Amazon Web Services has introduced a cloud service that delivered high-performance cluster computing as a service.
Launched overnight in Seattle, the Cluster Compute Instances for Amazon EC2 targets businesses and researchers with demanding network applications.
It typically runs as a cluster of up to eight instances, with a 10 Gbps connection linking them. Customers may request to run larger clusters.
Each instance has 23GB of memory, two Intel Xeon X5570 quad-core processors, 1690 GB of instance storage and runs on Linux.
Amazon EC2 general manager Peter De Santis said an 880-node cluster achieved 41.82 TFlops on a LINPACK test.
By contrast, the 145th and 146th most powerful supercomputers on the international Top500.org list have achieved 41.88 TFlops and 41.77 TFlops, respectively.
Amazon has completed beta tests with US customers including the Lawrence Berkeley National Laboratory, MATLAB developer MathWorks and Adapting Computing.
No Australian companies were involved and although it was a worldwide service it was confined to Amazon's North Virginia data centre.
It will be rolled out to Amazon's data centres in California, Ireland and Singapore according to demand, a company spokesman told iTnews.
Each instance is $US1.60 ($A1.82) an hour or $US0.56 an hour with a long-term subscription; these cost $US4290 or $US6590 for one or three years respectively.