Loading...
> >

GTX 970 or R9 390 for 3D/Video GPU rendering? CUDA or OpenCL way

winxtecwinxtec Posts: 2
in Technical Help (nuts n bolts)

Hi. I'm wondering which GPU should be better of a deal in a long term. I am planning to make use (single GPU) for 3D rendering and probably Video editing renders from the GPU, instead of the CPU.

Information to consider:

  • It is a budget build, only can spend $350 at best for the GPU, and both are in the same range of price.
  • Specs: Intel Xeon 1241 v3 @ 3.9GHz (4 cores with HT), 16GB DDR3 1600MHz (will add more RAM eventually), SSHD 1TB, Windows 7/10.
  • The purpose of the GPU it is also inteded for gaming (not above 1080p resolutions for now).
  • Programs for 3D and Video:
    • Adobe products: After Effects, Photoshop, Premiere Pro.
    • Autodesk products: Maya, 3DS Max, AutoCAD. I guess if a choose AMD I won't be able to use Mental Ray or iRay.
    • Other softwares: Daz3D, Octane 3 (now can use OpenCL), Zbrush, Cinema 4D, E-on Vue 15, Sony Vegas, RealFlow.

 

What are your thougths with these GPU's, which one do you recommend within these conditions?

Thanks for your time by helping. yes

Comments

  • nicsttnicstt Posts: 11,714

    Decide what software you primarily want to use first; some might accommodate both, but be better optimised for a specific card, or vendor's product; others will be vendor specific - Daz, using IRAY is Nvidia cards only; all allow CPU rendering, and you might find what you require gives you the opportunity to use a render farm online, which then changes your requirements.

    Daz and IRAY:  the most recent cards from NVidia: 900 series; the 970 probably gives best value, with the 980ti coming in closely after; 980ti, if it is within budget gives great render performance and a reasonable amoung of memory for scenes; the Titan X gives roughly the same performance as the 980ti, but provides twice as much memory. Older cards from NVidia can give good performance but can be memory limited.

    Don't use SLI and dual GPU cards don't offer more memory for rendering as it only currently uses the momory from one card. (It is unknown if Direct X 12, a windows 10 only option will change this.)

  • ToborTobor Posts: 2,300
    edited September 2015

    Even a few hundred CUDA cores will be faster than your CPU alone. Go with the nVidia card if you want to render with Iray. Lots of folks here use the 970. But be aware it's a 150W card, and it needs two 6-pin power connectors. Many PS's only have one 6- and one 8-pin, so you'll need an adapter for the second 6-pin. You can get an 8- to 6-pin aapter, or a 6-pin adapter that connects to SATA or Molex power. For the latter you'll need to tap from two (15-20 amp) 12V rails. 

    Post edited by Tobor on
  • digitalus49digitalus49 Posts: 20

    Stick with CUDA on Nvidia. It is by far the most efficent why to render I know. I have seen the GTX 980 Ti work wonders - it's a nice card. Similar specs to the Titan Black.

  • winxtecwinxtec Posts: 2

    Guys, the 980Ti is out of the price range. It is a budget custom built.

     

    Tobor said:

    Even a few hundred CUDA cores will be faster than your CPU alone. Go with the nVidia card if you want to render with Iray. Lots of folks here use the 970. But be aware it's a 150W card, and it needs two 6-pin power connectors. Many PS's only have one 6- and one 8-pin, so you'll need an adapter for the second 6-pin. You can get an 8- to 6-pin aapter, or a 6-pin adapter that connects to SATA or Molex power. For the latter you'll need to tap from two (15-20 amp) 12V rails. 

    I have no problem with an appropiate PSU for the card. I am planning to get a Corsair CXM 750W or might reduce the capacity to save some few bucks.

    It is kinda a bummer the thing with the 970, supposedly it slows down when it is using more than 3.5 GB of VRAM. Meanwhile, the R9 390 has 8GB with 2560 Shader cores, but no CUDA, but it seems more future proof than the 970 (3.5GB with 1664 CUDA cores but more efficent).

    Perhaps I am overthinking it, should I just go with Nvidia already. Hopefully DX12 can extend the performance of the 970 as in games as well as it is supposedly marketed.

  • mjc1016mjc1016 Posts: 15,001
    winxtec said:
     

    It is kinda a bummer the thing with the 970, supposedly it slows down when it is using more than 3.5 GB of VRAM. Meanwhile, the R9 390 has 8GB with 2560 Shader cores, but no CUDA, but it seems more future proof than the 970 (3.5GB with 1664 CUDA cores but more efficent).
     

    Don't worry about it...it is only important when gaming and frame rate is critical.  Basically if the same thing is being rendered for more than a fraction of a second, you will probably not notice the difference.  A render that will take several minutes or more is not really going to be impacted by something that slows 100+ frames per second down a few percent.  What has more of an impact is whether or not the card is GDDR5 or not.

  • nicsttnicstt Posts: 11,714
    edited September 2015

     

    Tobor said:

    Even a few hundred CUDA cores will be faster than your CPU alone. Go with the nVidia card if you want to render with Iray. Lots of folks here use the 970. But be aware it's a 150W card, and it needs two 6-pin power connectors. Many PS's only have one 6- and one 8-pin, so you'll need an adapter for the second 6-pin. You can get an 8- to 6-pin aapter, or a 6-pin adapter that connects to SATA or Molex power. For the latter you'll need to tap from two (15-20 amp) 12V rails. 

    Depends on the GPU the cores are on and the CPU, sorry but this is a misleading statement.

    The card I use for display until I upgrade again is a http://www.geforce.co.uk/hardware/desktop-gpus/geforce-gt-640-oem/specifications

    It is slower than my CPU; not only that it only has 2GB ram so is further limited. My CPU is an i7 but there are much faster out there. So whilst there will be many cards faster than a CPU, it is necessary to compare.

    Post edited by nicstt on
  • ToborTobor Posts: 2,300
    edited September 2015

    That would be true generically, but we're talking specifics here The OP provided his CPU, pretty close in to the 1230 Xeon I have (the 1241 is about 10-15% faster). I based my comment on real0-world results that I've gotten.

    Here are specs using a 740 as the main card, which has 384 cores, and a 620 which is for the monitor, and its measely 96 cores when the scene will fit inside 2GB. 

    Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info :     CUDA device 1 (GeForce GT 620): 137 iterations, 114.192s init, 759.590s render
    Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info :     CUDA device 0 (GeForce GT 740): 432 iterations, 64.344s init, 809.396s render
    Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info :     CPU (4 threads): 113 iterations, 30.352s init, 843.338s render

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