Multi-chip technologies to unleash computing performance gains over the next decade

Abstract

Datacenter and high-performance computing capabilities have continued their exponential improvements in performance over the prior decade, driven by the proliferation of devices and data through the internet of things (IoT), and new applications in the enterprise and cloud. This trend will continue over the next decade as the demand for compute performance continues to grow with exabytes of data being created daily and new use models incorporating machine learning and artificial intelligence become more prevalent. As Moore's Law has slowed in recent years, numerous techniques including system, architectural and software innovation have been used to extend the high-performance processor performance improvements. We examine these techniques and demonstrate that although some of these will continue, new innovations are needed especially at the system level to continue the performance trend over the next decade. We believe that multi-chip technologies and system level innovations are key to unlocking the performance gains in computing over the next decade.