For High Performance Computing, like Multi-physics FEM Simulations, CFD (Computational Fluid Dynamics), etc. even the fastest quad core processor can take hundreds of hours to do a decent simulation. And just one simulation is a single step in a real-world simulation project employing these tools.

The competitive solution is to setup a computer cluster. This is simply a group of computers that work closely and share the work load. Simply put, the speed is (approximately) N times faster with N nuclei working together. With 10/100M Ethernet, GigaBit Ethernet and the faster Myrinet, high packs of processed data can be transferred without impairing the overall performance.

Mini-ITX motherboards can mount the fastest Intel i7 processors and easily put together in standard 19" racks. Ethernet routers and power supplies units come also in 19" rack units, so the mechanical assembly is not only clean and aesthetic, but physically robust to vibrations and interference. Cooling down is also more effective than with a desktop PC even with the higher density of CPUs.

The Operating System most used in computer clusters is by far Linux, with all the others (Windows, Solaris, etc.) not so popular. If developing software for parallel computing, specialized compilers are available like Microsoft Parallel Studio.