Loading...
> >

Benefits of two different video cards?

SnowSultanSnowSultan Posts: 3,805
in The Commons

Just wanted to double check something with those of you who might have first-hand experience with this particular situation. Does Iray (or other 3D-related tasks) gain any advantages from having two *different* video cards installed?

I currently have an Nvidia GTX 760 and am considering upgrading at some point to an Nvidia GeForce 980 Ti. If I do, would it benefit me to keep the 760 as well? I mainly ask because I would have to buy a new power supply to run both (I have a 600W, calculations recommend I would need at least 711W to run both).

Thanks in advance.

Comments

  • Richard HaseltineRichard Haseltine Posts: 109,700

    Each card will be used, if checked in the Advanced tab of Render Settings, for any scene that fits its RAM.

  • fixmypcmikefixmypcmike Posts: 19,710

    You could also use the 760 for driving your display and other applications and the 980 for rendering, if you need to use your computer for other things while rendering.

  • SixDsSixDs Posts: 2,384

    Hey, Snow. The answer is that it depends on which two cards. If the cards are identical, then the answer is yes, since it brings double the number of CUDA cores to the table. Yours, however, are not identical, so the answer, for rendering purposes, is no. The problem is the onboard video memory. I don't know whether your GTX 760 has the standard 2 GB memory or if it is the 4 GB variant, but either way that card will bottleneck the 980 Ti, which has 6 GB of video memory. Essentially, when used in conjunction for rendering, the entire scene must be loaded into the memory of each card. This means that the scene must fit in the memory or frame buffer of the 760, and if it only has 2 GB (or possibly even if it has 4 GB), the render engine may default to CPU mode if the scene is too large. This will happen even if the scene will fit in the frame buffer of the 980 Ti, since the limits of the smaller card will prevail. In such circumstances, you are better off using the 980 Ti alone. There is another scenario: you could configure the system to use the 980 Ti for rendering, and use the GTX 760 to run your display, freeing the 980 from this task. That would put the 760 to use with at least some benefit, but, frankly, that isn't a terribly productive use of the 760;s capabilities, especially if you need to purchase a new PSU to run both.

  • namffuaknamffuak Posts: 4,470

    Well - yes, no, and - maybe.

    If the scene fits into the meory of the 760 then Iray can use both cards and finish the render faster. OTOH, you could keep the 760, NOT use it for Iray renders, and hang your monitor(s) on it. This will free up 50 to 300 MB of memory from the 980 - all the OpenGL for Studio as well as whatever else you are running during the render. In this case the 760 will be using minimal power and you shouldn't have power supply issues.

    FWIW, I have a 980 ti and at 97% utilization my power consumption (as reported by GPU-Z) runs just over 80% or 200 watts. I seldom get over 97% because I haven't tweaked the voltage parameters on the card and GPU-Z is indicating I'm capped by voltage limitation.

    One other item - once you get the 980 TI running, use the management tool that comes with it to change the default fan settings - ramp them up sooner - and the card will run MUCH cooler than the out-of-the-box settings. I went from a 45% fan speed and 83 degrees to a 63% fan speed and 63 degrees with no noticeable change in sound level.

  • SnowSultanSnowSultan Posts: 3,805

    Thank you all for the replies. I haven't had a problem with an Iray scene fitting on the 2GB of my 760 so far, but it's good to know how renders are handled with two different cards. It sounds like it wouldn't hurt to leave the 760 in and just uncheck it in the Render Settings if I have a feeling that a scene is going to require the 980 Ti (if I end up buying a new PSU).

    I've heard a lot of varying things about the power requirements for a card like the 980 Ti, and I'm honestly still concerned if 600W would be enough. Power Supply Calculator has shown recommended wattage values of 505-511 for my system if a 980 Ti was the only card installed. 

     

    Thanks for that inro Namffuak, I don't care about noise (my PC is already pretty loud), so I'll be sure to turn up any fan settings if I get the 980.

     

     

  • namffuaknamffuak Posts: 4,470
    SixDs said:

    Hey, Snow. The answer is that it depends on which two cards. If the cards are identical, then the answer is yes, since it brings double the number of CUDA cores to the table. Yours, however, are not identical, so the answer, for rendering purposes, is no. The problem is the onboard video memory. I don't know whether your GTX 760 has the standard 2 GB memory or if it is the 4 GB variant, but either way that card will bottleneck the 980 Ti, which has 6 GB of video memory. Essentially, when used in conjunction for rendering, the entire scene must be loaded into the memory of each card. This means that the scene must fit in the memory or frame buffer of the 760, and if it only has 2 GB (or possibly even if it has 4 GB), the render engine may default to CPU mode if the scene is too large. This will happen even if the scene will fit in the frame buffer of the 980 Ti, since the limits of the smaller card will prevail.

