P38: Co-Managing Performance and Power for High-Performance Scientific Computing Session: Posters: Systems Hardware Event Type: Poster Time: 5:00pm - 7:00pm Speaker(s): Sayaka Akioka, Shannon Kennedy, Konrad Malkowski, Mary Jane Irwin, Padma Raghavan, Lois Curfman McInnes, Boyana Norris Location: 4th Floor South Lobby Abstract: Our project concerns tools that can improve the performance of computational science applications while reducing the energy consumed by the hardware by exploiting low-power modes of processors, caches, memories and interconnects. Such power-aware high-performance computing is becoming increasingly important as microprocessors approach their packaging thermal limits, and the power-related costs of clusters and multiprocessors continue to grow as a quadratic function of peak execution rates and clock frequencies. We focus on sparse or irregular computations which often cannot execute near peak rates of the CPUS's, and there is potential for tuning them to reduce energy without adversely impacting performance. We will provide an overview of our project on co-managing quality, performance, and power trade-offs. Our initial results indicate that careful memory subsystem optimizations and the use of power control modes can significantly decrease energy while improving the performance of sparse matrix vector multiplication and some of the NAS benchmark kernels. Chair/Speaker Details: Sayaka Akioka The Pennsylvania State University Shannon Kennedy The Pennsylvania State University Konrad Malkowski The Pennsylvania State University Mary Jane Irwin The Pennsylvania State University Padma Raghavan The Pennsylvania State University Lois Curfman McInnes Argonne National Laboratory Boyana Norris Argonne National Laboratory