CAE and Structural Analysis HPC
HPC infrastructure for finite element analysis, crash simulation, and multiphysics simulation.
Computer-aided engineering (CAE) is a fundamental discipline that increases design reliability while reducing the number of physical prototypes in product development processes. Engineering problems such as structural integrity, crash safety, thermal management, and dynamic load analysis are solved on finite element models with millions of degrees of freedom. Computations at this scale require memory capacity and parallel processing power far beyond what conventional workstations can provide.
Engineering teams in Turkey’s automotive, aerospace, defense, and machinery manufacturing sectors need to run CAE simulations on domestic infrastructure, driven by both competitive pressure and global OEM requirements. Keeping design and analysis data within national borders is also becoming a mandatory requirement for KVKK compliance.
CAE Workloads: Software and HPC Requirements
Structural analysis and multiphysics simulation workloads exhibit a heterogeneous profile in terms of both memory and processor efficiency. A linear static analysis scales efficiently to hundreds of cores, while nonlinear contact and damage simulations require high memory regardless of core count.
Structural Analysis Software
| Software | Primary Application | Parallel Scaling | Memory Intensity | Typical Core Count |
|---|---|---|---|---|
| Abaqus / Abaqus Explicit | Nonlinear static, dynamic, damage, contact | Very good | High | 32–512 |
| ANSYS Mechanical | General-purpose FEM, thermal, fatigue, modal | Excellent | Medium–High | 32–1,024 |
| LS-DYNA | Crash, impact, blast, sheet metal forming | Excellent | Medium | 64–1,024 |
| MSC Nastran / MD Nastran | Modal analysis, random vibration, aeroelastic | Very good | High | 32–256 |
| OptiStruct (Altair) | Topology optimization, NVH, fatigue | Good | Medium | 16–256 |
| Radioss | Crash, penetration resistance, hydrodynamic impact | Very good | Medium | 64–512 |
| SimScale / Code_Aster | Research / validation (open source) | Good | Low–Medium | 8–128 |
Multiphysics Simulation Tools
| Software | Scope | Parallel Scaling | Typical Core Count |
|---|---|---|---|
| ANSYS Workbench (FSI) | Fluid-structure interaction (Fluent + Mechanical) | Very good | 64–512 |
| Abaqus + Fluent (co-simulation) | Thermal-mechanical interaction, thermal stress | Good | 32–256 |
| MSC Adams | Rigid and flexible multibody dynamics | Limited | 8–64 |
| Simufact Welding / Forming | Welding and forming process simulation | Good | 16–128 |
Sector Application Areas
Structural Integrity and Fatigue Analysis
Strength verification under static and dynamic loads for vehicle bodies (BIW), frame structures, and mechanical components is the most common category in CAE workloads. Linear static analyses performed with ANSYS Mechanical or Abaqus for large model sizes (5–30 million elements) require high memory and efficient solver capacity.
The fatigue life analysis workflow typically follows these steps: calculating load history via modal superposition, applying damage accumulation models, and identifying critical regions. Since these calculations proceed serially one after another, a single analysis cycle can take 8–72 hours.
Typical resource requirement: 64–256 cores, 512 GB–1 TB RAM per node
Crash and Impact Simulations
LS-DYNA and Radioss are industry standards for simulating short-duration dynamic events (crash, penetration, blast). Full vehicle crash models work with 10–20 million elements; typical computation times for a 100 ms frontal crash simulation:
| Core Count | Computation Time (approx.) |
|---|---|
| 32 cores | 12–24 hours |
| 128 cores | 3–6 hours |
| 256 cores | 1.5–3 hours |
| 512 cores | 45–90 minutes |
During certification periods (NCAP, FMVSS, UN-R95 side impact), running multiple scenarios simultaneously is mandatory. Flexible and rapidly scalable HPC capacity to manage these periodic peak periods is a decisive factor in time-to-market competition.
Thermal and Thermomechanical Analyses
Thermomechanical coupling analysis is increasingly common in the design of power electronics cooling, exhaust manifold thermal stress, and engine components operating at high temperatures. In these two-way coupled analyses, both the CFD solver and FEM solver run simultaneously; low-latency network connectivity is critical due to the communication overhead.
Thermal stress analysis in EV battery systems across the cell-module-pack hierarchy has a similar workload profile: multi-layer material models and complex contact conditions require nonlinear solver capacity.
Typical resource requirement: 64–256 cores, 256–512 GB RAM, high-speed scratch storage (NVMe)
Modal Analysis and NVH
Noise, Vibration, and Harshness (NVH) analyses involve solving large eigenvalue problems. MSC Nastran SOL 103 (normal modes) and SOL 108/111 (frequency response) workloads require high memory and strong I/O capacity in parallel solvers. Random vibration analysis (SOL 111 PSD) and aeroelastic analysis (SOL 144, 145, 146) are among the most computationally demanding Nastran analyses.
Typical resource requirement: 32–128 cores, 256 GB–2 TB RAM per node
Topology Optimization
Topology optimization workloads aimed at lightweighting goals involve large computation loops by their iterative nature. OptiStruct or ANSYS Mechanical topology and shape optimization may consume 24–96 hours of total computation time for an optimization cycle requiring 50–200 iterations. The frequency of these workloads is increasing with the growing adoption of Design for Additive Manufacturing (DfAM) applications.