    Ah - no.

    Iray will start with both (all available selected) cards; if the scene exceeds the memory capacity of a card it gets dropped. This is on a card-by-card basis, so you could have a 2 GB, 4 GB, and 6 GB set of cards and select them all. Most likely anything beyond a simple scene with one figure would exceed 2 GB and that card would drop off the render, leaving the 4 GB and the 6 GB cards to do the work. I don't know where the cutoff is for 4 GB - I apparently don't do anything complex enough. :-) The thing to remember is that Iray treats each card as a separate entity.

  • SixDsSixDs Posts: 2,384

    "if the scene exceeds the memory capacity of a card it gets dropped"

    Aha. I somehow managed to miss that salient point, Namffluak. Thanks for the correction. I realized that CUDA treats the cards separately, which is why SLI won't work, but didn't realize that it would drop only cards that didn't meet the requirements. It does raise an interesting point, however. What advantage would there then be to excluding any CUDA capable card from the render engine and using it for display purposes only, regardless of how much memory it might have? That seems to be the conventional wisdom. As long as the possibility exists that it might be used for some rendering tasks, then it would seem logical to let CUDA take care of it, no?

  • mjc1016mjc1016 Posts: 15,001
    edited October 2015

    Because to actually use a card just for display purposes, it must be the one with the monitor(s) attached to it.  By leaving it out of the 'pool' of available cards, you won't get into a situation where it will bog down the display refresh.  Unless you are a gamer, too, and need more than 1 GB for gaming, a 1 or 2 GB card is more than enough for that purpose...and most scenes would be very likely to be larger than 1 GB and many are bigger than 2 GB.  So, if it is likely to be dropped, anyway, why bother with it in the first place?

    It becomes a trickier question when you are dealing with bigger than 2 GB cards.

    Post edited by mjc1016 on
  • namffuaknamffuak Posts: 4,470
    mjc1016 said:

    Because to actually use a card just for display purposes, it must be the one with the monitor(s) attached to it.  By leaving it out of the 'pool' of available cards, you won't get into a situation where it will bog down the display refresh.  Unless you are a gamer, too, and need more than 1 GB for gaming, a 1 or 2 GB card is more than enough for that purpose...and most scenes would be very likely to be larger than 1 GB and many are bigger than 2 GB.  So, if it is likely to be dropped, anyway, why bother with it in the first place?

    It becomes a trickier question when you are dealing with bigger than 2 GB cards.

    Case in point; I have a GT 740 (4 GB). For a while, it was my only card - and if I used it for Iray renders, screen refresh pretty well stopped. Even Windows Solitaire became unplayable. I now have a 980 TI for renders - and if I add the 740 to the render I get very little improvement in render time at the cost of a slow and unresponsive monitor.

  • mjc1016mjc1016 Posts: 15,001
    namffuak said:
     

    Case in point; I have a GT 740 (4 GB). For a while, it was my only card - and if I used it for Iray renders, screen refresh pretty well stopped. Even Windows Solitaire became unplayable. I now have a 980 TI for renders - and if I add the 740 to the render I get very little improvement in render time at the cost of a slow and unresponsive monitor.

    I was thinking specifically of you when I wrote that...I know you've said that a couple of times but I didn't want to hunt down one of the threads you detailed it in.

  • namffuaknamffuak Posts: 4,470
    mjc1016 said:
    namffuak said:
     

    Case in point; I have a GT 740 (4 GB). For a while, it was my only card - and if I used it for Iray renders, screen refresh pretty well stopped. Even Windows Solitaire became unplayable. I now have a 980 TI for renders - and if I add the 740 to the render I get very little improvement in render time at the cost of a slow and unresponsive monitor.

    I was thinking specifically of you when I wrote that...I know you've said that a couple of times but I didn't want to hunt down one of the threads you detailed it in.

    Yeah - you want frustration, that's the way to go! Take this card - no, this card, this card and put it here - no,here - here!

Sign In or Register to comment.