Introduction

ANSYS has over the last few releases improved the GPU Solver for ANSYS Fluent. HPCBOX has been updated to make it seamless for end users to effectively use ANSYS Fluent on CUDA nodes with minimal effort.

Executing ANSYS Fluent GPU solver

HPCBOX delivers a fully interactive and integrated turn-key Linux desktop environment for ANSYS users to effectively use heterogeneous HPC clusters which have both high-speed network CPU nodes and now introduces support to also handle CUDA nodes with multiple NVIDIA GPUs for ANSYS Fluent.

A new GPU setting is introduced in HPCBOX 3.0.1, and, with just one checkbox, users can automatically start their ANSYS Fluent simulation on multiple nodes with multiple GPUs.

The GPU checkbox in the screenshot above can seamlessly use CUDA GPU nodes for starting ANSYS Fluent.

Scheduling methodology

HPCBOX provides exclusive node exclusive GPU access allowing users to fully utilize all the GPUs on the workers which are allocated to the Fluent simulation run. In addition, using the Processes Per Node (PPN) setting makes it easy to distribute the job across multiple nodes and use multiple GPUs on them with minimal effort. Fluent is always started in fully interactive mode with its GUI making it possible to work on an HPC cluster just as one would work on their local workstations.

In the above screenshots you can see ANSYS Fluent running on two CUDA nodes with two NVIDIA Tesla V100 GPUs each. This simulation was started using the “GPU” Checkbox in the HPCBOX Cluster App’s ANSYS Fluent panel.