080S. So, if you use 2080S + 970 you should get the speed of the 2080S multiplied with 1/1.4. So it could be expected 1:47 min but you get 2:02 min only. .... ?
        www.simulacrum.de ... visualization for designer and architects

        Comment


        • #5
          2:02 is correct. I may try to do some other tests on different scenes to compare the results. But in theory, the render time should technically be lower, correct?

          Comment

          Working...
          X