VRAM (Video Random-Access Memory), the dedicated
memory a GPU has onboard, is a key factor in determining
what GPU is best for the given simulation task. For Ansys
Fluent applications, a GPU's VRAM directly impacts the
maximum simulation model size (number of elements) it
can process. This varies based on physics being simulated
as well as the solving method utilized (segregated or
coupled).
During the study, it was observed that coupled solves
required anywhere between four to eight GB VRAM for
every one million elements in the model. Conversely,
segregated solves appeared to require somewhere
between one and two GB VRAM for every one million
elements in the model. While not a holistic view of all
GPUs tested, the charts below show the best performing
GPUs' associated maximum model size (cell count) and
required HPC Pack count:
GPU VRAM Impacts
©2024 Rand Simulation 9
The Impact of GPU Solving on Simulation Performance
