Technical sector guide · Foundation to specialised

HPC Simulation and Engineering AI for Manufacturing

CFD, FEA and design-of-experiments in production, with surrogate-model AI trained from simulation output.

Engineering compute is valuable when it shortens a design loop or increases the number of alternatives that can be evaluated. The fastest individual solve is not always the highest business throughput: a scheduler running many independent cases may finish a design-of-experiments campaign sooner.

Classify the solver and campaign

Determine whether the code is memory-bound, sparse-linear-algebra dominated, license-limited, GPU-enabled or strongly coupled across MPI ranks. Separate one very large transient solve from parameter sweeps, optimisation and post-processing. Record mesh size, cells/elements per rank, memory per rank, checkpoint rate and licence tokens.

Platform choices

X1140-C AxiKeystone handles large single-node CPU/memory jobs, preprocessing and compilation. X2160 AxiVector provides two 128-core nodes and direct 100GbE RDMA for codes that demonstrate MPI scaling. X3140 AxiLattice provides three scheduled nodes for shared CFD/FEA, parameter studies and departmental throughput.

Parallel CFD and FEA

OpenFOAM decomposes a domain for parallel execution; partition quality and inter-rank boundary area directly affect communication. PETSc provides scalable solver components and GPU paths for supported operations. Bind ranks to NUMA locality, test decomposition strategies and preserve a serial or validated reference result.

Do not infer accuracy from speed. Mesh independence, solver tolerance, convergence and physical validation remain engineering decisions.

Surrogate and reduced-order models

Use simulation campaigns to create a governed training set of geometry, boundary conditions and outputs. Train surrogates to screen designs or provide rapid estimates, while retaining a confidence/validity domain and routing critical candidates back to the full solver. Keep model predictions distinguishable from physics-based results in reports.

Production workflow

Submit jobs from versioned templates, stage input to NVMe, checkpoint long runs and return approved outputs to project storage. Scheduler integration should reflect licence tokens as a resource. Use CI to compile and regression-test custom solvers. Capture residual histories, solver versions and environment manifests with results.

Acceptance plan

  1. Choose small, medium and production-size reference cases.
  2. Validate numerical equivalence and one-node performance.
  3. Measure two- and three-node scaling where supported.
  4. Run a representative parameter campaign to test scheduler/storage throughput.
  5. Verify checkpoint restart, node failure procedures and licence controls.

Primary technical references

References are provided for software architecture and implementation planning. Validate the versions, licences, support matrix and regulated-use requirements applicable to the final deployment.

From technical concept to production system

Apply this technology through an Axiotech engineering work package

Axiotech can connect the compute, AI or analytics platform to the machine controls, data contracts, validation evidence and lifecycle-support model required for industrial use.

Relevant Hyperion platforms

Hyperion X1140-C

AxiKeystone

A CPU- and memory-led HPC platform for simulation, genomics, analytics, compilation and workloads with large in-memory working sets.

One 4U node; 128 AMD EPYC cores; 1.15TB ECC DDR5; one NVIDIA RTX PRO 4000 24GB; enterprise U.2 NVMe.

Hyperion X2160

AxiVector

A tightly coupled two-node simulation and mathematical-computing platform for MPI, CFD, FEA and optimisation.

Two 4U nodes; 256 AMD EPYC cores; 2.3TB aggregate ECC DDR5; two 96GB GPUs; direct 100GbE RoCEv2 RDMA between nodes.

Hyperion X3140

AxiLattice

A shared departmental CPU-compute facility for research pipelines, simulation, genomics and queued multi-user work.

Three 4U nodes; 384 AMD EPYC cores; 3.4TB aggregate ECC DDR5; three professional GPUs; switched 100GbE RDMA, 25GbE storage and Slurm.

Configuration and quotation

Validate this workload on Hyperion

Final architecture and price depend on representative code and data, concurrency, storage, networking, site infrastructure, component availability and export compliance.

Request Formal Quotation