Cordula Schnuer: The EuroHPC Joint Undertaking is a pan-European initiative. How did Luxembourg come to be the headquarters of the project?
Anders Dam Jensen: That’s part of a political discussion when a new agency or any EU body is being set up. It was before my time. Luxembourg, as I understand it, very early on wished to have this here. And I think rightfully so. They contributed quite a bit to the text that governs the joint undertaking. And what Luxembourg contributed greatly to is that there was room for everyone. The fact that we now have four very nice petascale machines--Luxembourg had a strong word in that.
You took up your role as executive director in 2020. Has the development of the network gone as planned?
The concept of the joint undertaking started back in 2018 and it was born out of the European Commission. A lot of work was done by the commission to get everything ready for September 2020, when we became an autonomous organisation, and being autonomous means that we have our own budget.
I started on 16 September, we declared autonomy on 23 September and, with that, I could start signing contracts. I don’t think a lot of people have had the pleasure of purchasing six supercomputers in the space of two weeks. That plan for the procurements was neatly laid out.
We’ve been hit by covid in a couple of places. We will inaugurate our biggest machine so far, Lumi in Finland, on 13 June. That was meant to have been in March, but basically covid caused a three-month delay on that. Lumi will be number three in the world. That’s a fantastic achievement only surpassed by the Japanese machine Fugaku, and then the US is coming in with a brand-new machine, which essentially is the same technology as Lumi, just a little bigger.
The EuroHPC project wants to strengthen industry and research in Europe by providing access to infrastructure. What has been the response of the players this network is aimed at?
Everyone is excited. If you look at the capacity we’re bringing to Europe, with the likes of Meluxina, it’s a significant lift compared to what was available before. We’re democratising it from the point of view that it becomes available to everyone. There always have been larger machines in Europe, but they’ve been national machines. There has been some good collaboration between countries, but it’s been more on a bilateral basis.
There’s no lack of good project proposals to run on the machines.
With the joint undertaking, we’re taking this to all the members, which by the way aren’t only the EU member states. We’re bringing high-performance computing to users that have not had as easy access to it as they will have now. To allocate the capacity, we basically go out and ask for projects to run on them. We’re about to close the next round. And there’s no lack of good project proposals to run on the machines. We have the tough job of choosing.
Can you give me some examples of these projects?
It’s very much across the field of science. The big one at the moment is climate change. Weather modelling is huge. The European Union runs a special project called Destination Earth, which is about making a digital twin of the Earth, allowing us to model climate change but also some of these weather phenomena that we’ve been seeing, not to mention what happened here last year. The idea there is to get models so that we can predict these early enough to be able to react. Destination Earth is desperately waiting for Lumi to be fully operational, Leonardo [in Italy] as well, but it is one of the projects that’s now gone to Meluxina because they can get started there, which is fantastic.
There’s a similar one on the human brain, there’s drug discovery, engineering and fluid flow modelling. All of those are the traditional applications for high-performance computing. As we broaden access to the machines, my hope is that we will discover brand-new applications to run on these things.
Is there a preference for one computer in the network over another?
There are definitely preferences for one machine over another. I think over time it will fade. We need to remember that we’re bringing in something new. When we have a computer, it’s hosted by a hosting entity. And, of course, there are some people who have worked with that hosting entity, and therefore know the people and so on. What we’re hoping to do is to bring all of those hosting entities together, so they also can help each other. Meluxina has been really good at that. There’ve been some projects where we needed to do some training, and they stood up and helped with that. People are getting to know other machines.
The interesting thing about Meluxina is that its architecture is a smaller version of the big machine we’re installing in Italy. From that point of view, there’s a lot of interest in Meluxina because there’s an opportunity to get your applications ready for when Leonardo becomes operational, hopefully at the end of this year. You’ll have the same machine, just 20 times the performance. There are some key projects that are waiting for Leonardo, but they now have an opportunity to get started on Meluxina.
Meluxina launched roughly a year ago. Are you satisfied with its performance?
Very. It’s 48th in the world. This is not a small machine. There’s some very interesting technology in that machine. We have a very good team out there with LuxProvide. In general, across our machines, we have very good teams. But Meluxina is dear to me. I live in Mersch, it’s in Bissen. This is where we take people when they need to see a supercomputer.
Each of the machines is unique. There are nuances. The technologies that went into them matter. You need to master them in order to get the full power out of them. Each of the teams are experts in the technology that was purchased for their machine. What we’re getting better at within the network is helping each other out. We’ve had examples when the technical team in one area looked at an applicant and said this doesn’t really fit well on our architecture, we should move into this one. They work with each other, and we get the application to run where it has most impact.
The machines help develop weather models to combat climate change, but large-scale computing can have a huge carbon footprint. How is EuroHPC addressing this?
High-performance computing requires electricity. It requires electricity to run the machine, but also to cool it. And that’s not insignificant. In the application to host a machine, there’s a strong focus on energy. We take a holistic view, because we can’t just choose low power, and then not get any compute. We look to get energy efficient solutions but, much more importantly, when we select the hosting entity, they come back in their application with how they’re going to deal with the power. Where are we going to get it from? What are they going to do with the excess heat and so on. Each of the machines has a story behind it.
