After 18 hours and 156,314 cores later, Amazon Cloud has hit a theoretical peak of 1.21 petaflops. Overall 8 Amazon data centers were involved including Virginia, Oregon, Northern California, Ireland, Singapore, Tokyo, Sydney and Sao Paulo. Why was this cluster being run and what needed this much computing power? The task of designing materials that may be suited to converting sunlight into solar energy. However an important distinction here is that even though the theoretical peak speed hit 1.21 petaflops, the Linpack benchmark(Top 500 performance benchmark) is disappointing. In order for a high Linpack benchmark, clusters of machines need to be physically close to reduce latency. Amazon fails spectacularly in this regard with its data centers spread across the globe. Amazon’s cloud currently has a real world peak of 240.1 teraflops rated at 127 on the top 500 list due to its hilariously high latency.
UM Honors 352: Cyberscience, Fall 2013 
I wonder if Amazon could implement a feature for high priority, high paying customers to be able to choose the data centers at which their computations are being done. This would allow Amazon to make a pretty penny and also offer extra processing power to the customers who really need it by allowing them to reduce latency by choosing the data centers closest to them.