IBM/Lenovo can do it too. Custom fabric for affinity of Lambdas and in memory S3 on the same rack for insanely low latency. Killer is an MPI Broadcast primitive

IBM/Lenovo can do it too. Custom fabric for affinity of Lambdas and in memory S3 on the same rack for insanely low latency. Killer is an MPI Broadcast primitive over a tree network and other bare bone equivalents of fundamental MPI operations. DOE alone will stop buying CRAY when they can use commodity low latency racks for HPC.