test scene for Superfly speed.



  • @piersyf

    Ohh sorry for confusion,I'm right now rendering my older scene which is bigger than that above and this has been my output log(those memory)
    Yup,GTX970 is powering displays then always will take some memory from GPU,if you want to have full 4GB I would use onboard GPU or get some cheap GPU only for displays
    Poser log is only calculating actual memory which he assigned or using for rendering which doesn't include Poser scene etc and Windows etc and doesn't output different VRAM usage for every GPU which is used for rendering,only due this you have there one output with rendering memory

    If you will have two GPU,then first GPU which is displaying the Windows/Poser etc will always take more VRAM like right now(2649MB in yours case) and second GPU will take around 1371MB,but in some instances can take same or similar memory as GPU 1 like in my case

    Splitting kernel of CUDA is not what I like or would rather have,with OCL 2.0 if yours GPU is 4GB and second GPU is 4GB,yours actual rendering memory is 8GB,in CUDA if you have 8GB VRAM and 12GB VRAM,you don't have 20GB VRAM but only 8GB,card with less memory will have priority in VRAM department and you will only gain on CUDA cores

    And sorry for confusion

    Hope this helps

    Thanks,Jura



  • I think I'm getting there... are you saying that only 1 card manages the texture memory, but both cards' cuda cores are used?
    Essentially what I'm getting is this;
    card 1 manages textures
    card 2 runs the actual render
    both cards cuda cores are available.
    ?



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

    BTW, I decided to push it a bit; I have the same scene but with 7 gen 4 characters in view (M4 and V4). Poser says it uses 5,664Mb of memory, so I've been using CPU and branched path because the GTX970 is nominally 4Gb... so I wanted to watch it fall over on memory usage on GPU-Z, but it didn't. Memory usage peaked at 3,403Mb with a bucket size of 64. So I tried a bucket size of 128, and memory usgae snuck up to 3,448...
    essentially, I'd like to be able to estimate whether I can use GPU and not have it fail, as well as understand the plusses and minuses of running 2 cards...

    Hi there

    Sometimes Poser output log will tell you how much he using etc,but actually rendering will take less memory if you are looking at GPU-Z
    You need to take to account if you add there few props and which do use several 4k maps then usage will be bit bigger and due this I would go with 6GB or 8GB cards there,with those cards you shouldn't have problems to render

    Regarding the bucket size,512 works well with new GPU's,with 1024 my times are worse as above I've said,my usual bucket size is 128 or 192,but sometimes I use 512

    Hope this helps

    Thanks,Jura


  • Poser Ambassadors

    I just did a few tests
    Rendering the scene - as is, without tinkering render settings - renders in about 17 minutes on my dual (3 years old) titan cards.
    Rendering this scene with bucket size reduced to 32 with CPU (i7, 6/12 core) did about 30 minutes.



  • @piersyf

    Hi there

    Sorry again for confusion,I will try to take you screenshots later and this should answer yours questions
    Both cards must have same assets(textures,geometries etc) or both must manage same assets and both cards are used for rendering and yes both CUDA cores are available

    Hope this helps

    Thanks,Jura



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

    I just did a few tests
    Rendering the scene - as is, without tinkering render settings - renders in about 17 minutes on my dual (3 years old) titan cards.
    Rendering this scene with bucket size reduced to 32 with CPU (i7, 6/12 core) did about 30 minutes.

    30 minutes on CPU is pretty much awesome result there
    Can you check something for me,can you check GPU usage on second GPU as with my mixed combination(Titan X and GTX1080),my Titan X is utilized not 100%,in GPU-Z looks like saw,not straight,I'm sure in Cycles my both GPU are used evenly

    Thanks,Jura



  • Here is the scene(excluding the V4,hair etc) which I'm trying to port,its pretty much done,just waiting on approval
    This has been rendered up to 70 samples and time has been 1 hour and 15 minutes,I think noise free should be in 100 samples

    0_1477280387594_Render 1.jpg

    Thanks,Jura



  • @jura11 yeah, I had set my bucket size to 32 for my test. 32 is the size i usually use for my bucket.


  • Poser Ambassadors

    @jura11 said in test scene for Superfly speed.:

    @wimvdb said in test scene for Superfly speed.:

    I just did a few tests
    Rendering the scene - as is, without tinkering render settings - renders in about 17 minutes on my dual (3 years old) titan cards.
    Rendering this scene with bucket size reduced to 32 with CPU (i7, 6/12 core) did about 30 minutes.

    30 minutes on CPU is pretty much awesome result there
    Can you check something for me,can you check GPU usage on second GPU as with my mixed combination(Titan X and GTX1080),my Titan X is utilized not 100%,in GPU-Z looks like saw,not straight,I'm sure in Cycles my both GPU are used evenly

    Thanks,Jura

    Yes, one of the card is alternating between 20% and 100% when progressive mode is active. This is not the case when progressive is off
    I did another test with progressive off and the render time is slightly faster (17 min 10 sec vs 17m 34s). Here the last bucket had only 1 card at 100% then other at 0%. So I did another test with the bucketsize of 256 to compensate for that and the render time was slightly slower again (17m 40s).

