After studying the case sizes, segregated and coupled solver methods, associated costs, and equivalent CPU cores,
the charts below was created to distill these dierent variables into one cohesive message focused on the two
variables simulation engineers care most about: a GPU's equivalent CPU cores and the associated hardware cost
for that configuration.
In the charts, each GPU configuration has two metrics: its equivalent CPU cores and the estimated hardware cost.
The equivalent speed in CPU cores shown are a range because, as previously discussed, their "eectiveness"
depends largely on the size of the simulation model (element count), the type of solver used (coupled or segregated),
and the physics being simulated. The minimum and maximum equivalent core values from each of the studies (model
size and solver method size) were used to create an all-inclusive performance range a simulation user could reasonably
expect when utilizing that GPU in CFD simulations.
Study Conclusions
©2024 Rand Simulation 10
The Impact of GPU Solving on Simulation Performance
