©2024 Rand Simulation 6
Study Results: Mixing Tank Model
(~4 Million Elements)
Continued...
The first experiment compares segregated and coupled
solving method performance on the Ansys mixing tank
model, the smaller of the two benchmark simulations. In
the previous charts, each GPU (y-axis) are grouped
based on how many HPC packs are required for that
configuration and lists its equivalent CPU cores (x-axis).
At the end of each bar, the associated hardware costs
are displayed.
When using HPC Packs for Ansys Fluent Simulations, the
benefit of using GPUs instead of CPUs for solving
becomes readily apparent. For CPU cases, 2 HPC Packs
allow parallel processing on 36 CPU cores, 3 HPC Packs
allow 132 CPU cores, and 4 HPC Packs allow 516 cores.
When utilizing GPUs, the equivalent number of HPC Packs
allow vastly more computing power. For instance, 128
CPU cores (the baseline case) require 3 HPC Packs with
an estimated hardware cost of $52,000. For approximately
half the hardware cost and same number of HPC Packs,
an NVIDIA RTX 6000 Ada Generation oers performance
that is equivalent to approximately 250 to 350 CPU cores.
When selecting to utilize segregated or coupled solving
methods, the coupled solving method delivered higher
performance in terms of equivalent CPU cores for the
Mixing Tank Study. The question now becomes, how do
GPUs perform simulating a model that is not only much
larger but also more complex?
The Impact of GPU Solving on Simulation Performance
