The world of high-performance computing (HPC) traditionally has been dominated by large engineering-intensive industries such as oil and gas and academic research institutions such as national science laboratories. These systems are energy intensive, but are space and process efficient, requiring fewer kilowatt-hours and less square footage to handle the same computational load as a traditional system. HPC systems typically cost tens or even hundreds of millions of dollars. Traditionally, financial institutions running sophisticated automated trading platforms were the only enterprise users with the capital to take advantage of their benefits.

In recent years, blade servers and mainframe performance has increased dramatically, allowing HPC clusters to become a more realistic option for enterprise data centers. In NYC, the HPC market has become considerably more robust due to the city’s high cost of electricity and even higher cost of real estate. As a result, managed service providers have expanded their offerings to include build-to-suit clusters, hybrid cloud and infrastructure to support increasingly denser server cabinets. With HPC becoming more mainstream in enterprise facilities, it is critically important that data center managers and IT directors consider the energy consumption and total cost of ownership for an HPC deployment.

Willdan recently conducted a study of a 20,000-core HPC cluster deployment, which found that careful planning can yield significant energy savings over a more traditional deployment of the same capabilities. The cluster originally was planned to be constructed with ‘conventional’ servers, which would have strained the existing infrastructure and compromised redundancy. The team instead focused on blade chassis — compact, powerful servers with higher core counts — which could be deployed in much smaller numbers. With higher densities in fewer cabinets, the number of servers could be reduced by over three-quarters, significantly reducing the space requirements of the data centers. The blade servers also brought dramatically higher performance per watt, with energy savings of over sixty percent from the standard alternative. For this installation, the high-density solution only carried a ten percent increase in hardware cost, which was quickly recovered on energy savings alone. The data center saved hundreds of thousands of dollars in utility charges in a single year — all without sacrificing performance. Additional savings were realized by avoiding the labor and materials required to install more physical servers, reducing the demand on support infrastructure and eliminating the need to procure new power and cooling equipment to support the larger power demand of a less efficient HPC solution.

As more and more companies look for HPC solutions to serve the needs of their enterprise data centers, the true cost of different configurations needs to be explored. No longer is the sheer performance of a cluster the only benchmark to be compared. In urban areas such as NYC, where the cost of electricity is high and there is a premium on white space real estate, highly efficient, high-density solutions can pay significant dividends in enterprise data centers.