Loading...
> >

Iray 4 GPU test

nickalamannickalaman Posts: 196
in Daz Studio Discussion

For my day job I need lots of CUDA cores for video, Adobe Premiere / DaVicni resolve, etc... so I'm lucky enough when I’m not editing wedding videos, I've got 4 GPUs available for Daz... Anyway, I’ve been reading here in the forums that studio can not handle more than 2 GPU unless you have some monster system with 8 cores and lots of memory. That's not my case, I currently have 32GB of memory and an I7 4770K overclocked to 4.4GHZ. I redid some of my tests by calculating the iterations per minute for each of my cards and then running them all at once. The point is to see how efficient the card are when running together vs single. 

   
980ti 111
Titan Black 82
Titan Black 82
780GTX 80
Total Theoretical (by adding up each card) 355
Total Actual (by measuring actual iterations) 321

 

My conclusions are that iray scales very well assuming you are using 4 identical cards, if the cards have mixed speeds, in my case the titan black are the slowest cards, running at about of 890mhz and the 980ti the fastest running at 1250mghz . It looks like iray bring everything down to the lowest common speed, in this case it would be the titan blacks speed of 890mhz. So the 980ti instead of running at 1250 it slows down by 30%. to match the speed of your slowest card. For those thinking of mixing pascal cards with older gen cards like mine, think twice. Those new pascal cards are running at close to 2000mghz overclocked, but if you pair one of them with say a 780, according to this logic it will drop it's speed by half to keep up with the 780. 

I've uploaded a video on youtube, i don't see any slow down with 4 cards....what I’m i suppose to see? can you point it out? is there something else i should be testing? is it a windows vs mac thing? i doubt the osx version of iray is more efficient than the windows version

 

Comments

  • Jim_1831252Jim_1831252 Posts: 728
    edited June 2016

    Interesting. Did you test your conclusion by Testing the two Titan Black cards? Also, I'd be interested to see some OctaneBench results for your rig.

    I haven't heard the comments about needing an 8 core CPU for more than two cards, I assume that is only if you want to run your CPUs at x16 PCIe lanes each, though from my own tests the performance difference between 16, 8, 4 lanes on PCIe 2 and 3 is not significant, at least not with my current builid.

    Edit: not that you can have four cards concurrently accessing 16 PCIe lanes, even with high-end LGA 2011-3 processors.

    Post edited by Jim_1831252 on
  • nickalamannickalaman Posts: 196

    It's actually a titan Z wich is eqivilent to 2 titan blacks.. I ran each card seperatley, calulated it's score, then I ran all 4 cards at the same time.

     

    Screen Shot 2016-06-27 at 8.17.59 AM.jpg
    911 x 460 - 112K
  • Jim_1831252Jim_1831252 Posts: 728

    Thanks. Did you find the performance of your cards behaving as you found with Iray? I was surprised (and disapointed) to hear your cards tended to bring their performance down to the slowest card. I have to 780ti cards, so I can't test it out.

  • nickalamannickalaman Posts: 196

    Beside the slowdown becuase of the titan, they work pretty much the same.

     

Sign In or Register to comment.