✓ Two decades of forging the future with High Performance Computing.
✓ We craft personalised solutions alongside partners to bridge the technology gap for clients.
OCF is a UK High Performance Computing (HPC) integrator. We design, build and support HPC, storage, cloud and AI systems for research institutions and businesses.
We are vendor-neutral. The system you get is the one that fits your workload, not the one that suits a single manufacturer. We stay with you from the first design conversation through installation, software and managed support, run from our headquarters in Sheffield with a network of more than 20 hand-picked technology partners behind us.
High Performance Computing is the use of many computers, called nodes, working together as one system to run workloads that a single machine, even a powerful server, cannot handle. The nodes share the work in parallel and return results in a fraction of the time a standard system would take.
It is the foundation for large-scale research, simulation, modelling and AI training, from climate science and drug discovery to computational fluid dynamics and machine learning.
Most of the systems that disappoint were bought on the wrong assumption. These are the four we see most often, drawn from years of designing systems for universities, defence and industry.
Bigger GPUs are not better GPUs. The largest GPUs sit idle when a workload cannot fill them. A teaching customer once asked for the current flagship GPUs to run training courses. The real usage was low and came in short bursts, and even as planned the design would not have given enough capacity for the courses they wanted to run. The answer was software that slices each GPU across many students during teaching, then pools the same hardware for researchers out of hours. A larger number of smaller GPUs is often easier to share than a handful of big ones.
Compute is not the system. A system is compute, storage and networking in balance. Spend everything on compute and you starve the storage, which is where the research data lives and where the bottleneck shows up. You can have the fastest system, the cheapest system, or the most power-efficient system. You only get to pick two.
The fastest network is often the wrong spend. One customer specified the latest, highest-speed interconnect across the whole cluster. On review of their scheduler data, more than 90 per cent of their jobs ran inside a single node, so the high-speed fabric was money that should have gone into compute. Right-sized Ethernet gave them the performance they needed and freed budget for more compute nodes. We have created a guide on Infiniband vs Ethernet.
The work starts at go-live, not before it. Hardware arriving is the start of the job, not the end. Storage that cannot feed the GPUs, a system sized for training but used for inference, or hardware that is right while the software stack is wrong: these are the failures that slow a system after it is installed. Add Cyber Essentials and ongoing software maintenance, and a cluster can no longer be deployed and left alone for years. Smaller teams often run HPC off the side of one engineer's desk and underestimate the time it takes.
Our first meeting is a conversation, not a pitch. We bring no vendor to it. We are there to understand your priorities and your targets before anyone talks about hardware.
We ask why. If you arrive asking for a specific GPU or a specific server, we ask what you are trying to do with it. The answer often points to a better system than the one in the request.
We read your data. When a workload is hard to describe, we use a platform that analyses years of your scheduler data and shows you what you run. The design follows the evidence.
We pick the right technology, not a vendor's catalogue. We will specify compute from one manufacturer and storage from another if that is the right system for you. Being vendor-neutral is the point: we work for the workload, not the brand.
We deliver and support your entire system, from initial design to live service.
Our technical architects and partners design a system around your operational constraints, not a template.
Our hardware and software teams build and deploy the cluster to your specification, in the right order so you can start using what is delivered first.
Air-cooled or liquid-cooled, x86 or POWER, GPU or FPGA, built with partners including NVIDIA, Lenovo, Dell and Supermicro and chosen for your workload.
Our OCF Steel cluster-management platform combines open-source tools with OCF software and support to deploy, manage, monitor and run your cluster.
Parallel storage built on file systems such as Lustre, IBM Storage Scale, BeeGFS and others, with InfiniBand or Ethernet sized to the jobs you run rather than to a spec sheet.
Support delivered remotely from Sheffield, set up before go-live and continuing after it. This is our answer to the HPC skills shortage: the expertise sits with us, the system stays yours.
Extend an on-premise cluster to the cloud for peak demand, or move workloads where it makes sense to.
