GPU Computing in Nuclear Engineering Faster speed and higher accuracy are, and will always be pursued by computational scientists and engineers. To achieve these goals, single-core micro-processors were assembled to arrive at “parallel computing”. This parallelism is recently extended to chip-level with the emergence of multi- and many-core architectures—that is, roughly, adding more cores onto a single chip—in both Central Processing Units (CPUs) and Graphics Processing Units (GPUs). The latter is lately realized to be a possibly stronger candidate than the former to solve computationally demanding problems. This is due to the fact that the floating-point computational capability of GPUs becomes generally 10× higher than CPUs. As a result, scientific computing is rapidly moving toward many-core parallelism. Modeling and simulation in nuclear science and engineering relies heavily on computational power. Nuclear engineers took advantage of parallelization to advance the simulations of neutronics, thermal hydraulics, materials, plasma physics, etc. As one of the first research groups introducing GPU computing into nuclear engineering community, we are trying to make use of this advanced parallel computing technique to solve some of the most challenging problems in our field.
NVIDIA Tesla S1070 GPU architecture |
Snapshots showing u and v velocity components for time-dependent, 2D Burgers Equation solved on GPU 
Performance comparison of CPU and GPU implementation of 2D Burgers Equation Publications Kai Huang, Rizwan-uddin, "GPU-Based Parallel Computing: A New Computational Approach and its Applications to Nuclear Engineering", Proc. of ANS Annual Meeting 2009(2009). On CD-ROM
People |