Comments by "" (@Gameboygenius) on "How Nvidia Won AI" video.

  1.  @dhvanitdesai1044  FPGAs are great if you need the full reconfigurability. However, this comes at the price of a speed and power use overhead for each logic gate. Thus if your work comes down to repeatedly performing tensor multiplication at large scale (as in the case of a typical neural network) a more or less specialized chip is preferable.
    2