Because of high quality cloud infrastructure, the era of high performance computing has arrived to mainstream applications. With 12+ core x86 processors as well as GPU and ASIC arrays, rack densities can easily reach or exceed 30kW per rack. That requires massive amounts of power and cooling in a very small space. Just four racks at that density equals as much energy as one lightning strike hitting our building every hour, every day of the year, all harnessed within 32 sq. ft. It is possible to cool this density efficiently with the right design, but not in a traditional raised floor data center space. HPC requires some custom method of cooling. Also, HPC often has unusual power requirements. In a computational array it can be very wasteful to provide Tier IV N+N power to all components. Providing Tier IV to the file system while limiting the compute cores to Tier I can dramatically decrease cost of implementation without hampering the bottom line. Likewise, since compute cores loads are highly variable, the entire system needs to be designed to scale gracefully with the load without over pressurizing.
To take on these challenges there are many custom solutions including: