Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Best CUDA performance - Fermi vs Maxwell vs Kepler vs Pascal vs Tesla
  • As I spent lots of time in front of my edit station for my feature, here are the results of my research into buying CUDA cards for Resolve, Hitfilm and Mercury Engine performance.

    Read this article first..

    > precision has also changed, with modern Pascal GPUs focusing primarily on FP32 (short of going GP100). The 780 Ti and its GK110 GPU support FP64 on a 1:24 ratio (1x FP64 enabled core, effectively, for every 24x FP32 cores).

    > 780 Ti GPU hosts 2880 CUDA cores with a reference Boost 2.0 clock of 928MHz. Modern datapath optimization, clock-gating, and compression tech (like advancements in color compression) mean modern architectures are vastly more efficient.

    > GTX 780 Ti include a 7Gbps memory clock – a bit slower than most modern devices – with 3GB of GDDR5 memory. This is on a 384-bit memory interface. The TDP of the reference card is 250W, about 70W higher than today’s significantly more powerful GTX 1080.

  • 10 Replies sorted by


    980 Ti = 2816 Cores @ 1000Mhz

    780 Ti = 2880 Cores @ 876Mhz

    the 780 ti is rated for 210 GFLOPS double precision

    the 980 ti is rated for 176 GFLOPS double precision

    the gtx 580 is good for 790 GFLOPS in double precision mode since after the 500 series, nvidia started disabling the dp units in geforce chips and saving them for the quadros/teslas


  • The best value cards for hobbyists who need CUDA are the GTX580 for under 100$, GTX780 Ti 3GB for under 200$ and the GTX 1060 6GB over that.

    You can OC the 780Ti 3G quite well, so thats also a factor!

    Power costs do matter, so its important to understand that I stated "hobbyist" - someone who will render or grade small scenes once a week on average.

    I have put in 2x GTX1060 6GB into a 2010 Mac Pro, as they use a single 6-pin PCIe connector each and with 128GB DDR3L and twin x5690 3.46GHx 6-core (total of 12C 24T), the 2009-2012 Mac Pro is the best value editing station in todays world.

    Just saying!


    Basically, GTX 1060 and GTX 780 Ti 3GB are the best value cards for CUDA today.

  • And GTX 780 Ti 3GB can be OCed to get higher performance - Windows folks should be still using these cards used for anything less than 100$

  • @zcream

    Most fun that many of this data is present in Nvidia GPU topic here.

    Also you make big error trying to push GTX 780 Ti.

    CUDA benchmarks are actually not simply applicable for video editing. And GPU video decoders also mean quite much.

    It is very hard to tell about value now, actually, due to miners.

    But at minimum you need 1050 Ti 4GB, best to use 1070-1080 8GB - but only if you are in Prores or raw land editing.

  • With my projects, both hitfilm and Resolve manage to go through 1080p footage in realtime for preview. AE is utterly hopeless. Renders are the time issue here, at least for me. I'm sitting here waiting for a 2 hour render for a 2 min clip on AE - though with multiple layers and pre-comps. Also, I am using CS6, where AE does not use GPU for render at all.

    I'm simply making a note that with todays GPU prices, GTX780 Ti 3GB is the best value GPU for 200$ - to get something faster, you have to go to GTX1060 at minimum. I dont think 1050Ti is faster.

  • I have the GTX 780 ti, and it has some serious limitations as far as driving a 4K monitor. The other thing is, I've tested it extensively, and most of the time it never goes above 35-40 percent GPU usage. It's a good card, but. realistically, it is a bit dated as well as far as HDMI.

  • @DrDave Not for driving a monitor but specifically for rendering in place of an expensive GTX1060. I bought mine before the prices went up.

    If you can have a very basic card for display, you can use the 780ti instead of the 1060 for Resolve.

  • I understand that totally, when I bought the card I bought the one with a lot of CUDA cores. It was a good card for its time, but I realize now that I didn't need all those cores even with lots of filters, it never runs at max. And if you are doing 4K, you need to drive a big, 4K monitor. Of course, you can work 1 to 1 in a smaller 2K monitor, but it isn't ideal, I think. It's still a decent card.


    Some old tests about gtx 780 6gb and gtx 780 ti. As stated earlier, thus only makes sense with a cheap card driving your 4k monitor and this card as your cuda card.