Penguin Computing is going into the cloud business.

Not your ordinary cloud, mind you, an HPC cloud, called Penguin on Demand (POD), an extension of its usual fare, and a first.

Penguin says it should enable access to scalable HPC by a much broader market and promote new and faster innovation.

Fussbudgets might object that POD's not a cloud because the high-end widgetry isn't virtualized. Customers will have their own full dedicated servers for maximum performance and the I/O for massive HPC workloads.

Penguin is targeting researchers, scientists and engineers who need surge capacity for time-critical analyses or can't afford their own HPC cluster. It's also handy for digitizing movies - both the Hollywood kind and the amateur's - as well as running web analytics.

Penguin CEO Charlie Wuischpard also figures that since supercomputers are hard to cover with even a two9s SLA, POD will make a nifty fallback in event of a blowout.

Dr Natalia Trayanova, professor of biomedical engineering at Johns Hopkins and chief scientist at CardioSolv, a company specializing in predicative cardiac modeling and simulation, reportedly got 30x better performance out of POD than Amazon EC2, solving a problem in 30 minutes compared to 18 hours.

And Wuischpard says POD is no more expensive than Amazon's top rung even though they're not comparable.

POD's highly optimized Linux clusters uses high-end Xeon quad cores, with 40GB of memory to a processor and Infiniband - not the kind of stuff Amazon offers. Its software configurations are specifically tuned for HPC.

Penguin describes it as a persistent compute environment that runs on a head node and executes directly on the nodes' physical cores. There's also state-of-the art GPU supercomputing with Nvidia Tesla technology. Jobs typically run over a localized network topology to maximize bandwidth and minimize latency.

Support is included for POD application set-up, environment creation, ongoing maintenance, data exchange services and application tuning. POD also includes persistent storage for local data and user-defined compute environments.