✓ Vendor-neutral, with more than 20 hand-picked technology partners
✓ OCF Steel, our own cluster-management platform, used by real customers
✓ End-to-end: design, installation, software, managed service and consultancy
✓ We stay after the sale, and we are the ones still there when something needs fixing
Our case studies show what an OCF system has done for research that depends on it.
We work across higher education, defence, engineering, manufacturing, life sciences and energy.
Higher education and research is where most of our work sits, with Russell Group universities running everything from traditional HPC to modern AI on systems we have built and support.
Defence, in the public sector and the supply chain, is our fastest-growing area. The hard part here is often security: systems built to meet defence requirements, delivered by people cleared to work on them, against a real skills gap in the UK.
We also support manufacturing, engineering, life sciences and energy, where simulation, modelling and analysis depend on the system keeping pace with the work.
Technical Director at N8 CIR
"Bede enables us to deal with data at a scale that other machines can’t.
It’s not just far faster, it enables us to tackle problems that were simply beyond our capabilities before."
University of Birmingham
"With OCF as our partner, we have the right guidance, skills and expertise to continually move BEAR forward. It’s not just about having the largest cluster, it’s about providing cutting-edge HPC tools for our researchers"
Head of HPC at UKHSA
"OCF’s engagement in partnership with NVIDIA and Quantiphi, has been extremely valuable.... As a result, UKHSA teams now have a clear pathway for efficiently evaluating AI models applied to domain-specific health security use cases."
What is high-performance computing? High-performance computing is the use of many computers, called nodes, working together as one system to run workloads a single machine cannot handle. The nodes share the work in parallel, so results come back in a fraction of the time. It is used for research, simulation, modelling and AI training.
What is an HPC cluster? An HPC cluster is a group of networked compute nodes managed by a scheduler that shares work across them. A high-speed network and parallel storage keep the nodes fed with data. To the people using it, the cluster behaves as a single system.
HPC or cloud: which is right for my workload? On-premise HPC suits steady, around-the-clock workloads and large or sensitive datasets. Cloud suits short bursts and unpredictable demand. For continuous use, cloud tends to cost far more than owning the system. The deciding factors are predictability of demand, data sovereignty and data size.
How much does an HPC system cost to run, not just to buy? The purchase is only part of the cost. Power, cooling, support and the staff time to keep the system running and secure all add up after go-live. Cyber Essentials and ongoing software maintenance make a system harder to leave alone for years. We help cost the running of a system, not only its purchase.
Do I need the latest GPUs? Not always. The largest GPUs sit idle when the workload cannot fill them, and a set of smaller GPUs is often easier to share across users. The right choice depends on what you run, which is why we look at your workload before recommending a GPU class.
What is an HPC integrator, and why not buy direct from the vendor? An integrator designs, builds and supports a complete system, bringing compute, storage, networking and software together from more than one manufacturer. Buying direct gives you parts. An integrator gives you a working system and the support around it. Because we are vendor-neutral, we pick the right technology for the workload rather than a single vendor's catalogue.
What support comes with an HPC system from OCF? A managed service delivered remotely from our Sheffield headquarters, set up before go-live and continuing after it. A managed-services contact is assigned early in the project, with a one-month period after handover where the project manager stays involved. Recurring issues become daily automated checks.
Can old and new hardware run in the same cluster? Yes. OCF Steel presents older and newer kit to users in the same way, so they submit jobs without knowing which generation of hardware runs them. Results take longer on older equipment, but the way people work does not change.
OCF works best with organisations that want a partner, not a parts list.
You are a good fit if you know there is a skills gap and you want the people who keep your system running to be free to focus elsewhere. If you want the system designed around your work, used well once it is live, and supported for the long term. If you value the advice your technical team gives as much as the price on the quote.
We are a professional services and managed services business first. That is where we add the most, and it is the work our customers (such as those highlighted below) come back for.
The systems that work are the ones designed around the workload from the start. Tell us what you are trying to run, and we will tell you what it needs.
