System-Level Design of Energy-Proportional Many-Core Servers for Exascale Computing

Abstract

Continuous advances in manufacturing technologies are enabling the development of more powerful and compact high-performance computing (HPC) servers made of many-core processing architectures. However, this soaring demand for computing power in the last years has grown faster than emiconductor technology evolution can sustain, and has produced as collateral undesirable effect a surge in power consumption and heat density in these new HPC servers, which result on significant performance degradation. In this keynote, I advocate to completely revise the current HPC server architectures. In particular, inspired by the mammalian brain, I propose to design a disruptive three-dimensional (3D) computing server architecture that overcomes the prevailing worst-case power and cooling provisioning paradigm for servers. This new 3D server design champions a new system-level thermal modeling, which can be used by novel proactive energy controllers for detailed heat and energy management in many-core HPC servers, thanks to micro-scale liquid cooling. Then, I will show the impact of new near-threshold computing architectures on server design, and how we can integrate new on-chip microfluidic fuel cell networks to enable energy-scalability in future generations of many-core HPC servers targeting Exascale computing.