A Project-based HPC Course for Single-box Computers

CARLOS S. BEDERIÁN; NICOLÁS WOLOVICK

Salt Lake City

Workshop; EduHPC-16: Workshop on Education for High-Performance Computing; 2016

Throughout three iterations and six years we havedeveloped a project-based course in HPC for single-box computerstailored to science students in general. The course isbased on strong premises: showing that assembly is what actuallyruns on machines, dividing parallelism in three dimensions (ILP,DLP, TLP), and using them incrementally in a single numericalsimulation throughout the course working in interdisciplinarypairs (CS, non-CS). The final goal is to explore how to useall the available transistors in a die. Assembly proved a greattool to show how bare-metal works, an alternative-semanticsapproach to programs, and a tool to demystify compiler technology.Parallelism is tackled gradually with a clear division intoinstruction, data, and thread parallelism. GPUs, through CUDAin particular, are used as a radically different approach to thethree dimensions of parallelism. Each dimension is explored ina gradual manner, starting from a sequential toy-yet-interestingnumerical simulation. After using each form of parallelism andsubmitting a short report, the experiences are put togetherin group discussion unveiling the strengths and weaknesses ofeach form of parallelism for each class of numerical simulation.Although there is a high variance in the students? background, CSand non-CS students pair well in project development, generatingunderstanding and value of the disciplines. The experience provedsuccessful, with former students producing parallel acceleratedcode of their own in their disciplines.