Performance Benchmarks

Below are performance analyses of FLOW-3D/MP v11.2 up to 640 cores for typical applications of the software, namely water & environmental, metal casting, microfluidics, and aerospace, as well as a quintessential CFD benchmark validation of a lid-driven cavity simulation that shows scaling up to 1280 cores.

Hardware Information

Skylake –  Intel® Xeon® Gold 6148 CPU @ 2.40GHz, 40 Cores/Node, 384GB/Node RAM
Broadwell – Intel E5-2600v4 Series CPU  @ 2.40GHz, 28 Cores/Node, 256GB/Node RAM
Interconnect – Intel Omni-Path®

8x Performance Increase
WATER & ENVIRONMENTAL
HYDRAULIC JUMP

8x performance increase hydraulics HPC benchmark

In this simulation, the hydraulic jump and the overall flow over a spillway were studied.
Mesh: 14.6 million cells
Physical Models: Free-surface tracking, gravity, air entrainment, and RNG turbulence model with dynamically computed maximum turbulent mixing length
Numerical Models: GMRES

4x Performance Increase
METAL CASTING
ENGINE BLOCK

4x performance increase metal casting HPC benchmark

In this simulation, the gravity casting of an engine block was studied.
Mesh: 3.6 million cells
Physical models: Free-surface tracking, gravity, heat transfer, solidification, and viscous laminar flow
Numerical Models: GMRES

8x Performance Increase
MICROFLUIDICS
INK DROP IN PRINTER NOZZLE

8x performance increase microfluidics HPC benchmark

In this simulation, the formation and discharge of an ink drop in a printer nozzle was studied.
Mesh: 2.0 million cells
Physical models: Free-surface tracking, laminar viscosity, and surface tension
Numerical Models: GMRES

5x Performance Increase
AEROSPACE
AIRCRAFT FUEL TANK SLOSHING

5x performance increase aerospace HPC benchmark

In this simulation, the fuel sloshing in an F-16 aircraft fuel tank at various flight conditions was studied.
Mesh: 0.7 million cells
Physical models: Free-surface tracking, non-inertial reference frame, gravity, electric potential, and RNG turbulence model with dynamically computed maximum turbulent mixing length
Numerical Models: Implicit advection, GMRES, and split Lagrangian VOF

40x Performance Increase
BEST CASE SCENARIO
LID DRIVEN CAVITY

40x performance increase lid-driven HPC benchmark

The standard lid driven cavity problem was simulated to demonstrate the scaling potential of FLOW-3D/MP. This is a fully-filled, perfectly load-balanced simulation that is often used to validate standard CFD codes.
Mesh: 10.0 million cells
Physical models: Viscosity and RNG turbulence.
Numerical Models: GMRES