While Nvidia's GPU-powered supercomputers aren't new Nvidia has been hard at work keeping the technology fresh and has inked some interesting deals (including one I wrote about with IBM) designed to keep supercomputer-like GPU processing power a relatively easy to add capability in your datacenter. Well, it has just gotten even easier, provided you are willing to experiment with the cloud for your supercomputing needs.
PEER 1 Hosting in the UK has launched a supercomputing cloud service based on the Nvidia Tesla S1070 and M2050 GPU computing systems. We're talking serious computing power here; the S1070 is a 1U rack mount that contains 960 processor cores and four teraflops of computing power.
The primary benefit of this type of cloud service certainly seems to be the ability to expand your computing capabilities in areas that you might not previously have considered. Programming for the Tesla machines and their Fermi processors can be done using the well-documented CUDA programming environment which probably supports the programming language you are already using for internal development, meaning that you have programmers with the skills to get you up and running in the supercomputing world already at hand.
Now granted, the applications of GPU-based supercomputing probably seem somewhat limited, but the broader availability of this technology, in an easy to access fashion, will lead more developers into looking for ways to enable their applications to take advantage of this kind of computing power. And to sweeten the deal, Peer 1 is offering free trials and proof-of-concept access to their cloud, so this might motivate new users to at least look into the benefits of GPU-based supercomputing.
The technology isn't for everyone, but if you have an application or a need for this type of computing power, it is worth taking a look.