The workstation is ASUS's version of the DGX Station GB300 and was supplied by Nextron. According to the two companies, it is the first of its kind from ASUS in the Nordics, and probably the first DGX Station GB300 of any make delivered in the region.
Inge Røinaas Gran, CEO of SINTEF Energy Research, says: AI can change both how we conduct energy research and which solutions we can develop for our clients. Our strength lies in combining in-depth knowledge of the energy system with the ability to adopt new technology. With this investment, we are giving scientists more room to explore the possibilities, with the aim of better research and a faster path from knowledge to practical value.
A power-efficient data centre in a box
The DGX Station is the size of a high-end desktop PC and, at full load, uses roughly as much electricity as a household space heater. Yet thanks to its extremely large and fast memory, it can run large language models that until recently required a full-scale data centre.
Hans Ivar Skjelbred, Research Manager for the Digital Solutions group at SINTEF Energy Research, says: With our own infrastructure, scientists can choose AI models suited to the task at hand, instead of sending every computation to global services designed for millions of users with very different needs. This allows us to match resource use to what the work actually requires.
Keeping sensitive data in-house
Research scientists at SINTEF Energy Research develop models of the power system, such as SHOP, ProdRisk, EMPS and Trident, which are used, among other things, to assess investments, calculate the effects of new regulations and plan the operation of critical energy supply. AI will help them write, test and document the code behind these models. Because some of the data is sensitive, this work is done on the institute's own hardware rather than in the cloud.
Ellen Krohn Aasgård, Research Director at SINTEF Energy Research, says: When our models are used to plan the operation of the power system, we need to know exactly what happens to the data. Having our own machine gives us that control.
The workstation is not intended to replace cloud-based services altogether, but to complement them. It gives the institute its own expertise and capacity, and thereby greater freedom to choose which tasks are run locally and which in the cloud.
Expertise for the future
With AI, researchers can build models faster, explore more alternatives and spend less time on routine coding. That leaves more time for understanding the energy system and for further strengthening the models that decision-makers already rely on.
SINTEF Energy Research believes that combining deep knowledge of the energy system with responsible use of AI will be an important strength in the years ahead, both for the institute and for an energy sector facing major change.
Bjørnar Fjelldal, Research Scientist at SINTEF Energy Research, says: We want to both use and develop this technology on our own terms, with security, transparency and societal benefit as our starting point. That means we need our own expertise and tools.
A new laboratory for combined intelligence
The SINTEF Intelligence Lab (iLab) will be a shared platform that brings together hardware, software, services and people, with both artificial and human intelligence at its core.
Fjelldal says: We deliberately chose not to call it an AI lab. Intelligence is about more than artificial intelligence; it is also about our own knowledge and creativity. This machine is the first step.
Lasse Sandstone, Sales Manager for AI and GPU at Nextron, says: We are seeing more and more organisations wanting to use large language models without sending sensitive data outside their own walls. The fact that SINTEF Energy Research is among the first in Europe to adopt the DGX Station shows that Norwegian research communities are at the forefront. We find it exciting that it will be used for research on the energy systems of the future.
Facts: NVIDIA DGX Station GB300
- Workstation for developing and running advanced AI models, based on NVIDIA's Grace Blackwell architecture.
- SINTEF Energy Research's machine is the ASUS version (ExpertCenter Pro ET900N G3). Supplier: Nextron.
- One Blackwell Ultra GPU and one 72-core Grace CPU.
- 748 GB of coherent memory, of which 252 GB is fast GPU memory (HBM3e).
- Up to 20 petaFLOPS of AI computing power (FP4), according to NVIDIA.
- According to NVIDIA, it can run language models with up to around one trillion parameters.
- Power consumption up to 1.6 kW.
