Not all computing systems are suited to the same tasks. But which tasks match with which architectures?
As a masters student at Ohio State University, Saniyah Bokhari set out to answer that exact question for her masters thesis.
Bokhari compared the conventional central processing unit architecture of the IBM 1350 Glenn Cluster at the Ohio Supercomputer Center, the novel general-purpose graphic processing unit architecture (available on the same cluster), and the multithreaded architecture of a Cray Extreme Multithreading (XMT) supercomputer at the Pacific Northwest National Laboratory’s Center for Adaptive Supercomputing Software.
To test the systems, Bokhari ran a subset-sum problem: an algorithm with known solutions that is solvable over a period of time proportional to the number of items included in the computation multiplied by the sum of their sizes. She also carefully timed the code runs for a range of problem sizes.
According to the OSC website, Bokhari found that the NVIDIA GPGPUs she tested were best-suited to small problem sizes, the IBM x3755 performed well for medium sizes, and the Cray XMT was the clear choice for large problems.
Want to learn more? Check out the Ohio Supercomputer Center's release on Bokhari's research. Or, if you want all the technical details, you can find the version of Bokhari's paper that was recently published in the journal, Concurrency and Computation: "Parallel solution of the subset-sum problem: an empirical study."