Why Acceleration?
Across the globe, engineers and scientist discover a plethora of problems they can’t reasonably solve using traditional paradigms, such as building prototypes and then testing them. As a consequence, they revert to simulations and modeling of behaviors in computers, and build the prototypes in a virtual world, test the virtual prototypes, and then refine the models to match the physical world and reality. This saves them time and money, but it requires large amounts of computational performance. Since their models and virtual prototypes must match reality or come as close as possible to it, the required computational performance is mostly floating-point performance.
In the past, just adding more General-Purpose processors was enough to satisfy the needs. However, with data centers approaching their limits in terms of volume, energy consumption and manageability, other and newer solutions are required. Here is where Floating Point Acceleration technologies come into the game.
Dedicated Floating Point Accelerators are coprocessors that do not need additional space other that occupying already existing slots in computers, consume very little additional electric energy, and are easy to manage. As a result, Floating Point Acceleration is one of new areas of technology that is being actively pursued by the architects of practically all Supercomputer Centers across the globe in their quest to provide ever more performance for their users. With energy consumption at today’s major supercomputing sites typically exceeding a megawatt and rising towards five megawatts or more, these same visionaries are exploring innovative ways to reduce the economic and environmental impact of their facilities.