Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Please, support PV!
It allows to keep PV going, with more focus towards AI, but keeping be one of the few truly independent places.
Nvidia GPUs design
  • GTX 1030, ultra budget card

    GP108-300-A1 chip, SMMs - 3, 64bit bus, 48Gb/s bandwidth, 942 GFlops

    image

    GTX 1050 / 1050 Ti, budget cards

    GP107-300(400)-A1, SMMs - 5(6), 128bit bus, 112Gb/s bandwidth, 1733(1981) GFlops

    image

    GeForce GTX 1060 3GB/6GB, medium card

    GP106-300(400)-A1, SMMs - 9(10), 192bit bus, 192Gb/s bandwidth, 3470 (3855 ) GFlops

    image

    GeForce GTX 1070 / GTX 1080, advanced cards

    GP104-200(400)-A1, SMMs - 15(20), 256bit bus, 256 (320)Gb/s bandwidth, 5783 (8228 ) GFlops

    image

    GeForce GTX 1080 Ti, top cards

    GP102-350SMMs - 28, 352bit bus, 484 Gb/s bandwidth, 10609 GFlops

    image

    You can see extreme savings on design, as all cards consist from literally same blocks, just different number of them.

    sample216.jpg
    332 x 334 - 33K
    sample217.jpg
    465 x 914 - 140K
    sample218.jpg
    540 x 694 - 119K
    sample219.jpg
    800 x 671 - 222K
    sample220.jpg
    800 x 458 - 151K
  • 12 Replies sorted by
  • Nvidia encoding

    image

    And decoding features

    image

    Note for high bitrates it is almost no reason to use software encoders

    image

    sample564.jpg
    800 x 316 - 47K
    sample566.jpg
    800 x 297 - 40K
    sample565.jpg
    800 x 290 - 32K
    sample565.jpg
    800 x 290 - 32K
  • GV100 architecture (Volta)

    image

    Single SMM close

    image

    sample454.jpg
    800 x 450 - 110K
    sample455.jpg
    641 x 843 - 187K
  • Nvidia GPU plans for 2018

    image

    sample336.jpg
    542 x 452 - 29K
  • Yup. 1080Ti, however I also have my eyes on Vega. I'm thinking of a few 1080Ti's for rendering- or Vega's.

    Will reassess when the star falls!

  • I was looking more at these sort of benchmarks. A 980ti is much faster than a 1060. However a 1070 is basically on par with a 980ti yet uses less power.

    I see zero reason to get 900 series cards now. Until may be if you get some amazing deal on top ones and use Resolve.

    Simplest thing is to just get 1080 Ti and go work if you use raw/ProRes and heavy grading. Instead of spending time.

    If just editing weddings and similar stuff, 1050 Ti /1060 6GB is enough for years ahead.

  • Nvidia GP106 GPU in Tesla cars computer

    image

    sample278.jpg
    698 x 720 - 109K
  • I was looking more at these sort of benchmarks. A 980ti is much faster than a 1060. However a 1070 is basically on par with a 980ti yet uses less power.

    http://www.phoronix.com/scan.php?page=article&item=14way-nvidia-madmax&num=2

  • With the 1070 being the best $ per CUDA core. I think that the 1080's memory bandwidth is the limiting factor.

    It is not true, see charts with dollar per SMM. Best are 1060 cards for US.

    Also made for you memory bandwidth per SMM chart

    image

    sample266.jpg
    800 x 413 - 99K
  • When looking at benchmarks 1070 & 1080 are the two GPU's that are high performance. With the 1070 being the best $ per CUDA core. I think that the 1080's memory bandwidth is the limiting factor.

  • I've compared GTX 1060 and GTX 1080 only.

    • H264, GTX 1080 encodes 70% faster than GTX 1060
    • HEVC, GTX 1080 encodes 140% faster than GTX 1060

    Interesting discovery - actually GTX 1080 have TWO encoder blocks.

    image

    Also note that outside Quadro they block more than 2 streams.

    sample265.jpg
    800 x 195 - 25K
  • image

    Chart for Russia (got close to minimum prices).

    sample222.jpg
    800 x 404 - 82K
  • Interesting view on prices if you just look how much they charge per SMM block

    image

    Two dips are due to cards with low memory amounts.

    sample221.jpg
    800 x 417 - 90K