On-Site HPC — On-Premises High Performance Computing Infrastructure

On-premises HPC brings high performance computing capacity into your own facility. It is the preferred model for organisations that need data sovereignty, low latency and full control.

Why On-Premises

Data Sovereignty and KVKK Compliance

Your data never leaves your organisation’s physical boundary. For organisations in healthcare, defence, banking and the public sector this is often a hard requirement. Assessments can be made against KVKK, BDDK and sector-specific regulation.

Low Latency

Compute and data source sit on the same physical network — a decisive advantage for workloads that move large volumes of data, such as genomics, CFD and manufacturing simulation.

Full Technical Control

Hardware configuration, network topology, user management and workload policy stay entirely under your control, allowing customisation that shared cloud infrastructure cannot offer.

Long-Term Cost Efficiency

Over a five-year-plus horizon, owning on-site hardware delivers a lower total cost of ownership than equivalent cloud capacity.

How the Mevasis On-Site Service Runs

1. Requirements Analysis and Feasibility

• Workload profile (MPI, GPU, data-intensive, mixed)
• User count and concurrent job estimate
• Assessment of existing facilities (power, cooling, network)
• Budget and TCO analysis

2. Architecture Design

• Node type and count
• Network topology (InfiniBand, Ethernet)
• Storage system (BeeGFS, NFS, Lustre)
• Job scheduler selection (SLURM, PBS)

3. Procurement and Installation

• Hardware sourced from approved suppliers
• Physical installation (rack, cabling, power)
• Operating system and cluster management software
• Network configuration and high availability

4. Software Stack Configuration

Rocky Linux 9 / AlmaLinux 9
Warewulf 4 (node provisioning)
SLURM (job scheduler + accounting)
Lmod (module management)
OpenMPI + MVAPICH2 + Intel MPI
CUDA + cuDNN (for GPU nodes)
Prometheus + Grafana (monitoring)

5. Testing and Acceptance

• HPL (Linpack) and GPU benchmark tests
• User acceptance testing
• Performance reporting

6. Handover and Training

• System administrator training
• User documentation
• Optional managed operations

Reference Configuration

An example mid-sized production environment with 32 compute nodes:

For sizing specific to your organisation, request a quote.