Cloud and HPC for Research, Agriculture, and Life Sciences
Research departments all face the same challenges: insufficient computing power, rapidly growing datasets, and computing software that is decades old. SUE builds HPC environments in the cloud or in hybrid setups, as well as data platforms that meet the requirements of funders and support CI/CD for reproducible research. Deltares, the KNMI, and Rijk Zwaan are among the organizations working with SUE in this way.
Challenges for Research
According to NWO and SURF, the demand for computing power in Dutch research is growing faster than the supply. At the same time, funding agencies are imposing stricter requirements on data management. Key challenges:
- Research data that is not discoverable or reusable
Genomics and phenotyping data in plant breeding, model outputs from research institutions, sensor and field measurements: datasets are growing rapidly and are often scattered across various locations. NWO requires an approved data management plan within four months of funding approval, and Horizon Europe mandates FAIR data management. This calls for an infrastructure that includes metadata, access management, and archiving. - Computing Power That’s Falling Behind
Simulations, model runs, and AI training require HPC and GPUs. National capacity is limited, and the Snellius supercomputer won’t be replaced until 2027. Setting up an in-house cluster to handle peak demand is expensive, and using the public cloud requires expertise to keep costs under control. - Outdated research software and platforms
Computational engines written in Fortran, clusters running on an operating system that is end-of-life, and knowledge that leaves with the developers who wrote the code. At Deltares, this involved approximately six million lines of code in the D-Flow FM computational engine. Anyone who doesn’t understand that code cannot safely modernize it.
Our solutions
SUE combines HPC, platform engineering, and Application Modernization for research organizations.
- HPC and Data Platforms
HPC clusters on AWS or in a hybrid setup alongside SURF facilities, with storage tailored to the workload: a parallel file system as the working layer and object storage as a data lake and archive. With access management and metadata for FAIR data. We built a research environment on AWS RES for the KNMI; now that AWS is discontinuing RES, we are assisting with the transition. - MLOps and reproducible pipelines
CI/CD, containerization of analytics software, and automated training and inference environments. Every result can be traced back to code, data, and environment. We built a scalable CI/CD infrastructure for Deltares. - Mapping and Updating Legacy Computation Software
With Re:App we map out legacy computing cores: call trees, dependencies, global variables, and the link to technical documentation. At Deltares, developers can now answer architectural questions in seconds, whereas it used to take hours or days. In addition, we migrate platforms—such as Rijk Zwaan from CentOS to RHEL.
Projects we are proud of
From 6 million lines of legacy code to immediate insight: Deltares unravels 40 years of software history
NXP achieves greater scalability and flexibility with HPC migration to AWS
Our use cases in this sector
-
Genomics pipeline that scales to the cloud during peak periods: At plant breeding and life sciences companies, the volume of sequencing data is growing faster than their in-house computing clusters can handle, causing jobs to sit in the queue for days during project peaks. We deploy the analysis pipeline in containers so that the same steps run both on-premises and in AWS. During peak periods, we scale out to the cloud, using FSx for Lustre for the computing stage and S3 for storage. Costs are transparent on a per-project basis.
-
A language model based on decades of research reports: Researchers search through reports, protocols, and measurement reports scattered across archives and drives, and that data cannot be sent to a public AI service. We run an open language model using vLLM on GPUs in the organization's own environment and connect it to the document archive via RAG. The model only provides answers from documents to which the researcher has access rights and cites the source.
Together with our partners
Modernize the code. Optimize the cloud. Run the platform.
SUE is a Dutch cloud native solutions partner that modernizes mission-critical applications, reduces cloud costs, and manages your platform. Specialists in Linux, Kubernetes, and the cloud.
29+
Years Experience
150+
Engineers and consultants
1.150+
Certifications
ISO
9001 & 27001
500+
Projects
75+
Research papers
Partner of progressive organizations
Relevant Blogs for Research, Agri & Life Sciences
All blogs
Frequently asked questions
All frequently asked questionsAWS will no longer accept new users for Research and Engineering Studio (RES) as of September 29, 2026. The latest version, 2026.09, will be supported through September 30, 2027. After that, you have three options: continue managing RES yourself or with a partner (it is open source), switch to AWS services such as Amazon WorkSpaces or AWS Parallel Computing Service, or switch to another portal such as Open OnDemand.
None of these options is a direct replacement, so plan the transition well in advance of September 2027. SUE has built test environments on RES and understands the dependencies involved in a transition.
Pay only for the actual computing power you use. Let clusters scale up and down automatically so that no nodes sit idle between runs. Run jobs that can be restarted on spot instances, and save intermediate steps of long computations so that an interruption doesn't result in lost work.
Categorize costs by project or grant, and you’ll see how much each research project costs. When it comes to GPUs, first assess how well they’re being utilized before scaling up. Using cloud cost optimization and FinOps, SUE identifies these costs and works with you to implement the necessary measures.
Yes. You run an open-source model on your own GPUs or in your own cloud account—for example, using vLLM on Kubernetes—so that the data never leaves your environment. An alternative is to use a language model via a cloud service in a European region, such as Amazon Bedrock, which does not use your input to train models.
In both cases, you connect the retrieval-assisted generation (RAG) model to your own documents and data, with the same access permissions as the source. SUE’s Data & AI team starts with a use case workshop so you can first determine which question you want to answer and with which data.
Discuss your situation with an expert
Discuss your situation with an expert
* required