GTC 2020 S21690
Presenters: Yi Chen,Altair Engineering
Abstract
We’ll talk about how to design proper algorithms and data structure to migrate parallel, sparse, and iterative linear solvers in a commercial CFD code that was previously optimized for CPUs to NVIDIA GPUs with CUDA libraries. Efficient linear solvers on GPUs allow for tighter tolerance of convergence of the iterative process with considerably less cost than that on CPUs, which leads to faster convergence of the non-linear Newton’s iterations, and thus much less wall time.
Watch this session
Join in the conversation below.