It seems like about once a year there is another breakthrough story about graphene and its potential for high performance computing. A recent paper describes a different transistor construction approach improves the performance-to-leakage ratio (Ion/Ioff) dramatically. This is crucial to lower power and reliable, higher performance. This article in NewScientist U-bend breakthrough for superfast graphene transistors, gives a bit more background for those interested.
New materials and approaches are going to be needed if we’re going to continue to have higher performance computers and not just continue to have more cores applied to problems. Some types of problems can be parallelized effectively, but not all.
On the other hand for those problems that can be broken into parallel execution, the advent of the 16 core Atom based processor may be the foundation for the next generation of data center hardware. I was part of team that took advantage of a structure similar to this in the mid-90s. We front-ended the I/O bound portion of a large ERP system with low cost, multi-core devices, freeing up the backend to do the heavy lifting. It was cheaper, more efficient and scalable. This may fly in the face of some organizations approach to IaaS where all VMs are targeted at the same type hardware but necessary as organizations expect different characteristics.