We try to make sure to minimise the carbon footprint of the machines
Lumi is a fantastic example. Because of its geographic location next to a river, we get the electricity from a hydropower plant, The excess heat is pumped into the city of Kajaani and used to heat houses. Meluxina equally has a colocation with an industry on the other side, where they do heat exchange.
We try to make sure to minimise the carbon footprint of the machines. What’s really exciting about that, though, is the whole research and innovation programme, where we’re funding research into HPC technology. We take a very strong focus on energy. We have a very interesting project developing a European microprocessor. The aim really is to have a low-power HPC processor, which we hope down the line will show up in our machines.
Can you estimate the impact of the energy crisis, which is also leading to higher electricity prices, on the projects in the joint undertaking?
It’s too early to tell. As energy costs go up, so does the cost of operating the machines. We are fortunate that most of our hosting entities have pre-purchased energy for a foreseeable future. We have some time to figure out how we deal with it on the other side. Energy is a very substantial part of the cost of operating each of the machines.
More than half of companies in the EU report difficulties recruiting ICT staff--67% in Luxembourg--according to the European Commission’s Digital Economy and Society Index. What has been EuroHPC’s experience?
Everyone is having challenges, not just Luxembourg. It’s across Europe, and maybe even the world that those people who know what they’re doing are in short supply. And HPC is an even smaller niche of people. It’s our mission to try and broaden the knowledge and educate more people in this area. Hopefully, we’ll be able to help ourselves.
There’s no question that it’s difficult to recruit. There’s only so much talent available. But that being said, we’re dealing with the larger machines in Europe. They’re also attractive for the people who are in this field. It’s not easy to recruit, but it’s not impossible either.
How are you preparing the transition from petascale to even more powerful exascale computers?
The core technologies that we’re seeing, even in a machine like Meluxina, is what scales up. What we need to spend a lot of time and effort on now is to make sure that we are able to use those machines to their full potential and take advantage of the special technology that’s in there.
We know how to make the compute node, the general node that does the computation. The challenge is making them all communicate together. You already have quite a large network of nodes in a machine like Meluxina. We’ll be scaling that even further into exascale. Managing that and taking advantage of it is an art that anyone in the world has to master.
With a little luck, somewhere in 2024 we’re aiming for exascale to be available to European scientists.
The US will be the first with an exascale machine, Frontier. The technology in Lumi is exactly the same. From that point of view, we’re getting ready. With a little luck, somewhere in 2024 we’re aiming for exascale to be available to European scientists.
The next big thing is quantum computing, with Luxembourg also already working on projects in secure quantum communications. What strategy are you developing to tackle this technological leap?
That is the coolest thing on this side of the Earth. It’s natural to see how we can take advantage of quantum together with HPC. Because the belief is that quantum will be very good at some problems that traditional computers are very bad at.
One of the things that Europe has been pioneering is the concept of a modular supercomputer. Instead of it just being one big machine, it’s modules, and each module is good at doing something. Your application passes the problem around the different modules to have different parts of it done where it’s most suitable. Meluxina has three different partitions where you can divide the workload depending on what type that particular sub-problem is. You throw it out there, combine it with something from another module, and in that architecture, quantum is a natural extra module that you add to it.
The role of the joint undertaking is very much to do what we’ve done for HPC, which is providing the infrastructure to the scientists. The aim is that EuroHPC will fund a number of quantum computers across Europe. What we’re looking for is technology diversity. There are different ways to produce a supercomputer, different methods, different underlying technologies. And I think it’s fair to say the jury’s still out on which one is going to be ‘the’ solution.
What we wish to do is fund as many different technologies as possible. We need to understand, and we need to find out, learn more, because this is very much still research. The aim is to give researchers access to the different technologies and build up knowledge. Europe is very far ahead in quantum technologies in several countries. This is also about spreading it out and making sure we get as many on this wagon as possible.
The joint undertaking has a budget of €7bn until 2027. That’s in less than five years. What will come after this funding period?
If we do a good job, I would imagine another X billion. We need to remember where this came from. EuroHPC is born out of countries realising that there was an underinvestment in HPC, and that Europe was falling behind. What we’ve done with the joint undertaking is closing that gap. Lumi becoming number three in the world demonstrates that we’re closing the gap rapidly. We’ll have a number of other machines coming online and with exascale, we will more than fulfil that mission and ambition. However, if we stop investing, then we fall behind again.
If we stop investing, then we’ll lose what we’ve won.
I would hope and expect that there will be another funding round. Because with the momentum and what this delivers to Europe, why not? And then we’ll see what other interesting new things the European Union will wish to throw at EuroHPC. It wasn’t obvious that quantum was coming to us. I’m really delighted that the new regulation chose to put quantum with the joint undertaking, because it makes sense. And it’s just super exciting. We’ll see what the next challenge throws at us over time. But the fact is, if we stop investing, then we’ll lose what we’ve won.