Together with the Export and Investment Fund of Denmark (EIFO) and NVIDIA, a world leader in AI computing, the Novo Nordisk Foundation is establishing the Danish AI Innovation Center.
Once operational, researchers from the Danish public and private sectors will have access to a state-of-the-art NVIDIA supercomputer optimized for large-scale AI projects, as well as leading NVIDIA software platforms, training, and expertise. The goal is to accelerate research and innovation in areas ranging from healthcare and life sciences to the ecological transition, supporting the development of innovative solutions to the world’s biggest challenges.
Denmark’s new supercomputer – a large-scale project NVIDIA DGX SuperPOD – is powered by NVIDIA H100 Tensor Core GPUs and interconnected using NVIDIA Quantum-2 InfiniBand networking.
The supercomputer and the collaboration with NVIDIA will enable Denmark to conduct large-scale projects in many areas where AI is a valuable tool, as well as in AI research itself. In some flagship project areas, researchers will be able to collaborate with NVIDIA’s expert teams to co-develop solutions to complex problems.
These include research in the pharmaceutical and biotechnology fields, including protein design, accelerating the green transition, and developing fault-tolerant (i.e. error-free) quantum computing.
Eviden, an Atos company and European leader in advanced computing, will deliver, install and configure the supercomputer, named Gefion, and support the start-up phase. The center is expected to be ready for pilot projects before the end of 2024.
The Foundation has established a public limited company, the Danish Centre for AI Innovation A/S, which will own and operate the supercomputer, with all revenues being reinvested in the initiative. EIFO is a minority shareholder, holding 15% of the shares.
Global data center provider Digital Realty will host the supercomputer in one of its AI-ready facilities in Denmark. The data center is designed and built to be sustainable and runs on 100% renewable energy.
The supercomputer will have the highest level of security and will contribute to Danish data sovereignty. Once fully operational, it will therefore be able to support projects involving sensitive data. No data will be stored permanently in the centre and the centre’s users will have full control over their data at all times.
Differential payment models are being developed to balance commercial and academic use to cover current and future costs of the center.
Frequently Asked Questions
When will the center be operational?
A pilot phase will begin in October 2024, during which DCAI will implement the necessary software and processes. Selected and sufficiently mature projects will be able to start during this phase, with users providing valuable feedback to DCAI.
The goal is then to launch most project types in the first half of 2025. A solution for projects involving sensitive data will come later – we will share more information as we have it.
Can I work at DCAI?
We will be hiring later this year and posting these positions on the Foundation website. LinkedIn.
Can you tell us more about the key areas of the project?
Several general areas have been developed and approved by the Novo Nordisk Foundation in collaboration with EIFO and NVIDIA. They are still evolving, but are currently as follows:
– Basic research on AI
– Developing democratic and human-centered AI healthcare solutions
– Development of major Danish language models, particularly in the field of health and biomedicine
– Pharmaceutical and biotechnology research, including protein design
– Acceleration of the green transition, including research into energy materials
– Data collection and AI for Earth-2 type projects, using multimodal sources and remote sensing to understand and predict climate, weather, forestry, agriculture, oceans, etc.
– Accelerating the development of fault-tolerant quantum computing and quantum state simulation
In projects in these areas, researchers will be able to collaborate with teams of NVIDIA experts to co-develop solutions to complex problems and push technological boundaries.
What if I want to use Gefion for a project outside of these areas?
Researchers and innovators from all fields are welcome to use the supercomputer, provided they meet access criteria regarding ethical use and accept the terms of use. (both will be specified in the Access Charter).
NVIDIA will be able to provide general support to all projects, including skills upgrades and train-the-trainer training. However, resources allocated by NVIDIA teams to co-develop new solutions will initially be reserved for flagship project areas.
Can the supercomputer be used to develop large language models (LLMs) that meet Danish needs?
Yes, this is a possible use case. The development of large Danish language models, especially in the field of health and biomedicine, has been identified as a possible point of collaboration between Danish researchers and NVIDIA experts. Gefion’s architecture is optimized for training and deploying large AI model computing workloads, and LLM training is a good example of this.
Who will prioritize the large language models (LLMs) for which Gefion will be used?
As long as sufficient capacity can be allocated, users are welcome to use Gefion to train or refine LLMs. For such projects, as for many others, DCAI and/or NVIDIA would be able to provide technical expertise to help accelerate the training process. For the development of health or biomedical LLMs – one of the flagship project areas – researchers will be able to collaborate with NVIDIA expert teams to co-develop solutions.
How can I access it?
The details are currently being worked out. We want to establish a process that will allow researchers and innovators from the Danish private and public sectors to access the facility’s services relatively quickly and smoothly, while ensuring that all users meet certain basic criteria, including ethical use.
I may want to use the facility in 2025. How can I prepare for this?
Gefion is a powerful compute platform designed for large-scale computing and AI workflows using GPUs, so a good understanding of the NVIDIA software stack and how to scale your computing workload on GPUs will be a good starting point.
We’re also hosting a series of educational events. In August, NVIDIA experts will lead a workshop at the Niels Bohr Institute focused on quantum computing as part of NVIDIA CUDA-Q. (Registration is now closed.)
In September, join DCAI and NVIDIA for an instructor-led workshop, “Model Parallelism: Building and Deploying Large Neural Networks,” to learn how to scale the training and deployment of LLMs and neural networks across multiple nodes. Led by a DLI-certified instructor, this workshop offers the opportunity to earn an industry-recognized certificate of competency through assessments. Attend onsite in Copenhagen or virtually. Learn more and apply here.
How much will access to the supercomputer cost?
The aim is to ensure that all researchers, innovators and companies in Denmark who wish to use the facility will be able to do so, while ensuring that DCAI can cover the running costs of the centre and the cost of future upgrades.
So there will be a differential pricing model, with higher costs for some types of users than for others. We will communicate more information as it becomes available.
What if the data is sensitive?
The environment will be designed to – eventually – be able to facilitate a secure, multi-tenant configuration, where each user will have their data and computing tasks isolated and secure from others.
We want to have this solution as soon as possible, but building a platform that offers this level of security is a complex process. We will share more information on the expected timeline as soon as we know it.
How can I find out more?
This page will be updated as new information becomes available. We will also share important announcements on the Novo Nordisk Foundation channels. Later in 2024, an official website will be created for the company and the supercomputer.
