Supercharging Cloud-Based Computational Fluid Dynamics with the NVIDIA Ampere Architecture
, Altair
, Google LLC
Learn how the new NVIDIA A100 GPUs and latest-generation NVLink and NVSwitch supercharge cloud-based computational fluid dynamics (CFD) with Altair ultraFluidX on Google Cloud’s A2 VM family to enable workflows that were impossible before. For decades, automotive OEMs working on aerodynamic design faced the limited availability and high costs of purely CPU-based systems when performing highly resolved transient CFD simulations. In recent years, the situation has changed with the advent of CFD solvers that are designed to leverage the massive power of GPUs, which allowed for simulations of fully detailed production-level passenger vehicles overnight. Now, using our low-level CUDA C++ implementation of the Lattice Boltzmann Method that uses CUDA-aware Open MPI for GPU-to-GPU communication, we will demonstrate that such simulations can even be performed on a single node during working hours, facilitating completely new possibilities for simulation-based design in time-critical situations.