By Cleve Moler, MathWorks
The LINPACK benchmark, which involves solving a 100-by-100 linear system of equations, was once the industry standard benchmark for measuring performance in scientific computation.
In this article, originally published in 1994, Cleve Moler discusses LINPACK and five other benchmarks used to gauge computational performance and megaflops: combining for loops and matrix allocation using zeros, completing the triangular factorization of a 167-by-167 matrix, solving a sparse system of linear equations, measuring the speed of polygonal fill graphics, and measuring the speed of line drawing graphics. Benchmark execution times for a variety of computers available in 1994—including a PC with an Intel® 486DX2 chip, running at 66 MHz—are included.
Published 1994 - 92029v00