How can GPU acceleration benefit modern UAM/UAV applications? The explosion in machine learning and artificial intelligence research we see today is not due to fundamentally new computer science theories. Indeed, most of the math involved in these algorithms is decades old. The truly revolutionary thing in AI and ML today is the realization that we can accelerate the vast number of computations involved—which were once considered computationally intractable—using graphics processor units (GPUs). GPUs offer large arrays of computational cores. Provided that we write our algorithms in such a way that they can take advantage of the GPU cores to perform our computations, we can offload much of our algorithms to the GPU and leave the CPU free to continue running the system. This is true acceleration. This paper discusses the calculations involved in the use case of modeling the flight dynamics of a quadrotor, discusses which parts of the calculations are candidates for GPU acceleration, and demonstrates the importance of determinism to the execution time of these calculations.

Product developers often conduct trade studies before selecting components for their designs. One of the components that is often considered for these studies is a Graphics Processing Unit (GPU). GPUs have become a rather ubiquitous staple of most any electronic device or computer; however, the parameters that need to be collected to choose which GPU best fits a given application have grown increasingly complex. This paper addresses both discrete and embedded GPUs by comparing performance, cost, development environments, obsolescence, and certifiability. When all GPU variables and parameters are identified and quantified (or “costed”), a truly comprehensive design trade study can then be conducted by product/platform developers.