After attending an Exascale presentation at this year's Linux Foundation's Collaboration Summit, an interest was inspired in establishing an understanding of the challenges in producing an ExaFLOP cluster. To gain some perspective, the first question was what happened in the growth from TeraFLOP to PetaFLOP clusters. Looking back should help in looking forward and gives a context for contrasting and comparing.
The type of nodes expected to be part of an ExaScale cluster are nodes built in the future from the standard processors of that day. The node's performance capability will be extrapolated based on Moore's Law. It also assumes that no extreme measures will be undertaken to modify the HPC applications to benefit from the performance. The exception to this is the need to support a larger number of CPUs and nodes.
|Author||F. Michel Brown|
|White Paper||The Road to Exascale|