High Performance Computing – Increasing Computational Performance with Hardware
The traditional definition of high performance computing focused on achieving supercomputer performance through clustering of computers. With the advent of graphical processing units most notably led by Nvidia and AMD, the high performance computing now stands for using lower-end computers and customizing the code to make use of their graphs processor units.
The IDC reports strong growth in the HPC segment with a shift to higher performance including the petaflop range and the move to fiber for interconnects.
A rich resource of information is the general purpose computation using gpus (graphics processing units). Additional information can be found at GameDev.Net.
Players in this space include Acceleware which makes turn-key solutions with vertical solutions in seismic and imaging applications. Their partners include the usual suspects, Nvidia, AMD, Dell, Sun, and HP.
Rapidmind offers a modular based solution with an elegant API that focuses on single-threaded, multi-core solutions using CPUs and GPUs. They steer clear of the multi-threaded approach due to the complexity of the code. They offer solutions to the desktop and medical imaging and financial analysis markets.
ClearSpeed takes the accelerator approach building custom cpus onto its own proprietary hardware and then optimizing its software to work with it.