Supercomputing as a commodity, available on Amazon
They really do have everything!
- By Henry Kenyon
- Dec 07, 2010
Amazon.com is where you go to buy books, DVDs, clothes, video games ... and supercomputing as a service?
Yes indeed, Amazon recently announced the launch of a high-performance number-crunching capability attached to its existing cloud computing capability. Supercomputing has traditionally been limited to specially-built machines at government or academic facilities, and time and access was limited.
The new Amazon Elastic Compute Cloud uses clusters of Nvidia’s Tesla M2050 GPUs — the same high-speed graphics processors used in some Chinese supercomputers — to provide the massive parallel processing muscle for a variety of tasks such as graphics rendering, modeling, simulations and other computational heavy lifting. It’s also very affordable, at $2.10 an hour for teraflop-level processing.
According to Nvidia, the GPUs open up affordable supercomputing for businesses of all sizes. In a statement, Nvidia noted that hundreds of applications have been ported to the company’s CUDA massively parallel architecture on which the firm’s current crop of GPU’s are based. CUDA-accelerated applications benefit from the ability to access hundreds of cores in a GPU to run much faster than conventional processors. Some of the applications ready for use on the Amazon Cluster GPU instances include mental images’ RealityServer for interactive photO-realistic rendering and MathWorks’ Matlab technical computing software.
According to Venturebeat, the alliance with Amazon Web Services is important for Nvidia because of Amazon’s large e-commerce computing infrastructure that farms out computing capabilities to other companies via the Elastic Compute Cloud. The supercomputing instances will allow organizations from startups to cost-conscious government agencies to run a variety of high-performance applications.