In many simulation systems, dead reckoning is used to minimize network bandwidth utilization. The Distributed Interactive Simulation (DIS) standard is one example protocol that uses dead reckoning. Many game engines also use the technique. Until a few years ago graphics hardware used a fixed pipeline. In recent years PC video cards have been built with a programmable architecture. Collectively, the programmable pipeline is referred to as the Graphics Processing Unit (GPU). As GPU programming has progressed, a growing research field into applying non-graphical algorithms onto the GPU has started. Image processing, numerical equations and illumination computation are some examples of what is called General Purpose GPU programming.
We performed a computational study of dead reckoning comparing the GPU with the Central Processing Unit (CPU). We tested various quantities of simulated entities using a variety of CPUs and GPUs. GPUs have the possibility of dead reckoning millions of entities in a single pass, but suffer the requirement of data readback from the video card, which is often slower than "outbound" data transfer. The study is presented and then analysis of the results discussed.