Computer giant Hewlett-Packard has teamed up with Nvidia to build a server containing up to eight graphics processors designed for the high performance computing market. The two have worked together since 2007 to deliver GPU-powered servers and have found design wins inside some of the world’s top supercomputers, including the one dubbed the “Greenest Production Supercomputer.” The latest HP server will use Nvidia’s brand new Tesla GPU.
The combination of GPUs and CPUs is becoming more common as companies strive to gain processing power without boosting energy consumption. For parallel processing, GPUs, which contains hundreds of cores, can deliver speed without a corresponding jump in energy consumption. However, software and workloads must be optimized for the still rather unique architecture. For a comparison, check out the difference between the greenest supercomputer running Nvidia chips and Jaguar, a machine that actually uses AMD x86 processors, from Nvidia’s blog:
Tsubame 2.0 is an incredible achievement in performance per watt, especially if you compare it to Jaguar, which is listed as the #2 system on the Top 500 list. Both have similar peak petaflop performance, but Tsubame 2.0 does it with 92 percent fewer servers and consumes only 1/7th the power. Take that into consideration and it’s easy to see why GPUs have made such an impact in HPC.
The new machine is the 14th in a line of servers from the two, but its real significance might be an indication of HP’s continuing commitment to innovation on the server side and its relationship with Nvidia. Nvidia plans to make a GPU chip that will use an ARM-based processor known as Project Denver, and a close relationship with HP may mean the company will build a box around the planned chip. Regardless, today’s news is a boon for Nvidia, and high performance, energy-efficient computing in general.