Dual Video Card Setup

I found my renders taking too long and ordered a Titan Z to replace my GTX 970.

Is there a way to set up my system with both cards, so I get the benefit of all of the CUDA cores?

Are there any guides on how to do this to ensure that the renders are usingthe CUDA from both cards?

Comments

  • Good timing on this question as I just added a second card to my system.  I have a 970, and was able to pick up a 960 for dirt cheap to hold me over until I order a couple of Titan X's in the spring.

    No real guide is necessary.  Just install the new Titan in your system.  However, you might consider doing a clean install of the Nvidia drivers.  While not totally necessary, it doesn't hurt to do it just to make sure you're starting from a clean slate.  There is a free utility called Driver Sweeper that will completely remove the existing drivers.  I think it's on guru3d.com, but you can find it via Google.  Once that's done, just install the latest drivers from Nvidia's web site.  The same driver will support both the Titan and the 970.

    Once your system is up and running on the new drivers, you just need to go in to the Advanced tab of the Render settings pane in Daz Studio and click the check boxes next to both the 970 and the Titan.

    The thing to remember is that the 970 will be used as long as the scene fully fits into the 4GB of memory it has.  If it's larger, then only the Titan will be used.  In your configuration, I would run the displays off of the Titan, since it has more RAM.  

    Something just occurred to me, though.  The Titan Z is a dual-GPU card.  It may show up twice in the render settings pane, though I don't know this for certain.  If so, just check both instances.  The other thing is that I believe the 12GB of RAM is split between the two GPUs, so scenes must fit in 6GB to be handled by the Titan without failing over to CPU rendering.

    One more thing, if you're on Windows 10, make sure you have installed all updates.  There was some funkiness with dual Nvidia setups where the cards were from different generations (the Titan Z is lumped in with the 700 series). I think this was resolved in a November update.  

    Ok, so this ended up almost being a guide.  Good luck, and enjoy your new horsepower!  Even without the 970, you'll be rocking over 5000 cuda cores.  Jealousy ensues.

     

  • edited December 2015

    Great, thanks for the advice. Looking forward to its arrival early next week.

    Sounds like it s as simple as leaving both cards plugged in?

    Post edited by bueller1998_df4ca4b697 on
  • nicsttnicstt Posts: 11,714
    edited December 2015

    deleted

    Post edited by nicstt on
  • edited January 2016

    Titan Z Just Arrived. GTX 970 also still in system. Almost 7500 CUDA. Six Core i7 CPU running.

    Rendering "Dream Closet." 1 G3 male, 2 G3F. 2 Power Hair, one Pen hair. Sort of Near Mirror. No extra lights. Dream Closet has built in illuminating lights that reflect in mirror. The list of items is so long I can't figure out how to stop those lights from illuminating.

    Two hours, thirty minutes into the render and I am at 92% completion.

    Seems like it should work better than this.This is my test render and maybe I could experment or some of you can suggest some tricks. But I think Iray needs some improvement, too.

    :(

     

    Post edited by bueller1998_df4ca4b697 on
  • JCThomasJCThomas Posts: 254

    What dimensions are you rendering at? Honestly, that sounds about right.

    You're probably using more than the 970s memory and so you're going "out-of-core," which is what it's called in octane. If that's the case, something else could be bottlenecking the render. Any subd? COuld be a lot of polygons too.

    Try it with just the Z and no cpu render either. If you consider that the scene you're describing would probably take days in Reality or Luxus, 2 hours 30 minutes is pretty good.

  • I know that feeling haha. After spending that much money you want it to just render in light speed. However, it is all relative. Renders that used to me 7 hours are now done in 30 minutes. Renders that took 24 hours are now done in a few hours. It still seems like a long time but it is so much faster than it was!

  • hphoenixhphoenix Posts: 1,335

    Bueller, you are going to be kicking yourself when the Pascal core nVidia cards debut in the next few months..... But yes, I know how hard it is to resist the urge to make Daz render faster......

     

  • hphoenix said:

    Bueller, you are going to be kicking yourself when the Pascal core nVidia cards debut in the next few months..... But yes, I know how hard it is to resist the urge to make Daz render faster......

    Any word on exactly when and how much it will cost? I've started thinking about my next system, and what little I've read is tempting. VERY tempting.

  • JCThomasJCThomas Posts: 254

    Nothing official, but I'd expect four figures based on performance specs. If these next cards are going to make Titan X obsolete and have twice as many CUDA cores, they're not going to cost 500 bucks.

  • hphoenixhphoenix Posts: 1,335
    edited January 2016

    What HAS been posted indicates about 1.5 - 2x the memory (for the same class of card) and about double the CUDA cores as well.  nVidia stated that it was a '10x the performance' over Maxwell.  ( http://wccftech.com/nvidia-pascal-gpu-gtc-2015/ )

    No hard and fast data yet, but estimates are putting the 1080ti as having 8-16GB of HBM2 memory (the newest type) and the Titan equivalent having 16-32GB.

    Pascal's being fabbed at TSMC, using their new 16nm FinFET process.  Huge boosts there.  17 billion transistors on die.

    I'd expect PEAK performance numbers listed like above.  Real-World performance boosts I'd estimate as half what is claimed.  Which is still HUGE.  And the amount of memory the new cards will sport is crazy (HBM2 is 3D structured memory, stacked cells, higher bandwidth, and AMD already has a shipping product, its Fiji XT GPU, with it so it's proven to work.)

     

    Note, the pattern for nVidia is pushing out the highest-end professional cards FIRST, then consumer grade.  So we'll see Quadro and Titan first (and possibly an updated VCA) and after a few months of drooling over specs they'll start releasing the consumer-grade stuff.  I'd expect the first 1000-series (which is also just a guess, no idea what they'll number/call it) consumer cards around July-Sept, with all the main variations available just before xmas 2016.

     

     

    Post edited by hphoenix on
Sign In or Register to comment.