Best Nvidia GPU's in cMP..

Discussion in 'Mac Pro' started by violst, Feb 9, 2017.

  1. violst macrumors 6502

    violst

    Joined:
    Jun 14, 2012
    #1
    I was wondering what the best dual Nvidia GPU set up is in the cMP without having to use an external power supply?

    I would like to start using octane render for OS X and would like to maximize GPU render power without to much tinkering with external power supplies if possible.

    Thanks!
     
  2. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
  3. jeff7117 macrumors regular

    Joined:
    Jul 22, 2009
    #3
    While not internal, this is an alternative solution that takes the internal power supply out of the equation. Take a look at this thread on the Amfeltec GPU expander https://forums.macrumors.com/threads/el-capitan-10-11-and-cubix-expander.1925054/#post-22288628

    I ran one of these with 3 GPU's in my 5,1 for about a year when I got started with Octane. I've since switched to a PC but the Amfeltec was a workable solution, though the 1x PCI connection was noticeable on heavy scenes.
     
  4. PowerMike G5 macrumors 6502

    Joined:
    Oct 22, 2005
    Location:
    New York, NY
    #4
    Yeah, I think x2 Quadro M4000 would work as well, if you need single height solutions to save PCIE slots. But definitely more expensive.
     
  5. violst thread starter macrumors 6502

    violst

    Joined:
    Jun 14, 2012
    #5
    Thanks guys,

    Thats my dilemma! I have worked on a Mac For all of my career and now its limitations are just to glaring to avoid. I currently have the nMP but was thinking about going back to a 5,1 so I can use nvidia GPUs's with octane and stay on OS X.

    I have years of time machine backups, fonts, software and so on. It seems daunting to make the switch. How was your transition jeff7117.

    Thanks
     
  6. jeff7117 macrumors regular

    Joined:
    Jul 22, 2009
    #6
    Yeah, I've worked with Macs for a long time. Overall, the switch to PC has been relatively painless. Except for Prores, and a few random programs that weren't available for PC, jumping ship was pretty easy. Octane was really the driving factor for me. After the Amfeltec chassis, I built a small PC for network rendering with Octane. Once I got my feet wet with Windows, it was about 6 months before I finally switched for good. The hardware was just too fast and flexible to overlook, and I'm glad I did switch. There was recent news on the Octane forum that they may never be able to release V3 for ATI cards (apparently ATI is dragging their feet on some driver fixes) so you may never be able to use Octane with the nMP.


    http://www.schoolofmotion.com has a nice three part series for artists and designers who are thinking about switching to PC from mac. They cover a lot of software solutions and have a great hardware breakdown for building or buying a new system.

    Unfortunately there really isn't an elegant solution to using Nvidia cards if you have a nMP. For the money you would spend on an older 5,1, you could really build a nice PC for Octane.
     
  7. PowerMike G5 macrumors 6502

    Joined:
    Oct 22, 2005
    Location:
    New York, NY
    #7
    I believe you can use one of those external thunderbolt eGPU boxes with the nMP if you wanted to have an Nvidia option with your current machine. I think Barefeats showed the benchmarks using GTX 980ti.
     
  8. violst thread starter macrumors 6502

    violst

    Joined:
    Jun 14, 2012
    #8
    Thanks jeff7117 for your experience with the switch to a PC and the links to the PC to mac article!

    PowerMike G5 I thought that the eGPU still wasn't working correctly with the nMP and thunderbolt 2?
     
  9. itdk92 macrumors 6502

    itdk92

    Joined:
    Nov 14, 2016
    #9
    Titan X (Maxwell) is the best card atm, beside higher end stuff like the M6000 and pascal line.

    GTX980Ti is the closest competitor, and more affordable.
     
  10. PowerMike G5 macrumors 6502

    Joined:
    Oct 22, 2005
    Location:
    New York, NY
    #10
    I believe this one works with the nMP:
    https://bizon-tech.com/us/bizonbox2s-egpu.html/

    I'd research more into it. though as I don't use it.

    Also, I use the Titan X in my cMP and its fantastic for my Adobe CC work. Might be something worth looking into as well, given its almost the same CUDA cores as 2x 970s or 2x M4000.
     
  11. itdk92 macrumors 6502

    itdk92

    Joined:
    Nov 14, 2016
    #11


    ^ totally agree.

    Also:

    1 - not all software can take advantage og two GPUs, so one good GPU might be better in some cases.

    2 - the price of 2 x M4000 is crazy and definitely not worth it.

    3 - 2x Titan Xs or 2x GTX980Ti might still be the best option for the highest GPU performance, with 2x 980Ti being the most cost effective
     
  12. h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #12
    Since the initial OP post mention about "without external power", and for Octane render.

    Unless OP willing to go through the Pixlas mod. I suspect 2x970 actually work better than a single TitanX. Technically, 2x970 can safely draw as much as 300W from two PCIe slots and two 6pins. A single TitanX can only safely draw 225W. Even though we know that the stock TitanX (1000-1075MHz) 250W power draw is OK in the cMP, but that's technically exceed the official limit. And really right at the edge of the real world limit. If the card draw a bit more may activate the cMP's shutdown protection.

    On the other hand, stock 970 (1050-1178MHz) only 148W TDP, just within the 150W limit. So, even without OC, 2x 970 has more CUDA core, higher clock speed, and most likely lower temperature. In fact, since everything still within the official limit, there is actually some room to OC the 970s and make them even faster.

