HPC in Europe Taskforce

Large computing facilities come in two flavours: capacity computing and capability computing. Capacity computing typically fills the needs of scientific disciplines which do not need a low-latency, high-bandwidth interconnect architecture between processors. Prices for such capacity facilities continue to drop, and it is not expected that combining these national resources into a few capacity superclusters in Europe will benefit from the economy of scale. Surely there exists a need for large cluster facilities, but these tend to fall well within the financial possibilities of users themselves.

Capability computing, on the other hand, need a European approach. In general capacity computing requires access to many processors in parallel, large memory, and low-latency, high-bandwidth interconnects. The availability of significantly larger capability systems than available now is urgently required for increasing the scientific opportunities for many application areas, since more processors and memory will be available in a low-latency, high-bandwidth interconnect architecture. Hence, the availability of a number of capability systems will facilitate economically important fields of science and research in Europe in their efforts of running increasingly accurate simulations.

Within the class of capability systems, there are still trade-offs to be made. Actual architectures range from purely shared memory vector computers, through clusters of SMP systems (connected by state-of-the-art interconnects, like Infiniband) to NUMA systems, with a large amount of processors with direct access to a single address space. Since specific application areas run more efficiently on specific capability architectures, it may be desirable to represent these types of systems in a European Supercomputer infrastructure.