There is a performance-at-all-costs mentality at most of the nation's supercomputing centers that has resulted in significant and growing energy use.  Energy consumption and the resultant heat dissipation are becoming important performance-limiting factors that we believe will eventually come to bear on high-performance computing users. The goal of our research is to design and implement system software that will allow HPC programs to consume less energy (generating less heat) with no more than a modest performance penalty---and to do so without burdening computational scientists.

This talk first discusses the energy consumption and execution time of applications from a standard benchmark suite (NAS) on a power-scalable cluster. It presents results showing that many standard scientific applications executed on a such a cluster can save energy, without a significant increase in time, by reducing the processor-state (i.e., frequency and voltage).  Next, it describes a system that dynamically changes processor state at program phases boundaries. It reduces performance, lowers the processor state, when the CPU is not the bottleneck resource.  Last, it presents a system that transparently detects communication regions at runtime.  It saves energy by reducing the processor state during these regions where the CPU is not on the critical path.