As the technology for high-speed networks has incredibly evolved this last decade, the interconnection of workstations at gigabits rates and low prices has become a reality. These clusters, based on regulars workstations (e.g. PCs), can now be used in place of traditional parallel computers with no possible comparison on the prices! In this article, 3 applications (high performance computing, distributed shared memory system and parallel simulation) that were traditionally executed on expensive parallel machines are ported on a Myrinet-based cluster of PCs. The results show that the performances of these new architectures can be very close to those obtained on state-of-the art parallel computers. \textcopyright Springer-Verlag Berlin Heidelberg 1999.