Poor performance with non-flashed GeForce?

Discussion in 'Mac Pro' started by bwillwall, Feb 25, 2018.

  1. bwillwall macrumors 6502a

    bwillwall

    Joined:
    Dec 24, 2009
    #1
    So I installed a 2GB GTX 960 in my Mac Pro 3,1. I know it's not a high end card or anything but I figured if I upgraded the GPU in this machine it might make it still usable coming from a GTX 8800. I'm using the Nvidia web drivers of course and I even heard that using the latest drivers hurt performance big time on macOS 10.13.3 so I used a 3rd party installer that patched the driver 378.10.10.10.25.106 to work on High Sierra 13.3. All seemed well until I tried doing anything graphics intensive and realized my performance was no better if not worse in Final Cut Pro X, in gaming it seems better but certainly not to the extent I would expect a 960 to perform at.

    I read something about how non-flashed cards don't interface with the Mac Pro 3,1 through the highest bandwidth the PCI slot can handle? Could this be causing my issue? I have it in slot 2 16X. If flashing the card would improve performance could somebody show me where you find compatible cards for flashing for Macs? This one is an EVGA SuperClocked GeForce GTX 960 with 2GB GDDR5.
     
  2. h9826790 macrumors G5

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #2
    No, it has nothing to do with the bandwidth. A x16 slot is more than enough for a 960. And web driver enable PCIe 2.0 in MacOS anyway (if the hardware supported). Flash won’t make any difference.

    Your gaming performance may be CPU limiting.

    And FCPX is particular bad in optimising for Nvidia GPU.
     
  3. bwillwall thread starter macrumors 6502a

    bwillwall

    Joined:
    Dec 24, 2009
    #3
    I could check the CPU use to confirm on gaming, but can't imagine basic games are maxing out all 8 cores. I get that Final Cut runs better on AMD cards but why would a GTX 8800 run better than a GTX 960?
     
  4. h9826790 macrumors G5

    h9826790

    Joined:
    Apr 3, 2014
    Location:
    Hong Kong
    #4
    Games may no even able to utilise 2 cores, of course can’t max all 8 cores. It’s about CPU single thread performance, not multi threads performance. 99.99% software on the world cannot utilise all 8 cores of the CPUs, but they can be CPU single thread limiting.

    Don’t know, I never own the 8800, not sure how good the optimisation in FCPX. But which parts of video editing you actually experience that the 8800 do better than 960?
     
  5. orph macrumors 68000

    orph

    Joined:
    Dec 12, 2005
    Location:
    UK
    #5
    for games, some games are limited to 1 thread (cpu core) some can spread better, but even games that can use more than 2-4 cores may have a "master" thread that is limited by the speed of a single core and after that cant scale well (ie one thread that runs a job that must be done and jobs cant be handed of to more cores until that task is done which limits how well the game can scale to more than 1 cpu so it's common to see one core at 100% then 2-3 cores at 30-40%).
    and on the mac side, games tend just to run bad compared to windows that simple.

    the3.1 cpu is slow by today's speeds, my 5.1 is a lot faster with the same GPU - GTX660/770 (i have a 3.1 2.8gzh 8c + 5.1 3.33ghz 6c) and my 5.1 3.33ghz 6c is also slow by today's speeds.

    when comparing speeds in FCX make shore your doing fair comparison of the two cards and that you are not introducing something that is making the change in speed & that the task you are doing is actually using the GPU or limited by the GPU.

    id be amazed if the 960 is slower than a GTX 8800 unless there's some massive bugs, the GTX 8800 must be bairly suported with only 768mb of vram which is just there for 1080p editing for most apps even with not the best drivers the 960 must be a lot faster.

    might be worth checking some benchmarks
    for open CL speed LuxMark is worth a go http://www.luxrender.net/wiki/LuxMark#Binaries
    (just do the simple tests think the complex one is buged in osx)
    for CUDA OctaneBench is a good way to compare nvidia cards
    https://render.otoy.com/octanebench/

    mp5.1 3.33ghz gtx770 4gb (in osx)
    LuxMark v3.1 - neumann TLM-102 se (medium benchmark) scores 2738
    OctaneBench 3.06.2 my gtx770 scores 48.30
     

Share This Page

4 February 25, 2018