4 ms·

Can you explain how this is done? Does this mean that VM and the GPUs can be on different host machines? Will it affect the GPU-CPU communication performance?

• detaro 10y ago
  Sounds like if you max the GPUs per CPU core they have CPU cores "left over" that they then use for CPU-only instances, instead of splitting a host to a predefined pattern of just GPU instances that use up all the CPU as well.
  • boulos 10y ago
    Yep!