Abstract
We analyze COSBIs Open Source Mark (OSMark) benchmark [1] in measuring CPU performance in desktop personal computers. We select and focus on selected tests of the benchmark targeting the CPU which we also profile. We run the benchmark on two personal computers with single and dual cores. We then collect performance event counts for the selected tests to characterize the benchmark's workload and we correlate the performance event counts with the CPI. The results reveal that although this benchmark provides a nice and convenient alternative to existing commercial benchmarks [2-7], it falls behind in total instruction coverage and application coverage, and does not stress enough multi-core processors as well as competing benchmarks.