There are a number of different tests or benchmarks used to measure the performance of a computer. One common benchmark is the "LINPACK benchmark", which measures floating point performance. This benchmark solves a 100'th order dense system of linear equations. Here are some numbers based on the LINPACK benchmark (in millions of floating point operations per second, Mflops) that show the relative performance of Athena workstations: DEC 5000/25 ~2.5 Mflops Sun SparcClassic ~3.3 Mflops Sun Sparc 5 ~7.2 Mflops Sun Sparc 4 ~9.5 Mflops SGI Indy ~12.8 MFlops If you would like to see more detailed information, the numbers above were generated with the "bench35" command in Matlab. However... BENCHMARKS CAN BE MISLEADING! In order to asess any set of benchmarks, you have to know what is being measured, and how it is being measured. Consider the following results obtained running an example from "Optimization and Tuning Guide for the XL FORTRAN and XL C Compilers", IBM AIX version 3.2: Machine Compile Command Time MFlops DEC 5000/25 f77 18.6 sec 8.57 DEC 5000/25 f77 -O3 18.6 sec 8.57 Sun SparcClassic f77 61 sec 2.67 Sun SparcClassic f77 -O3 30 sec 5.33 Sun Sparc5 f77 21 sec 7.62 Sun Sparc5 f77 -O3 7 sec 22.86 Sun Sparc4 f77 16 sec 10.00 Sun Sparc4 f77 -O3 5 sec 32.00 SGI Indy f77 7.9 sec 20.23 SGI Indy f77 -O3 -mips2 3.6 sec 44.20 As you can see there is large difference for this particular code between the DEC, Sun, and SGI workstations and that sometimes optimization matters greatly. The source code used to get this times is located in the "fortran" locker: IBM RS6000 /mit/fortran/src/benchmarks/Flop.f DEC 5000 /mit/fortran/src/benchmarks/Flopunix.f Last updated: 7/7/97