test scene for Superfly speed.

  • @johndoe641 said in test scene for Superfly speed.:

    I replaced my old 620watt with an 850watt PS today, so I decided to try my mega old 560Ti 448 Core and see what happens. Well it's only got 1.5gb of ram so it wouldn't render the Poser test scene, but doing the Blender test scenes... holy crap this old card literally almost cuts my render times in half.

    BMW Scene:
    (test scene spreadsheet result) GTX 980: 03:30.49

    My single GTX 970: 4:07.19

    GTX 970+560ti: 2:36.40

    Classroom Scene:
    (test scene spreadsheet result) GTX 980: 08:53.30

    (test scene spreadsheet result) GTX 1080: 7:57.13

    GTX 970+560ti: 6:32.08

    It's such a shame that my ancient 560ti only has 1.5GB of ram, it would make such an improvement on rendering speed in SuperFly I assume, since it did so much in Blender. If a 560ti could do that much, I wonder what a 1050ti or 1060 6GB could do. Even if they're not the most powerful cards, any help is good help I would guess, right?

    Hi there

    Not sure, but 1.5GB should be enough for some scenes which are not very populated or where you have one figure or two fully clothed and with hair and few props

    Usually V4 will take around 350-550MB, its all depends on if you have enabled subdivision,how many maps do have V4 and you can try to lower VRAM usage by lowering resolution of maps from 4k to 2k etc

    Any GPU will make difference if you pair two GPU, its doesn't matter what GPU just needs to be Nvidia

    Newer GPU offers good performance increase in Poser SuperFly, if you are thinking putting extra GPU to yours then I would go with GTX1070 if budget allows, GTX1060 6GB offers similar performance like GTX970 in games, in rendering not sure I don't have concrete data to say will perform better or worse in rendering, I'm thinking getting one extra GPU like is GTX1060 6GB for rendering, but I still thinking getting one more GTX1080

    In yours case depends on budget, if budget allows then GTX1070 I would choose amd if budget is smaller then GTX1060 6GB should be great option too, but still GTX780 6GB offers great performance, in many cases is faster than my Titan X

    Hope this helps

    Thanks, Jura

  • I would stay away from the GTX1050ti for rendering. It is basically a low wattage mobile GPU that found its way onto a desktop card.
    There is no additional power connector, and because of that it runs at a lower clock frequency as well. You have to sacrifice something to get the wattage that low.

    It has only 48 TMUs as well, which is very low for rendering usage.
    128 bit memory bus, again, not real wide for a GPU
    768 cuda cores, etc

    When you consider a GPU, TFLOPs can tell you a lot. TFLOPS is literally what you would get out of it full bore, if you could keep it full bore.

    GTX 10 series Tflop potential at base clock speeds (approx)

    • 1050: 1.8
    • 1050ti: 2.1
    • 1060: 3 gig card 3.9, 6 gig 4.3
    • 1070: 6.5
    • 1080: 9.0

    See a pattern here? Yep, the base cards are gimped in more ways that one.
    The 1050 and 1060 GPUs, are also a smaller nm, which is going to limit overclocking. All 10 series GPUs self overclock as well. They do not work the same way the older GPUs did in that department. Some companies push the factory over clock so far, you wont get much more out of it either. You can't do that on the 1050 and 1060 very well at all.

    My factory overclocked 1070 will plug along just fine at 2 gig. It costs just over twice what 1050ti does and will run circle's around it. Do the math on the Tflop potential of a 1070 running at 2 ghtz, it is a very fast card.

  • @shvrdavid In the case of AMD FirePro GPU's the professional line has drivers optimized for several dozen applications in addition to standard windows WHQL certified. The drivers are hard coded to enhance the user experience and performance of these independent applications. That driver development and the standards certification testing and QA/QC is where much of the additional cost goes. I suppose the same goes with Nvidia Quadro. They don't intend to sell many pro cards as distribution usually is limited to workstation OEMs and integrators rather than retail outlets. They are created to develop a market for the consumer versions while ulta/extreme clocked versions satisfy the top of thier markets and deliver maximum profit. GPU value is deceiving because these hyped and overclocked boards cost (almost) no more to produce than standard versions yet they command much higher price.

  • Ok folks

    new computer is built. AMD Ryzen5 1600x air cooled no overclocking, 16GB ram, GTX970 and GTX980

    Test scene with 512 sized tiles: 921.31 seconds
    Test scene with 256 sized tiles: 979.85 seconds

    When running this test the 512 bucket size made my fans come on pretty strong but the 256 seemed like it wasn't that big a deal for the cards and only a little slower.

