Loading patent…
OwnerAdrien Loseille, Michael Thomas Mara +2inventors
InventorsAdrien Loseille, Michael Thomas Mara, Dries Kimpe, Juan José Alonso
Estimating the term…
A technique manages graphics processing units (GPUs), as well as associated algorithm advancements, to improve and effectively hide computational costs of adaptive mesh refinement (AMR). A cloud-based framework dynamically utilizes a distributed pool of the GPUs to parallelize computations of physical simulation software (solver) partitioned across multiple GPUs. A highly parallelized and distributed architecture of the framework provides a hybrid approach that operates partly on GPUs, e.g., to perform the solver computations, and partly on general-purpose processing units (CPUs) to accelerate portions of AMR and mesh processing of an AMR module. In an illustrative embodiment, the technique leverages the GPUs to perform as much of the AMR module processing (e.g., algorithms that rely on fixed topology) that can be naturally and efficiently implemented on the GPUs, while maintaining those portions of processing that do not naturally map onto the GPUs (e.g., algorithms with dynamic topology changes) on CPUs.
A non-transitory computer readable medium including program instructions for execution on hardware resources, the program instructions configured to:
A method comprising:
A system comprising:
Dependencies are inferred from the claim text.