Programming CUDA in nvlink mode and 2 RTX 2070 Super's

  1. Is this even supported?
  2. If supported, do we have access to all 16 Gb’s as if they’re one memory system?
  3. Is it, by chance, automatically supported by the current drivers, such that we can fully utilize the cores and memory on both cards and write the programs with no extra or special libraries or calls? Doubling the number of cores and memory with no changes in our code at all?
  4. If 3 is not valid, where would we find the programming model to follow?
    The goal, of course, is to cut execution times in half or so.