HINT: A new way to measure computer performance

The computing community has long faced the problem of scientifically comparing different computers and different algorithms. When architectures, methods, precision, or storage capacity are very different it is difficult or misleading to compare speeds using the ratio of execution times. We present a practical and fair approach that provides mathematically sound comparison of computational performance even when the algorithm, computer, and precision are changed. HINT (Hierarchical INTegration) removes the need for pseudo-work measures such as MFLOPS or MIPS, producing a work measure we call QUIPS (quality improvements per second). It reveals memory bandwidth and memory regimes, and runs on any memory size. The scalability of HINT allows it to compare computing as slow as hand calculation to computing as fast as the largest supercomputers. It ports to every sequential and parallel programming environment with very little effort, permitting fair but low-cost comparison of any architecture capable of digital arithmetic