High-Performance Computing (HPC )

Abstract

Computing with nano-chip technologies spans highly different applications. The corresponding processor-chips span a range of 1000:1 in performance and in their energy-efficiency. From 2010 to 2014, their performance per product-unit almost doubled every year, while their energy efficiency doubled only every three years, so that their total power consumption quadrupled in three years. With the evidence of this energy crisis , the projections for 2015–2020 have been corrected somewhat to a global performance improvement of 20-times in 5 years, while the energy efficiency shall improve an optimistic 7-times in 5 years, which would raise the total required wall-plug power another 3-times, an unlikely scenario, accompanied by a “Green” movement. This “green ” strategy lowers the energy per operation by reducing the operating voltage. However, this reduces the operations per second so much that, wherever the completion of operations in a given time is an issue, more computing units would have to be run in parallel, with a worse balance in required hardware and energy. Therefore, GREEN is not enough. A new throughput figure-of-merit sheds new light on the task of simultaneously improving performance and energy-efficiency, demanding disruptive innovations.