While these slides don't provide any new information, they emphasize the benefits of AMD's memory architecture: The more CPUs you add, the more memory bandwidth you gain. It is, however, very important to understand the HyperTransport architecture, to be sure to get the most performance out of it - simply put, add memory to all NUMA nodes. If you have a VM requiring more memory than it can find at its closest NUMA node, you will experience performance penalties according to how far away it finds its memory space. Dedicate enough memory to each CPU for such a VM, or at least make sure it runs at the CPU with the greatest amount of memory. ESX is very NUMA aware, and will try to achieve the best scenario.