    So, IMO, even OP didn't ask for dual card setup in the original post. As long as OP have empty PCIe slots (need 2 slots for the 2nd 970). 2x 970 should be a better option under that specific condition. Two card not always work better, however, an extra slot do always able to provide an extra 75W, which can eventually turn into more performance. For the same Maxwell architecture, 300W power avail should provide better result than only 225W avail.
     
  13. itdk92 macrumors 6502

    itdk92

    Joined:
    Nov 14, 2016
    #13
    But...

    to safely power 1 x 970, you need:

    1 x mini6pin -> 8pin
    1 x mini6pin -> 6pin

    and to safely power 2 x 970, you need:

    (Card 1)
    2 x sata power -> 6pin
    1 x mini6pin -> 8pin
    +
    (Card 2)
    2 x sata power -> 6pin
    1 x mini6pin -> 8pin

    ... my point being - that to achieve something in the league of the Titan X, you actually need to drop 4 SATA ports with the GTX970

    For 1 x Titan X you would need to drop 2 x SATA ports, and wire it like this:

    2 x SATA power + 1 x mini6pin -> 8pin
    1 x mini6pin -> 6pin


    So, strategically speaking, the Titan X might be better.

    With Pixlas mod, a 2 x GTX970 setup would be a piece of cake, safer, and more cost effective.

    Pixlas mod + dual Titan X powered internally remains the most powerful setup.
     
  14. h9826790, Feb 11, 2017
    Last edited: Feb 11, 2017

    h9826790 macrumors 604

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #14
    TBH, I didn't expect to drive any card by any SATA port.

    I expect to drive the TitanX simply by mini 6pin -> pin & mini 6pin -> 8pin. And that's why I mentioned the TitanX can just fit inside the real world cMP limit without trigger the shutdown protection. If using the SATA port's power, the TitanX should never able to shut down the cMP.

    The standard reference 970 is just dual 6pin, technically no difference then a single 8pin's power draw limit. And it's been proved that a single mini 6pin -> dual 6pin works fine on the 970. Also, it's not about how many pins, but how much it actually draws. The 970 will draw a bit more than 150W under max loading. I suspect that's why Nvidia use 2x 6pin on the 970 rather then single 6pin for a 148W TDP card. However, since it's just a 148W TDP card. Even though the card come with 8+8pin, it won't really draw 300W via the cables, but just the cable are rated up to draw 300W safely.

    I don't have a 970 on hand. However, I doubt if both 6pin on the stock reference 970 will draw more than 55W at the same time. If not, then we can still drive a 970 with single SATA port + mini 6pin. In this case, each card can theoretically draw up to 205W safely, the power avail is way above the card's TDP. Actually, this can even provide more room for dual 970 to OC.

    Anyway, I totally agree that Pixlas mod + dual TitanX is the best can go at this moment.
     
  15. violst thread starter macrumors 6502

    violst

    Joined:
    Jun 14, 2012
    #15
    To take full advantage of octane you really need multiple nvidia GPU's the more the merrier. The Bizon box looks great but it can only handle 1 GPU. Thats why I asked about the cMP because if I could put 2 980's in it, it may be worth a try.

    unfortunately for me the PC is looking like a more realistic option in the future. Unless Apple can pull something out of their hat with a new and improved Mac Pro, but that doesn't look likely!
     
  16. PowerMike G5 macrumors 6502

    Joined:
    Oct 22, 2005
    Location:
    New York, NY
    #16
    The 'cleanest' way for the cMP would be using 2x Quadro M4000. They each will run off 1x 6pin power, so the 2 internal ports will work fine. They are also single height, so they will not block the other 2 PCIE slots, which you can use for other upgrades. And each card has 8GB VRAM.

    Whether the cost is worth it for you is something only you can answer.
     
  17. jeff7117 macrumors regular

    Joined:
    Jul 22, 2009
    #17
    Nvidia Gaming cards (Titan X, 980ti, ect...) are just as fast if not faster than Quadro cards when it comes to Octane. Quadro cards are terrible when it comes to rendering with Octane, especially when cost is considered.
     
  18. deppest macrumors member

    Joined:
    Oct 6, 2009
    #18
    I am running a titan x and a 980ti with pixlas mod, using mostly thea render. Couldn't be happier. Individually, 980ti is a bit faster.
     
  19. violst thread starter macrumors 6502

    violst

    Joined:
    Jun 14, 2012
    #19
    Thanks for all of the great info everyone! It gives me a lot to think about.
     
  20. itdk92 macrumors 6502

    itdk92

    Joined:
    Nov 14, 2016
    #20
    But cost aside, how would it compare to a 980/980Ti/TitanX double setup?

    Would double precision even work on the Quadro, since the drivers are kind of stuck at the K5000 version?

    How about a 3 or 4 x M4000 setup?

    Do you have any experience with any of this Mike? :)
     
  21. PowerMike G5 macrumors 6502

    Joined:
    Oct 22, 2005
    Location:
    New York, NY
    #21
    A double 980/980Ti/TitanX setup will certainly be faster than 2x M4000s. But the OP asked for the best dual card solution to run in the cMP without having to deal with external power, which the 2x M4000s can do. I run a single Titan X and already with that, I am routing power from 2 SATA ports from the optical bay, in addition to using the 2 min-6pin connectors on the motherboard. The 2x M4000 will run off the 2 mini-6pin connectors alone.

    I believe Octane Render scales linearly with your GPU power, so the more CUDA cores overall, the faster it will run, given the same level of GPU architecture.
     

Share This Page