Announcement

Collapse
No announcement yet.

Twinmotion GPU / system ranking or benchmarks

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

    Twinmotion GPU / system ranking or benchmarks

    Hi all

    First just a quick thanks to unreal for making Twinmotion free, I was just in the process of putting a business case together to buy twinmotion licences so now I don't have to do that, yay!

    I was wondering if there would be much interest in setting up a 'benchmark' tread, or feel free to point in the direction of one that already exists.

    Simply that I couldn't find much information even owned by abvent the benchmark data was limited. There is information at https://epicsupport.force.com/twinmo...m-Requirements which basically states a quadro or gtx 1080ti / 2080ti are best. Which is what you would expect but we all know things are not andas clear cut and spending more doesn't always get you more in a linear fashion. So what is best raw performance is one question but what is best bang for the buck or if I have X card is it worth upgrading to Y card. Is it worth having two cards as there is a very vague statement about that.

    The above page makes note that 6,000 points is required and links to https://www.videocardbenchmark.net/directCompute.html

    Which shows the 1080ti with 9,353 points in passmark however the best gpu is surprisingly inside a CPU it is the Intel iris plus 655.... Now either there is an anomaly in the data or the best gpu to render on is an igpu? Which means an Intel NUC with a i7-8559U for £550 performs better than a £1,000 rtx 2080ti.

    Which is why I would be interested in actual benchmarks. We just need a test scene and a defined render output that could be shared as a twinmotion file so as a community we can run it and compare results?

    Is anyone else interested in this idea or just me?

    #2
    I have gone away and run the same render 1 minute full HD 60fps render on 3 computers to give an impression of what is to be expected. If there is interest I can probably upload the twinmotion files so others can run it to add more scores.

    Pc 1

    Spec
    • Intel Core-i7 3930K 6 core, 12 thread, Water cooled,
    • 32Gb 1600mhz DDR3 ram,
    • 250Gb SSD with twinmotion loaded on it and writing the render to a separate 250gb SSD.
    • Nvidia 1070ti 8Gb, water cooled.
    • Win 10 home
    Result
    • GPU clocks, Core 2012mhz, Ram 8202mhz,
    • GPU core usage 65-86%,
    • GPU memory usage 35%
    • CPU clocks 3.375Ghz
    • CPU usage 8-14%
    • System memory usage 22%
    • Render finished in 5 min 59 seconds
    Pc 2

    Spec
    • Ryzen 7 1700 8 core, 16 thread 3GHz.
    • 16Gb 3200mhz DDR4 ram,
    • 240Gb m.2 SSD with twinmotion loaded on it and writing to it.
    • AMD sapphire nitro+ RX 570 4Gb.
    • Win 10 home
    Result
    • GPU clocks, Core 1300-1400mhz, Ram 7800mhz,
    • GPU core usage 83-86%,
    • GPU memory usage 74%
    • CPU clocks 2.8Ghz
    • CPU usage 5-8%
    • System memory usage 47%
    • Render finished in 15 min 9 seconds
    Pc 3 a 5K imac (just for fun)

    Spec
    • Intel Core-i7 6700K 4 core, 8 thread,
    • 32Gb 1866mhz DDR3 ram,
    • 1Tb fusion drive with twinmotion loaded on it and writingwriting to it.
    • AMD R9 M395X 4Gb,
    • OSX High Sierra
    Result
    • GPU clocks, Core 909mhz, Ram 5460mhz,
    • GPU core usage 98%,
    • GPU memory usage 90%
    • GPU temp 91 C !!!
    • CPU clocks 3.5-4.0Ghz
    • CPU usage 29-32%
    • System memory usage 39%
    • Render finished in 41 min
    As expected the 1070ti finished above the others with the mobile M395X in the imac limping in last. Passmark direct compute gives the 1070ti 8Gb a 7,381 score and I finished my test in 5 min 59 sec. The RX570 with 4gb gets 3,370 and finishes the same scene in 15 min 9 sec.

    The cheapest RX 570 5GB on amazon uk is £172,

    The cheapest 1070ti 8gb on overclockers uk is £428 (which is way overpriced, but the 2060 and 2070 according to pass mark perform worse)

    The 1070ti is 119% better than the RX570 on pass mark and completes 150% quicker based on % increase but you pay 148% more so really you get what you pay for lol.

    But the processor really doesn't make any difference, nore does the storage media as long as it is an SSD. Ram usage on 16gb is less than 50% so in a pinch 8gb may suffice but 16gb would be best, 32gb is overkill. System memory speed doesn't seem to have any great affect and as can be seen with my r7 1700 pc the processor even underclocked itself as the loading was so low.

    The test with the 1070ti was done with the game ready driver so I may test the creator ready driver to see what affect that has. Also I tested the 1070ti stock at 1683mhz core and 8000mhz mem and it took 2 sec longer. So an overclock can help as long as it's stable. For example my Rx 570 can OC to 8400mhz on the ram in gaming but would hard crash in twinmotion so I had to ease back.

    What I really want to try an Intel skylake i7 with the UHD630 as passmark does seems to be a good indicator of performance in twinmotion but I still can't believe a UHD600 equipped processor score 28,723.
    Last edited by ArchiDryburgh; 05-23-2019, 04:40 PM.

    Comment


      #3
      Hi,

      I find your proposal very interesting, especially now the old forum posts on the old Abvent website are not available anymore.

      In regards to your method of comparing GPU's based directly on their respective scores in the Passmark benchmark might not be so correct. Please correct me if I'm wrong. The GTX cards can't use RTX cores, obviously. RTX videocards would have a distinct advantage over GTX and AMD GPU's if/when Twinmotion supports direct raytracing.

      Check this benchmark where RTX is turned on/off on the and compare the gain/loss in framerates between GPU's.
      https://tweakers.net/reviews/7018/3/...enchmarks.html or https://benchmarks.ul.com/compare/be...OP&minRating=0
      Last edited by guapper; 09-13-2019, 01:25 PM. Reason: Added 2nd benchmark

      Comment

      Working...
      X