Amazon Web Services is launching compute clusters for high-performance computing applications like genomic research.

The offering, dubbed Cluster Compute Instances for Amazon EC2, is designed for intensive computational workloads like parallel processes used by the likes of Lawrence Berkeley National Laboratory for research.

High-performance computing (HPC) is a competitive space that’s closely watched for jaw-dropping throughput and speeds. With its move into HPC, AWS is looking to capture some of the market typically dominated by the likes of HP, Cray and IBM with their supercomputers. AWS reps say that customers have been asking for instances to run complex computing workloads.

Amazon Web Services’ (AWS) cluster computing move will either be an interesting showpiece for a limited market or end up democratizing supercomputing. In the meantime, Lawrence Berkeley National Labs was a early tester of AWS’ Cluster Compute Instances. AWS’ HPC service features pay-as-you-go pricing and the ability to scale up or down on demand. These cluster computer instances operate the same as Amazon’s Elastic Compute Cloud.

According to AWS, Cluster Compute Instances will run $1.60 per instance hour. A one-year reserved instance will be $4,290 upfront and carry a usage price of 56 cents an hour per instance. A three-year reserved instance will be $6,590 upfront and 56 cents an hour per instance.

The big difference is speed. AWS’ says that applications on its HPC instances can garner 10 times the network throughput of its largest EC2 versions. In a statement, Peter De Santis, general manager of Amazon EC2, said “in our last pre-production test run, we saw an 880 server sub-cluster achieve a network rate of 40.62 TFlop.” AWS says it hasn’t tried to establish a theoretical peak load, a metric that’s watched in the supercomputer rankings, but larger clusters would provide more performance.

Overall, Amazon’s cluster computing efforts fit in with its other large-scale efforts such as public data sets and services to popularize Hadoop, a tool to analyze a lot of data.