A company that specialises in building high-performance clusters running on Linux for jobs such as weather modeling and product design, is taking its business into the cloud.

Penguin Computing has launched Penguin on Demand (POD), which is a pay-as-you-go service that provides access to a HPC (high performance computing) cluster built by Penguin and hosted at a data centre in Utah.

The service is aimed at startups such as biotech companies that need HPC capacity but can't afford their own supercomputer, and larger companies whose scientists and engineers need more capacity than their data centres can provide.

The POD service will compete with Amazon Web Services (AWS), which has signed up some biotech customers for its EC2 Elastic Compute Cloud, and with on-demand offerings from IBM and several smaller companies.

Penguin CEO Charles Wuischpard argued that AWS isn't ideal for HPC work because of the performance overhead from using virtualisation, and because customers don't know where the servers they have been allocated are located. If they are far apart, that can cause latency that impairs performance for some HPC jobs.

"In high-performance computing you're sending messages back and forth many times per second, so that latency adds up," said Joshua Bernstein, a Penguin software engineer.

Penguin isn't offering performance guarantees, however, except to say that its optimised Linux clusters will run HPC code faster than EC2. Wuischpard said the company can offer service-level guarantees to customers who require them.

It also hasn't published a price list, saying only that it will be "at or below" Amazon's rates. Amazon's pricing starts at $0.10 (£0.06) per virtual machine image per hour. Penguin will charge per CPU core per hour, and also on a monthly subscription basis.

William Fellows, a UK-based analyst with The 451 Group, said many traditional HPC customers have been experimenting with Amazon and other cloud-based services. "The Penguins of the world need to have an answer for that," he said.

Penguin has a good track record designing specialised HPC clusters, he noted. Its customers include The National Weather Service, Caterpillar and Sandia National Laboratories, as well as Web 2.0 companies such as Digg and Plaxo.