EU Regional School - Hammond Seminar - Part 2
Dr. Hammond - Scientific Programming at Petascale and Beyond
Argonne Leadership Computing Facility
Argonne National Laboratory
Computer hardware, particularly for high-performance computing, is becoming more complicated due to the emergence of GPU computing and the growing parallelism even in commodity CPUs. At the same time, state-of-the-art supercomputers possess an ever-increasing number of nodes, more powerful networks and hierarchical filesystems. This growing hardware complexity poses a significant challenge to scientific programmers. I will present case-studies of scientific programming for high-performance computers, describing both the programming techniques and software tools employed, for a variety of architectures. Two examples include the implementation of quantum-chemical many-body theory on GPUs and massive-parallelization of umbrella sampling on Blue Gene/P using a one-sided programming model. These and other examples should provide insight into the path forward for scientific programmers attempt to efficiently utilize high-performance computers at petascale and beyond.