Facing space and cooling limitations at its on-premises site, Helmholtz Munich moved its HPC cluster, including 360 GPUs, to NTT Global Data Centers’ Munich 2 colocation facility. The result is a greener, more flexible setup and a business case showing the advantages of renting. Dr Alf Wachsmann, Head of Scientific IT, Helmholtz Munich, tells us more.
Medical research requires a powerful IT infrastructure. Helmholtz Munich has relied on NTT DATA’s expertise for many years. Moving its server infrastructure – including the high-performance computing (HPC) cluster and 360 GPUs – to the NTT Global Data Centers (NTT GDC) Munich 2 colocation site laid the foundation for efficient next-level scientific computing. Compared with its on-premises data centre, Helmholtz Munich benefits from this economically sound and significantly more sustainable solution.
Business need: Strengthening biomedical research capabilities with a future-proof high-performance data centre
Modern research increasingly relies on high-performance computing power. At Helmholtz Munich, the top priority is processing the majority of its scientific data on dedicated in-house infrastructure.
Relying on cloud-based resources is not an option – only a few dedicated research tasks that exceed its own computing power are performed on national resources such as the supercomputer in Jülich, Germany. As scientific computing and AI-driven applications have continued to advance, the demand for scalable computational acceleration has increased steadily. In life sciences research – where traditional CPU-based IT must work alongside high-performance GPU computing – this development has led to the deployment of an outstanding onsite HPC cluster.
However, the existing data centre had reached its limits in terms of space, and energy-efficient operations could no longer be ensured. GPUs – with their massive parallel processing power – require far more electrical capacity per rack than CPUs and the resulting heat output exceeded the centre’s cooling design.
To maintain both sustainability and efficiency, a modern, future-ready solution was needed. The midterm plan to switch from air cooling to liquid cooling was another core technical requirement. Internal cost-benefit analyses strongly favoured moving to a colocation data centre. This was when the IT department began looking for a trusted partner to tackle Helmholtz Munich’s current and future challenges.
“With a total of 300 HPC servers – 100 of them equipped with 360 GPUs – we are among the leading IT clusters worldwide,” said Dr Alf Wachsmann, Head of Scientific IT, Helmholtz Munich. “NTT Global Data Centers now provides us with a high-performance, efficient and future-oriented data centre infrastructure. Through joint projects, we benefit from their extensive expertise in data centre design and liquid cooling. Their strong commitment to sustainable energy usage and the short distance to the Munich 2 site further convinced us.”
Solution: Colocation data centre enables resilient and sustainable next-level IT infrastructure
Working on a long-term project with NTT DATA to optimise its core IT infrastructure, Helmholtz Munich evaluated several options and ultimately decided to enter into another strategic collaboration within the NTT Group. The new Helmholtz Munich data centre is now located at the NTT GDC’s Munich 2 site only 10km from the association.
Helmholtz Munich drives world-class medical research, exploring critical questions such as how climate change affects human health. It goes without saying that environmental compatibility was a decisive factor that guided Helmholtz Munich when selecting its colocation provider. NTT GDC’s Munich 2 site particularly stands out for its sustainable cooling concept. Due to its proximity to the Alps, a large-volume flow of naturally cool groundwater runs beneath the colocation data centre. Following successful evaluations, NTT GDC implemented an intelligent and sustainable groundwater cooling system. In addition, the data centre is powered exclusively by green electricity. Both factors were key arguments for Helmholtz Munich in choosing this location.
Initially, 15 servers were installed as a learning experience within the leased colocation space. Subsequently, 300 HPC servers already in operation at the previous site, along with approximately 400 servers and storage devices, were relocated by a specialised service provider in three phases. The transfer of the storage systems was handled by the system manufacturer. Helmholtz Munich is connected to the data centre via two redundant dedicated lines, ensuring high bandwidth and reliability.
At present, both CPU and GPU servers are cooled using a conventional air-to-water cooling approach. In the future, the racks for the HPC cluster will be directly liquid-cooled. NTT GDC already operates one of the world’s largest liquid-cooling data centres in Utah, US. For years, NTT GDC has designed its data centres to allow liquid cooling to be retrofitted at any time. The cooling backbone is separated from customer colocation areas and its architecture is designed to allow the direct installation of liquid-cooling systems.
Outcomes: Dynamic partnership drives innovation for reliable high-performance computing in bioscience
Until a few years ago, all IT-related tasks at Helmholtz Munich were handled by the research teams themselves. A few years ago, the IT organisation started transitioning to a centralised model to streamline operations and optimise the use of IT resources. Today, the IT team enables researchers at Helmholtz Munich to focus entirely on their areas of expertise, while they manage the complex computational infrastructure. Typically, 600-700 users rely on the research centre’s computing power across roughly 500 projects running in parallel. By moving the data centre into NTT GDC’s colocation space, Helmholtz Munich has reached a new level of efficiency and flexibility.
Computing power can be added easily, resulting in shorter lead times. This higher agility is especially valuable in time-sensitive projects.
New cooling options can be explored with the guidance of NTT GDC’s experts. Addressing today’s needs and future requirements is key to remaining at the forefront of life sciences research.
Additional services – such as preparing dedicated client reports or handling smaller adaptations at the racks in the data centre – simplify the team’s day-to-day work.
The collaborative partnership with experts at NTT GDC is already paying off. While many research centres hesitate to build and operate their own HPC clusters, Helmholtz Munich demonstrates that doing so within a colocation data centre is both feasible and worthwhile.
Dr Alf Wachsmann, Head of Scientific IT, Helmholtz Munich, offers further insight.
How has the move to NTT’s Munich site made things more economical and sustainable compared to how you operated previously?
As a government-funded research centre we’re required to consider sustainability – this includes understanding where the electricity comes from. Our main campus runs entirely on green electricity, so it was important to find a data centre that could guarantee the same, and NTT could.
On cooling, NTT uses free coolers on the roof, supplemented with groundwater. It was very important to us that a lot of electricity wasn’t being used to produce cold water to cool the facility. Those two things meant our energy monitoring and auditing could sign off on it. Even though the site isn’t on our campus, we still have to report those numbers and they align with what we’re already doing or planning on campus. So, it was a good match.
Helmholtz Munich researches how climate change affects human health, so how do you respond to the widely reported environmental concerns around data centres?
It’s a real concern. The amount of compute we use across the Helmholtz association is tiny compared to the hyperscalers and neoscalers, but the scale of AI expansion raises a genuine question about what it’s worth to humankind. Internally, I’ve been questioning whether we should have something like an ethics committee for AI jobs. They consume so much electricity and produce so much heat that, as with animal experiments, perhaps you should have to justify how many GPU hours you require and why it’s necessary. A committee could then approve or reject.
What advice would you give to those research institutions who are hesitant to build and run their own HPC clusters?
It depends why they’re hesitant. Typically, it comes down to investment and complexity, rather than environmental concerns. In Germany, if you run the numbers on total cost of ownership, there’s a clear benefit to owning your own hardware compared to the cloud, except when it comes to the building. We expected building our own facility to pay off after 10 years or so, but even after 20 years, renting space was still more cost-effective. It was a very clear business case.
