Research Computing Benchmarks

Performance of supercomputers and clusters at UNC Research Computing is one of the major issues that users are interested in knowing more about. To this end, two kinds of benchmarks have been performed on our machines:

The quantities used to measure performance include total execution time and floating-point performance for serial jobs and latency and bandwidth for parallel jobs.

Two more benchmarks have also been implemented. One is an I/O benchmark to measure the performance of machine input and output. The other benchmarking reports machine performance on a daily basis for each of the production Research Computing machines. This is done by running a few preselected/standard performance I/O benchmark codes every day at some preselected time. Results are plotted and published online.