High-performance computing
CPU and accelerated clusters for simulation, modelling and scientific research.
High-performance computing serves workloads that AI clusters often do not: reservoir simulation, computational fluid dynamics, weather and climate modelling, genomics and finite-element analysis. These codes are frequently bound by memory bandwidth and interconnect latency rather than accelerator count. We supply dense CPU nodes, accelerated nodes where codes support them, low-latency InfiniBand fabrics, and parallel file systems such as Lustre or IBM Storage Scale, balanced against the profile of the applications you actually run.
We work from benchmark data and scheduler logs where they exist, so node count, core-to-memory ratio and storage tiering reflect real job mixes. Procurement covers compute, head and login nodes, burst buffers, archive tape or object tiers, and the cluster management and scheduling software. Rack-level power and cooling are confirmed against your facility, and acceptance testing against agreed benchmarks is written into the contract before delivery is scheduled.
Benefits
Balanced for real codes
Memory bandwidth, interconnect latency and I/O are matched to the applications in your job queue.
Benchmark-based acceptance
Performance targets are agreed in advance and verified at handover, not assumed from datasheets.
Tiered storage design
Scratch, project and archive tiers are sized separately, controlling cost without starving active jobs.
Vendor-neutral comparison
Platforms from several OEMs are compared on fit, availability and lifetime cost.
Reference architecture
HPC cluster stack
Case study
Illustrative scenarioA national university consortium, Morocco
- Challenges
- Researchers shared an ageing cluster with long queue times, and the consortium needed a replacement suited to climate, materials and bioinformatics work within a fixed public budget.
- Relevant services
- We analysed scheduler records to profile the job mix, compared configurations from several OEMs and supplied a CPU-led cluster with a smaller accelerated partition and a tiered parallel file system.
- Outcome
- Queue times fell markedly, groups with GPU-ready codes gained a clear path forward, and the tender file satisfied the public procurement auditors.
Frequently asked questions
Can you respond to a public HPC tender?
Yes. We prepare technical compliance matrices, tender-specific OEM authorisation letters, bid bonds where required, and pricing in the tender's currency and Incoterm.
How is performance verified?
We agree acceptance benchmarks before order, for example HPL, STREAM, IO500 or runs of your own applications. Handover is signed off only when the installed system meets them.
What lead times should we plan for?
Standard CPU nodes usually ship within weeks. Accelerated nodes, high-radix switches and some storage controllers can take several months. We confirm OEM-quoted dates with each quotation and again at order.