We present a comparative analysis of the maximum performance achieved by theLinpack benchmark on compute intensive hardware publicly available frommultiple cloud providers. We study both performance within a single computenode, and speedup for distributed memory calculations with up to 32 nodes or atleast 512 computing cores. We distinguish between hyper-threaded andnon-hyper-threaded scenarios and estimate the performance per single computingcore. We also compare results with a traditional supercomputing system forreference. Our findings provide a way to rank the cloud providers anddemonstrate the viability of the cloud for high performance computingapplications.
展开▼
