Processor Research Team (Taiji)

The performance of parallel processing supercomputers can be raised by increasing the number of compute nodes employed. There are two broad approaches used to raise performance: the “weak scaling” approach employs more nodes to deal with a larger portion of the computation task, while maintaining the same computation time; the “strong scaling” approach aims to attain faster computation for the same sized task by increasing the number of nodes used. The latter is more difficult to accomplish. In order to realize strong scaling performance, we seek to fine-tune application programs and improve the efficiency of a processor, a major component of the compute node, with the aim of achieving full utilization of the K computer and to set the course for a next-generation supercomputer.