Typical CAE HPC Configuration
Login / Pre-Post-Processing Nodes (2×)
├── CPU Compute Nodes (16–64 units)
│ └── 2× AMD EPYC 9654 or Intel Xeon Platinum 8592+
│ 256–512 GB DDR5 ECC RAM/node
│ — Crash (LS-DYNA, Radioss), modal (Nastran), FEM (Abaqus, ANSYS)
├── High-Memory Nodes (2–8 units)
│ └── 2× AMD EPYC 9654
│ 1–2 TB DDR5 ECC RAM/node
│ — Large nonlinear models, aeroelastic, large NVH matrices
├── GPU Nodes (optional, 2–4 units)
│ └── 2× Intel Xeon + 4× NVIDIA L40S
│ — ANSYS Mechanical GPU solver (linear static acceleration)
└── Storage Layer
├── NVMe Scratch (BeeGFS parallel file system)
│ — 50–200 TB, 10+ GB/s read/write bandwidth
└── Capacity Layer (SAS/SATA or S3-compatible object storage)
— Result archiving, project history
Network: Mellanox InfiniBand HDR 100G or HDR200, fat-tree topology
Operating System: Rocky Linux 8 / 9
Job Scheduler: SLURM + MUNGE authentication
License Management: FlexLM / RLM license server (ANSYS, Abaqus, Nastran)
Why On-Premise or Rental HPC?
CAE workloads are among the most economically challenging for cloud use. The primary reasons are:
Large data movement: A crash simulation or large NVH analysis generates tens to hundreds of GB of temporary files during runtime. Scratch storage and data transfer costs in cloud environments significantly add to compute costs.
License economics: Commercial solvers such as ANSYS HPC, Abaqus, and Nastran use core-count-based licensing models. At high core counts, minute-based license rental in cloud environments can reach many times the cost of on-premise licenses.
Data security: Product design data and simulation geometry are in the most competitively valuable trade secret category. On Turkey-located, KVKK-compliant infrastructure, this data never leaves the customer’s facility or a Turkish data center.
Predictable capacity: Certification periods are predictable. Fixed-capacity rental or on-premise clusters eliminate the risk of spot instance scarcity or price surges during peak periods.
Mevasis CAE HPC Services
Mevasis provides end-to-end technical support on structural analysis infrastructure for CAE-focused engineering teams:
- Workload analysis and sizing: Your current simulation portfolio is profiled to determine the required core count, memory capacity, and storage bandwidth.
- Turnkey cluster installation: Plug-and-play HPC infrastructure including hardware procurement, InfiniBand network design, and SLURM job scheduler configuration.
- Software environment configuration: ANSYS, Abaqus, LS-DYNA, Nastran, and FlexLM/RLM license server installation and parallel scaling tests.
- Performance optimization: MPI parameter tuning, SLURM partition strategy, prioritization of high-memory nodes.
- HPC Rental: Flexible rental models for temporary capacity bursts specific to certification periods or new platform launches.
- On-Site HPC Installation: On-premise cluster infrastructure installed at your facility and managed by Mevasis.
- HPC Consulting: Bottleneck identification in existing infrastructure, software upgrade support, and simulation workflow automation.
All infrastructure and management services are delivered on Turkey-located data centers and sites; KVKK-compliant data processing and access control policies are applied as standard.
Contact us to discuss your CAE infrastructure needs.
Frequently Asked Questions
How many cores and how much RAM are needed for Abaqus? It depends on model size and analysis type. For a linear static Abaqus workload with 5–10 million elements, 64–128 cores and 256–512 GB RAM per node are generally sufficient. For nonlinear contact or damage analyses, memory requirements can reach 512 GB–1 TB independently of core count. Workload analysis on your existing models is recommended for accurate sizing.
Is InfiniBand mandatory for LS-DYNA crash simulations? Not mandatory, but it provides meaningful improvement above 128 cores. LS-DYNA MPP (Massively Parallel Processing) mode relies heavily on inter-node MPI communication. Compared to 100 Gbit Ethernet, InfiniBand HDR100 provides 15–30% better parallel efficiency and lower-latency synchronization in typical crash workloads.
Can I set up an on-premise cluster without an ANSYS HPC license? The compute infrastructure can be set up independently of licensing; however, running ANSYS Mechanical in parallel requires an HPC Pack or Task Scheduling license. Mevasis evaluates the suitability of your existing license portfolio for parallel use and provides consulting on license optimization. Integration of the FlexLM license server into the cluster infrastructure is carried out as part of the installation.
Is a high-memory node required for Nastran large modal analysis? For SOL 103 (normal modes) and SOL 111 (frequency response) workloads, matrix size grows cubically with the number of model elements. Models below 1–2 million degrees of freedom are manageable with 256 GB RAM; beyond that, high-memory nodes with 512 GB or 1 TB RAM are necessary. Aeroelastic workloads (SOL 144/145) are the most demanding Nastran analyses in terms of this requirement.
How is simulation data security ensured? On on-premise infrastructure, all data never leaves your facility. For simulation data held in Turkey-located data centers under Mevasis managed services, access control lists, encrypted transfer (SSH/TLS), and audit logging are applied as standard. All processes are conducted in compliance with KVKK domestic data processing requirements.
How can capacity be temporarily increased for certification periods? The HPC Rental service offers a flexible option for short-term capacity increases specific to NCAP, FMVSS, or DO-160 test periods. Extending existing on-premise cluster capacity with Mevasis rental nodes can be seamlessly integrated via overflow queue configuration. Contact us for details.