Accelerating Urban Microclimate Simulations with multi-GPU Computing
68% of the population is projected to live in urban areas by 2025 (Blocken, 2015). Simulating urban microclimates in real time, accounting for relationships between surface temperatures and wind patterns, is crucial to smart city planning and urban mobility development.
To produce high-quality real-time simulations, researchers need to balance the speed of calculation with the high computational costs that quality simulations incur. Running on multiple GPUs allows for a meter-scale resolution, but certain challenges to maximizing the efficiency and precision of these calculations without oversimplification remain unsolved.
In his recent work titled “Multi-GPU-based real-time large-eddy simulations for urban microclimate,” Dr. Mingyu Yang of Yonsei University addresses these challenges by integrating advanced numerical methods and the enhanced computational efficiency of a multi-GPU setup for large-scale calculations. The new solver allows for real-time simulations with a 1.25-meter resolution on a 1.526 km2 area.
Based on work accelerated at the 2021 KISTI Hackathon, his presentation delves into the conditions under which real-time micro-urban climate simulations become feasible and discusses the parallelization methods using multi-GPUs that address efficiency and accuracy.
Watch Dr. Yang’s in-depth discussion of the new solver for high-fidelity real-time simulations that is 6.4x faster and 11.7x more efficient than a CPU socket. Learn how OpenACC and CUDA-aware MPI help optimize running simulations on multiple GPUs.
Author
