Modern high-performace computing systems are usually set up with large numbers of nodes, each with many CPU cores, each node being linked by very fast networks.
Recently Graphics Processing Units (GPUs) have begun to be used within such supercomputers, where large amounts of data can be streamed and processed with relatively simple
instructions very quickly. Alternative current developments in computer hardware work with hardware such as Field-Programmable Gate Arrays (FPGAs) which can be flashed with a custom circuit design to process
streamed data in more complex ways than GPUs.
Encoding these designs used to be the speciality of chip designers, but recent innovations have led to compilers which can translate conventionial style programs into circuit designs for such FPGAs, allowing algorithms to be easily converted directly to hardware, with speedups measured in the hundreds when compared to CPUs.
We are involved in the EXTRA consortium (https://www.extrahpc.eu/) which seeks to inform the next generation of exa-scale computer hardware. We are encoding Quantum Monte Carlo techniques (which are used for high-accuracy calculations of energies of materials) on FPGAs and the feasability to rework these algorithms to take advantage of new computer architectures.
A presentation on this is at:
http://paginas.fe.up.pt/~specs/events/cse2015/index.php?page=specialsess...