Power Efficient Supercomputing

Datacenters house server farms that perform many different jobs ranging from search engines to supercomputing. In aggregate these computer systems already consume enormous power, and their energy requirements are increasing both in absolute terms and in proportion to the total capacity of the United States electrical grid. With higher power consumption has come greater difficulty in keeping these systems cool and running reliably, which threatens to curtail the significant benefits to society they provide. We have developed a novel computer processing system based on certain aspects of human brain computation. The novel system runs embarrassingly parallel software written for standard computer clusters, but provides much higher performance for the same amount of power or, alternatively, consumes much lower power for the same performance. The work conducted in Phase I developed both the hardware implementation and the programming tools to better test the performance and performance-per-watt of the system at standard tasks. Benchmarks in six categories were run and demonstrated that the processor is able to achieve power efficiency improvements ranging from ~10x to ~160x (1000% - 16000%), averaging about 80x. Testing was conducted in Field Programmable Gate Array hardware and cycle-accurate simulation. The Phase II project further develops the programming and simulation tools to allow new users to test their software on the novel system prior to its manufacturing in Phase III. 92 benchmarks have been selected by potential Phase III financers and will be run to provide product performance information helpful for arriving at a procurement decision or other Phase III financing decision. Many benefits to society are anticipated after concluding Phase III. For example, pharmaceutical companies are anticipated to be able to leverage larger scale supercomputers to pre-test more candidate medicines and eliminate drug testing dead- ends with more accuracy, resulting in an increased probability that cures for diseases will be found. In the case of an outbreak of a new infectious disease, the ability to find cures more rapidly increases the likelihood a cure will be found before the disease becomes a global epidemic.