    The larger bucketsize computers faster as the lower bucketsize and that just compensates the idle time of the second card. In other scenes it may just be the other way around. It is a careful balancing act between bucketsize, rendersize and complexity distribution(SSS, transparency, etc take longer to calculate). Having 2 cards makes this more obvious



  • Thanks for the scene.

    It Renders in 55 minutes on a GTX 960.

    I Will launch test on CPU is an oldy I5 760 but just before going to bed hopping it will be done tomorrow...
    How do you set up such scenes : open gl preview is very poor ? lot of quick low quality renders or use of raytrace preview ?



  • I've done the test without changing anything with my GTX980 Ti:

    • 20 minutes with the scene, as is
    • 16 minutes if I hide the hair
      which seems reasonable, as some hair props may explode the time while rendering


  • OK guys I would upload the scene during today/tonight as they allowed me to post the scene and I would need to put across,we shouldn't compare renderers as this scene is made for Corona Alpha and due this too we shouldn't compare in this therm too,some materials are not translated well,but I done best job as I could

    I will create separate thread on this where people can post their results

    Thanks,Jura



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

    @jura11 said in test scene for Superfly speed.:

    @wimvdb said in test scene for Superfly speed.:

    I just did a few tests
    Rendering the scene - as is, without tinkering render settings - renders in about 17 minutes on my dual (3 years old) titan cards.
    Rendering this scene with bucket size reduced to 32 with CPU (i7, 6/12 core) did about 30 minutes.

    30 minutes on CPU is pretty much awesome result there
    Can you check something for me,can you check GPU usage on second GPU as with my mixed combination(Titan X and GTX1080),my Titan X is utilized not 100%,in GPU-Z looks like saw,not straight,I'm sure in Cycles my both GPU are used evenly

    Thanks,Jura

    Yes, one of the card is alternating between 20% and 100% when progressive mode is active. This is not the case when progressive is off
    I did another test with progressive off and the render time is slightly faster (17 min 10 sec vs 17m 34s). Here the last bucket had only 1 card at 100% then other at 0%. So I did another test with the bucketsize of 256 to compensate for that and the render time was slightly slower again (17m 40s).

    The larger bucketsize computers faster as the lower bucketsize and that just compensates the idle time of the second card. In other scenes it may just be the other way around. It is a careful balancing act between bucketsize, rendersize and complexity distribution(SSS, transparency, etc take longer to calculate). Having 2 cards makes this more obvious

    Thanks Wim for clarifying me this

    This I've wanted to know as I tried few bits and still trying to find perfect balance between the speed and quality
    Larger bucket size yes sometimes in some scenes is really better to use and in some larger bucket size will result in slower times

    Thanks,Jura



  • @jura11 said in test scene for Superfly speed.:

    @ghostship
    @Miss-B

    For CPU only renders I would tweak the settings for CPU as this bucket size will kill any CPU,I would suggest around 16 bucket size and please try with Branched Path Tracing,this should help you with bit faster renders and try like with Progressive or without,I find sometimes Progressive rendering on CPU can put not necessary stress on CPU which can be used for rendering

    Hope this helps

    Thanks,Jura

    Thanks for the tidbit Jura. I usually use 32 rather than the 64 I usually see by default, but I'm now going to try 16 and see how that helps.



  • I've now rendered the scene with the same render settings using my cpu, an i7-2600k @3.40ghz overclocked to 4.2ghz with a bucket size of 64 and it took 1 hour 55 minutes. First time I rendered with the cpu I forgot to change the bucket size and it took 3 hours and 9 minutes! :)



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

    31 minutes 33 seconds on a GTX 970, with other stuff running.

    How much ram does that card have? Most that I see only have 4GB.



  • 970 is nominally 4Gb, except Nvidia cheated and made it a 3.5Gb card with a 512Mb tacked on slower memory module. Nvidea have just lost a class action suit in the US over it.

    I'm currently eyeing a GTX 1070 8Gb card for $650...



  • @ghostship

    Hi there

    As above GTX970 using or have 4GB and I would suggest get something like is GTX1060 with 6GB which should be faster or similar to GTX970 and will have 6GB VRAM or I would check few classified and get GTX 980Ti which is still good performer as in general 980Ti is cut down Titan X and with right card you will have great temps and great OC potential there

    Just be careful with new "Pascal" those card are not currently supported by IRAY in DS and there is limited support in Octane too,Poser works with new "Pascal" which is very surprising but awesome too

    Hope this helps

    Thanks,Jura



  • That's fine by me... I can't afford Octane and I wouldn't use Studio if they paid me...



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

    That's fine by me... I can't afford Octane and I wouldn't use Studio if they paid me...

    Lol...