XXXIII Jornadas de Paralelismo
The Tyranny of Benchmarks: Past, Present and Future Challenges in High Performance Computing
Fecha: Jueves 20 de Septiembre de 2012 / Hora : 09:00 a 10:00
Lugar: Aula Magna (Edificio Altabix)
High performance computing (HPC) is an expanding field with a challenge-filled history. This webinar will briefly examine the history of HPC starting at the move from vector machines to massively parallel processor (MPP) systems. We will also explore current and future challenges for HPC systems and software looking toward the next decade. In particular, HPC has moved from a realm of ever increasing single thread performance to the more challenging realm of rapidly increasing concurrency as increases in processor clock frequencies has largely ended. Through this all, we will explore the difficulties of measuring the productivity of a machinefor a given set of applications.
Historically, the performance of the largest supercomputers has commonly been compared using the LINPACK benchmark and has been tallied on the TOP500 list. When introduced, LINPACK was representative of the typical application workload, but new applications have reduced its relevance. In a like manner, many microbenchmarks have been introduced to measure specific features of HPC systems, but it is often difficult to quantify their relevance to a given application. As we move forward into the next stage of HPC, the lingering desire to optimize for benchmark performance will continue to cause added challenges in achieving the proper machine balance required for scalable application performance.
(Sandia National Labs – USA)
Scott Hemmert is a Principal Member of Technical Staff at Sandia National Laboratories, where he leads the advance supercomputer interconnect research. He’s also a member of the joint Sandia/Los Alamos National Laboratory Alliance for Computing at Extreme Scale (ACES) design team, which is responsible for defining architectural requirements for and overseeing procurement of the alliance’s advanced technology supercomputers. Hemmert has a PhD in electrical engineering from Brigham Young University.