date - May 24, 2026

Optimizing an HPC Powerhouse.

Many professionals, small firms, and even medium-to-large companies are interested in understanding how to best invest in R&D hardware for CAE analysis, especially for open-source software.
Given the high costs of professional solutions, we also had to evaluate how to maximize our investment.
AMD holds a long-standing history of strategic partnerships with open-source software, including detailed collaborations with OpenFOAM. This commitment highlights the company's role in advancing open-source engineering simulations. Read more about AMD's open-source initiatives on their official blog:

cgx models

Leveraging the high performance-to-value ratio of the 7000 series described in that 2020 above article, we configured our HPC server with:


Dual AMD EPYC 7B13 processors on a Supermicro H12DSi-N6 motherboard,
equipped with 256GB of ECC RAM and high-speed NVMe storage.


To fully unlock the potential of the Zen 3 architecture, specific optimizations are mandatory. By utilizing Ubuntu Server 26.04 LTS (Noble Nightingale) in a headless configuration, we ensure the OS footprint is minimal, directing all 128 cores toward the workload.
This setup allows the system to leverage AVX2 instruction sets effectively, which are critical for accelerating the heavy floating-point operations in solvers like OpenFOAM, CalculiX, Elmer Multiphysics, LIGGGHTS, OpenRadioss and many more.
Combined with the latest OpenMPI and OpenBLAS libraries tuned for the 7B13's specific topology, these configurations significantly reduce solving times,


transforming raw hardware into a finely-tuned HPC engine.


Every detail, from the BIOS settings to the latest firmware updates, has been meticulously tuned to squeeze every drop of performance out of the available hardware.

The effectiveness of this optimization is best demonstrated by a robust benchmark: running a modified version of the classic simpleFOAM MotorBike tutorial scaled to 12 million cells, the simulation completed 300 iterations in just 20 minutes utilizing only 64 of the available 128 cores.

cgx bellow.frd

The benchmark was further enhanced by a custom geometry:


a stack of two S-shaped forward-swept winglets acting as a Venturi S-duct.


This complex aerodynamic model was designed by engineer Salvatore Dona using his proprietary aerodynamic surface modeler, FlusurFlusur logo.

This new HPC server represents a strategic leap forward, providing an unprecedented boost to our consultancy services.
By combining this massive computational power with our strictly open-source oriented philosophy, we can now offer all our clients faster, more scalable, and high-fidelity simulation workflows without the burden of proprietary licensing.

If you are interested in the source program file don't exhitate to contact us.



Back to Homepage Back to FEM/FEA Consultancy Back to FEM/FEA Tutorials Back to FoamBuilder Back to Software Development Services Back to Software Development Services foamBuilder Back to Software Development Services DAKOTA interface Back to Software Development Services Post-processing Back to FEM/FEA Courses Back to Who we are Back to Contact Back to Request Back to open resource Back to News