help question


I saw that nvidia jetson nano recently came out, I would like to have some questions are the following;

  1. On Google I saw that some users created super computer clusters with nvidia jetson.
  2. After creating the nvidia jetson clusters will it have the graphical calculation power ?.
    I saw the hardware requirements of nvidia jetson nano has 128 CUDA cores and 472 GFLOPs.
    If for example I buy 50 of nivida jetson nano nor do I have 23.600 in what result GFLOPs or TERAFLOPs of graphic calculation ?.
  3. If I buy a mini computer with Intel processors with Windows 10 operating system, I have a notebook laptop; eg milk panda alpha can I use the graphical calculation power of the 50 jetson nano envy cards or should I take Raspberry P3 model B + or does it work with linux this graphical computational power ?.
    I would like these clarifications
    Thank you and I await answers.

Hi ricky89, each Jetson Nano is capable of up to 472 GFLOPS FP16 or 236 GFLOPS FP32. The total number of FLOPS in your system would be the number of nodes multiplied by this. So if you had 50 nodes, 50 * 472 = 23,600 GFLOPS FP16.

Yes, there are many packages that you can install from the Ubuntu apt repository, you can search through them here. Others you can build from source to run on ARM aarch64 architecture (what the Jetson runs).

Thanks of the response

if I buy a mini pc with windows I can’t use all the power calculation of all the 50 nodes nvidia jeston nano?
this result has 23,600 in which video card is the GTX 1080 Ti compatible?

If your mini-PC is on the same network as your Jetson Nano’s, then you could use your mini-PC as the “head node” for command and control of the individual Jetsons. You could also utilize a software multiprocessing framework like OpenMPI for distributing the processes across the Jetsons.

FYI, any software for that situation must be specifically built for it. There won’t be any automated use of the cluster…it isn’t like putting the NVLINK between a pair of SLI video cards.