Comparison of the PVP and MPP situation

• In chemistry the user already today has the choice, whether to use a PVP or an MPP. In structural mechanics and CFD there is however still a strong dominance of PVPs.
• The codes in structural mechanics are almost all finite element codes. The effort for the partitioning of unstructured grids is pretty high. On the other side this is primarily a workstation business (except crash). So it seems that software vendors are not really attracted too much by MPPs.
• In CFD things are different, because even everyday-problems are more CPU-intensive. There are more activities going on in this field.
• The Convex SPP-1000 and the IBM SP/ machines can be used for throughput-purposes also. The reason for this is, that they are compatibele with workstations based on the same processor.
• Convex and IBM seem to dominate in the MPP market. On the Convex it is however not always clear, which codes really run in parallel and in which state the codes are, as parallelisation could simply mean to run it through the compiler and use the machine as a shared-memory machine without any optimisation.
• On the SGI shared memory parallel multiprocessors information was not available, which codes run in parallel.