Announcement

Collapse
No announcement yet.

Vray RT - 1070 - disappointing performance...

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

  • Vray RT - 1070 - disappointing performance...

    Thought i'd share this - as I struggled to gauge what the real world performance would be on the new 1070 / 1080 GPU's versus my 780TI cards.

    Currently I run a 780TI (MSI version with slightly higher clock speeds) - and I wanted faster Vray RT performance.

    I struggled to find out how much better the new cards would be than the 780TI due to CUDA cores / clock speeds etc - so in the end I took the plunge and bought a 1070 reference card.

    In a scene i'm working on - using Vray RT as the production renderer - the 780TI (set as the display card also) did 3.03 mins per frame.
    The 1070 (with no displays connected) did 2.47 per frame.

    So - not a significant leap by any means

    Anyone else any thoughts / experiences to add?
    I was going get a 1080TI - but i'm really not sure if the small extra performance is worth ?700.... for now i'm just using both cards to render my scene and getting a frame time of 1.47 using both - which seems about right

  • #2
    780ti has 2880 CUDA cores compared to the 1920 CUDA cores of 1070 .. Those 2880 are Kepler cores, which are not as efficient as Pascals, so 1070 should be faster, but I can't tell by how much. Proper comparison would be 780ti vs 1080ti, since they are in the same segment. Also, keep in mind that 1070 has 8GB memory compared to the 3GB memory of 780ti and that it uses 150W compared to the 780ti 250W.

    Btw, is that V-Ray 3.5? Is GPU set as production (and have you changed any default settings) and are you sure you are the GPU is well utilized during the reddening ?

    Best,
    Blago.
    Last edited by savage309; 25-03-2017, 09:58 AM.
    V-Ray fan.
    Looking busy around GPUs ...
    RTX ON

    Comment


    • #3
      Did a few other tests using the Demo's I have installed of FStorm and Octane.

      Octane showed a very slight speed increase using the 1070 over the 780TI - but on a 30 second render - we are only talking about a 1 second speed increase for the 1070.

      FStorm - on the other hand - with the same scene - rendered almost TWICE as fast with the 1070 versus to 780TI.... so with FStorm - I was seeing 14 seconds a frame with my little test scene.

      Also - opened up the same scene I was using Vray RT on and that appeared to render much faster - with less noise using FStorm, hmmmm.
      Last edited by ripe; 25-03-2017, 10:03 AM.

      Comment


      • #4
        Hi Blago - Yes Vray 3.5.

        Not sure how to know if GPU is well utilised - when I use RT - I see 80/90% usage.

        I have the 780TI in slot one on the PC and this runs the monitors too.
        I have the 1070 in slot two and have zero monitors connected.

        When using Vray RT as production - I am using lower settings and Denoiser - as I need to achieve sub 3mins per frame for the animation.

        Comment


        • #5
          Probably FStorm is not well optimized for Kepler GPUs, in contrast to Octane and V-Ray.

          Also, hard to say about the render speed without a scene. I did some tests a while ago on performance and things were looking rather well for V-Ray GPU. Just keep in mind the small (probably unbiased..) tricks like, "blurring the frame buffer and applying sharp after that, to hide the noise", "lens effects on by default to hide the noise", "clamping GI intensity to 1", "clamping GI depth to 1", etc . If you however think that there is a problem with the performance of any scene, just mail it to blagovest.taskov@chaosgorup.com and I will be happy to check.

          Best,
          Blago.
          Last edited by savage309; 25-03-2017, 10:06 AM.
          V-Ray fan.
          Looking busy around GPUs ...
          RTX ON

          Comment


          • #6
            That 80% utilization sounds rather bad. There is something wrong with that .. This is definitely not okay. Are you rendering on the both GPUs ?

            Also, keep in mind that during denoising the GPU is not rendering and that having the many additional render elements for the denoiser can have some affect on the performance.

            Best,
            Blago.
            V-Ray fan.
            Looking busy around GPUs ...
            RTX ON

            Comment


            • #7
              Hi Blago - appreciate the replies.

              Ignore the 80 / 90 - that was simply when I was using 1 GPU and it was running the screens too.

              I have just re-set Vray and rendered my scene using VrayRT as the production renderer and get an almost consistent 100% GPU usage in the stats - dropped to 99% for a second here and there - but generally 100%.

              So - I re-rendered my scene - checked I got 100% GPU usage - and here are the results.

              With the 780TI - my scene took 4.36
              With the 1070 - my scene took 4.22

              I would have thought that the 1070 should be quite a bit faster than the 780TI - but clearly it is not.
              In FStorm - the same scene is almost twice as fast with the 1070, versus the 780TI?

              Comment


              • #8
                1070 is low end GPU vs 780ti which was top of the line. This performance is somewhat expected. New CUDA compilers sometimes tend to have worse performance with older GPUs ... you have to ask FStorm if that is the case with them.

                Best,
                Blago.
                V-Ray fan.
                Looking busy around GPUs ...
                RTX ON

                Comment


                • #9
                  Hi Blago.

                  I really appreciate the replies - I bought the 1070 mainly to see what the performance would be like - to judge if a 1080TI would be worthwhile - I got a great deal on the 1070 - so it won't be wasted

                  From what I have seen - I think the cost associated with moving from the 780TI (which I have 2 of) - isn't worth the investment yet - certainly for Vray RT.

                  Also - I hoped other people - in a similar position to me would appreciate the info - currently for a significant improvement in speed I would need a couple of 1080's at a minimum - which is around 1000 pounds for not a huge improvement in speed.
                  Last edited by ripe; 25-03-2017, 10:29 AM.

                  Comment


                  • #10
                    Well if the investment is worth it depends from use case to use case, of course. Testing it by yourself is the best way to tell how well it works for you, so I guess you did the right thing
                    1080ti is about 65% faster than 1070.

                    Best,
                    Blago.
                    V-Ray fan.
                    Looking busy around GPUs ...
                    RTX ON

                    Comment


                    • #11
                      Yep, I have my eyes on 1080 ti's...I'm going to skip the 1080's. It helped me to read the user benchmarks to see the 1070's and 1080's were, what I thought, not as good as an investment for Vray RT for me as the 1080 ti's might be.

                      Might work well for gamers, but I'm looking at cuda cores and clock speed.

                      Comment


                      • #12
                        Yep 1080ti is what's needed to make the cost / performance worthwhile.

                        I have decided to spend a little on a new 1000W PSU and a new case so I can run 3 GPU's - my exisitng 2 x 780Ti cards and my new 1070 card.

                        Comment

                        Working...
                        X