Interesting, looks like #ibm 's approach to the #cache and #sharedmemory problem is just to do away with L3 and L4: https://www.anandtech.com/show/16924/did.. Will definitely be watching this space.