I am going to make use of this on my next HPC most likely... with CXL I can finally ditch the very expensive high memory nodes :D at SOME cost of latency...

But looks like the MPI market is going to evolve into something else :D