NVIDIA (NVDA) announced that its Tesla GPU accelerators have helped the new Eurora supercomputer set a new record for data center energy efficiency. The supercomputer was deployed this week at the Cineca facility in Bologna, Italy and reached 3,150 megaflops per watt of sustained performance - a mark 26 percent higher than the top system on the most recent Green500 list of the world's most efficient supercomputers.
Built by Eurotech, Eurora set the efficiency benchmark by combining 128 high-performance, energy-efficient NVIDIA Tesla K20 accelerators with the Eurotech Aurora Tigon supercomputer, featuring innovative Aurora Hot Water Cooling technology, which uses direct hot water cooling on all electronic and electrical components of the HPC system. It packs 64 compute nodes with 2 Intel Xeon E5 processors and 2 Tesla K20 GPU accelerators into a single rack. It delivers 350 teraflops of peak performance, which would rank it among the top 100 supercomputers in the world.
Enabling scientists to advance research and discovery across a range of disciplines, the Eurora is available to members of the Partnership for Advanced Computing in Europe (PRACE). "Advanced computer simulations that enable scientists to discover new phenomena and test hypotheses require massive amounts of performance, which can consume a lot of power," said Sanzio Bassini, director of HPC department at Cineca. "Equipped with the ultra-efficient Aurora system and NVIDIA GPU accelerators, Eurora will give European researchers the computing muscle to study all types of physical and biological systems, while allowing us to keep data center power consumption and costs in check."
Euora's efficiency is derived partially from the Tesla K20 accelerators, and partially from the use of warm water to keep server components at optimal operating temperatures. Heated water produced from the supercomputer can be re-purposed to heat buildings or drive absorption chillers, and then returned back to the supercomputer at a cooler temperature. Over its five year lifespan, Eurotech estimates that Cineca will save a whopping 2.5 million kilowatt-hours, or $500k savings in energy costs, while eliminating over 1,500 tons of CO2 emissions as compared to a typical CPU cluster of similar performance.
"GPU accelerators are inherently more energy efficient than CPUs, and Tesla K20 accelerators widen this gap considerably," said Sumit Gupta, general manger of the Tesla accelerated computing business at NVIDIA. "Energy efficiency has become the defining element of computing performance. And GPUs enable data center computer systems of all sizes - from small clusters to future exascale-class systems -- to achieve performance goals within an economically feasible energy budget."