The FEATFLOW benchmark is a set of test calculations which is very similar to the 1995-DFG benchmark `Flow around a cylinder'. However in contrast to the original one, all input parameters and even the triangulation is fixed. The remaining interesting point is the total CPU cost and how they are distributed to the separate tasks in a CFD code, as mesh generation, assembling of matrices and right hand sides, solving linear subproblems or postprocessing steps.

The original idea behind this benchmark was to have a tool at our institute which helps us in selecting the "optimal" workstations for our classes of problems. Being an university institute, no money is available for supercomputers. However, these might be even useless since most of our working time is spent with developing and testing methods and codes. Modern workstations are much more appropriate for these kind of tasks since, for even less money than needed for one single supercomputer, every member of the institute may get his own workstation.

Furthermore, the new generations of processors in workstation are very fast, even for complex data structures. In contrast, these data structures which are necessary for complex CFD applications often behave in such a way that they lead to performance problems on vector computers.

Moreover, parallel machines require additional work in the design of algorithms and implementations which is in general a very hard job. Hence, modern powerful workstations with 1 - 2 GByte RAM are the typical workhorses for numerical analysts, and this at relatively small cost.