High Power does not equal High Performance Computing
Wednesday July 25, 2018
The number one fastest supercomputer in the world (at the time of this writing) is the Summit supercomputer at Oak Ridge National Laboratory. This computer uses over 2 million cores and almost 3 Petabytes of memory. However, what is more extraordinary about this supercomputer is that it’s also the #5 greenest supercomputer in the world with a power efficiency of 13.889 GFlops per watt.
Indeed, companies like Facebook, Amazon, and Google have found that the finite limitation when it comes to the systems of tomorrow is not compute power, but electrical power. Each have created their own initiatives to find a way to maximize high performance computing needs while avoiding excessive power usage.
In 2010, Google took the approach of becoming their own power company. Google Energy was authorized by the Federal Energy Regulatory Commission to buy and sell electricity like any other power utilizing company. In so doing, Google purchased and built wind farms and solar energy plants. In fact, just in April of this year, Google became 100% renewable energy powered. By owning their own power plants and selling excess power back to the grid, Google was able to eliminate power costs as well as the power ceiling. If their needs continue to climb, they just need to build more power plants. Unfortunately, not every company is big enough to be their own power company.
Facebook took a different approach. Facebook worked with Intel to develop a new server chipset. The Xeon-D family was solely created to address the primary goal of extracting as much compute capacity for the least amount of power consumption. The result is a much denser compute cluster using the same amount of power.At GigeNET, we too have been looking at providing HPC to our customers without ruining the environment. Our complex system clusters including our Public Cloud, Private Clouds, and dedicated server clusters run on the Xeon-D family chipset. By making the intelligent choice of switching to a new server standard, we have greatly increased our customers computing power while also reducing our power draw.