  • @ghostship said in test scene for Superfly speed.:

    Ok folks

    new computer is built. AMD Ryzen5 1600x air cooled no overclocking, 16GB ram, GTX970 and GTX980

    Test scene with 512 sized tiles: 921.31 seconds
    Test scene with 256 sized tiles: 979.85 seconds

    When running this test the 512 bucket size made my fans come on pretty strong but the 256 seemed like it wasn't that big a deal for the cards and only a little slower.


    Did you rendered only with GPUs or with CPU as well?

    512 bucket size with GPUs is probably best what I've tried, 256 bucket size I use only for test renders to check lighting and how shaders looks etc

    What is killing render times is transmapped hair or any hair where is trans map used, EZSkin can slow down render times as well,don't have right now too much time on my hands but will do few tests later on when times allow

    How long yours previous renders took with single or dual GPUs?

    I didn't tried that scene on my cards and not sure how yours renders compare but will do if time allows later on

    Hope this helps

    Thanks, Jura

  • @jura11 Render times are about half when running both cards. My original test on my old i5 2400 (CPU render) took 3 hours and I was only about halfway done when I stopped it. So 15 min is pretty good compared with 6+hours.

    the difference between running 256 pixel times and 512 is only 6% decrease in render time so not really a big deal to run with smaller tiles and keep my machine running cooler and happy.

  • Dell R610 Server:

    2 x Xeon X5670 Processors (24 total cores)
    64 GB of RAM (32 GB for each processor)
    2 x 250 GB SATA SSD in a Raid 1 Mirror set
    Windows 10 Pro
    Poser and all files are located on the SSD mirror set

    I had to turn the bucket size down as the processors were only being used at 30% with a bucket size of 512. I also turned off progressive rendering for similar reasons - processors were only be used at 80%.

    Render time at 64 bucket size was 49.517 minutes.

    I actually expected better than this. I am a little disappointed.

  • @mr_phoenyxx For CPU renders I would keep my bucket size to about 32 pixels. Yes, turn off progressive rendering in CPU mode, you will get better performance with it off.

  • @mr_phoenyxx sorry, was confused there a bit. 64 pixels for bucket size, not 32.

  • @mr_phoenyxx This is why I should read a post carefully before spouting off.LOL You HAD a bucket size of 64 already.

    The GPU rendering blows the doors off of CPU rendering. This is why I made sure my new computer had 2 GPU card slots. You can use two GPU's even if they don't match. I have a GTX 970 and a GTX 980 and they both work together.

  • @mr_phoenyxx Ok, running CPU test right now. will have results soon.

  • On CPU: AMD Ryzen 5 1600x (6 core/12 thread), 16GB ram: 58 min. bucket size 64 pixels.

  • @mr_phoenyxx

    CPU render on my Xeon X5680 system:

    MacPro 5.1

    2x Xeon X5680
    48 GB RAM (triple channel configuration)
    OS X 10.11

    Bucket size: 64K

    Render time: 37.98 minutes

  • @nagra_00_ is that MacPro one that is in the older form factor like the old G5's or in the newer "trash can," shape?

  • @ghostship It has the old G5 form factor. AFAIK at that time nobody put CPUs into a trash can ;)

  • @ghostship No worries about not reading everything in detail, I knew where you were headed. :)

    I agree that GPU certainly produces far more processing power for the $$ than CPU does, but I thought I'd read that certain features of SuperFly don't work with GPU rendering. Not just BPT, but I thought caustics or something didn't work with GPU rendering?

    I'm also playing around with different settings to see what happens, and which is more efficient. I agree that 64 seemed to be the best option, which is why I tried it first.

    I just finished one with a bucket of 128 and it took 53.43 minutes. So slightly slower even though it appeared to be going faster.

    A test at 256 bucket is running right now. I expect it to be slower, as it's only using about 85% of the processors.

  • Poser Ambassadors

    Just in case people want to know - test scene rendered in 9 min 4 sec on a dual 1080ti
    On my previous set (2 x original titan) it took 17 min.
    Not as much improvement as i expected
    Octane is significantly faster with the new cards, but they upgraded CUDA to version 8 which has better support for the new cards.

  • @wimvdb Very interesting!

  • @mr_phoenyxx you might be able to squeez some more speed out of it by turning on BPT (set all branches to 1) and then lower the number of samples (start by lowering it to 20) I think I might try this as well.

  • @ghostship If we are trying to get a good idea of comparative speeds though, then don't we all need to be rendering a similar number of samples?