Frontiers in Supercomputing at LLNL

Predictive simulation and knowledge discovery are key drivers that will shape the future supercomputing and expand our thinking about this field.  In large-scale simulation, the implications of developing multi-physics /multi-scale computer models and uncertainty quantification are augmenting traditional considerations of performance scaling in the design of supercomputers and software architectures.  To understand massive datasets, we are being driven towards architecture designs that are considerably more I/O intensive in comparison to traditional supercomputers used for simulation. In this talk, I will discuss how the Laboratory is approaching these challenges. In particular, I will review activities and motivation in our development of petascale supercomputers, work in a new software infrastructure to build multi-physics simulations, and new research in data-intensive supercomputing architectures.