Primeur Magazine:When you talk about a big supercomputer, what do you call big?
Prof. Dieter Kranzlmüller:Big is in a sense, according to our task from the Gauss Supercomputing Centre in Germany that we should be setting up the largest capability systems for German science and research and this is the next phase of setting up these systems. Big means the system has 26.9 petaflops peak performance and about 311,000 cores working together on solving the largest application requirements we have in Germany.
What is your main goal with these machines? What are your main users?</em>
Prof. Dieter Kranzlmüller:The three centres work together because they have individual characteristics in regard to what kind of applications they are supporting and by having these three centres we are able to support a bigger number of applications covering all the domains that are needed for German science and research. We also do this by three different kinds of machines. In our case, the Leibniz Supercomputing Center has the most general-purpose system supporting a wide range of applications but also, of course, providing an architecture that offers general-purpose computing capabilities.
Primeur Magazine:Can you give a few examples of applications?
Prof. Dieter Kranzlmüller:The applications are covering quite a wide range, from astrophysics to earth sciences to biomedical examples, just to go from large to very small.
Primeur Magazine:Of course, it is not that important but everybody asks the question what will be the position in the TOP500 of that machine?
Prof. Dieter Kranzlmüller:It is a good question but my answer to that question is always: 'I don't care'. The truth is we assume that we will be ending up somewhere in the top 20 of the TOP500 list. However, that doesn't matter to us because what we need is the best system for the science available. If we can make the scientists happy, then we have achieved our goal.
Primeur Magazine:What is the expected lifetime of this system?
Prof. Dieter Kranzlmüller:From what we have seen in the past the lifetime is usually about six years. We bring it into operation now which means we're also shutting down the other system that is already operational for six years. In the next couple of months, the new system takes over.
Primeur Magazine:Are there plans to have an upgrade in the near future?
Prof. Dieter Kranzlmüller:Yes. We usually buy a new system every six years and in between, around three years later, we usually upgrade some capabilities. The planning is also the same for the SuperMUC next generation. This is phase one formally, and there is phase two in the current funding scheme. The system is funded by the federal government and the state of Bavaria. The funding is already granted for the current version and the next phase.
Primeur Magazine:One of the things that makes your machine special is that you also look at the energy consumption.
Prof. Dieter Kranzlmüller:That is indeed a speciality of the Leibniz Supercomputer Center which is in practice also forced upon us. The electricity costs in Germany are relatively high. We have to deal with the question as how to do as much computation as possible with as little energy as needed. We have been doing some research on the topic in the past. I think with the SuperMUC next generation we are making another step forward in being more energy efficient for the computing that we are getting out of it.
Primeur Magazine:How much power is the system using?
Prof. Dieter Kranzlmüller:That depends. The average power consumption should be around 3.5 to 4 megawatts around. What we are doing is to adjust the power consumption to the power that is actually needed. If you a Linpack benchmark for instance, the power consumption will go up. If you perform a regular application, the power consumption will go down usually.
Primeur Magazine:Do you have any plans to increase the power input to your centre?
Prof. Dieter Kranzlmüller:This will happen for the next system after the SuperMUC next generation. We are thinking about exascale-class systems. We have done a feasibility study to see whether that can be done on campus. The feasibility study was positive and included an increase in terms of power provisioning to the building. We now need to ensure that politicians also understand what their role is in this equation.
Primeur Magazine:Do you have an idea what to expect in power consumption from an exascale machine?
Prof. Dieter Kranzlmüller:That is hard to say six years in advance or somewhere in between but what we want to have is a power provisioning of around 20 to 25 megawatts.
That is what we have been inquiring and we found out that this is possible.
Primeur Magazine:When you talk about exascale, you are of course talking about an exascale machine which is proven and which suits your users. It will not be one of the first experimental exascale systems, I assume?
Prof. Dieter Kranzlmüller:My answer is the same one as for the TOP500. I don't want to have the first system. I want to have the best system and the best system needs to have some reliable and proven quality, also in terms of technology. I would neither limit it to a particular technology but we will take whatever technology is there that is proven and reliable and the best one for the applications.
Primeur Magazine:If you look at the European scale, EuroHPC has now been established, as well as the EuroHPC Joint Undertaking. A lot of money is going into that direction. Has this any importance for you directly?
Prof. Dieter Kranzlmüller:Indirectly, it is important because it drives technology. If the technology improves, our users will be able to get better technology for their applications but for me it is not that important to be the first one which I think is one of the criteria for EuroHPC.
Primeur Magazine:In EuroHPC, apart from funding machines and systems, it is also about supporting others than just the high-end researchers with with HPC. It is also about supporting SMEs and supporting long tail researchers. Is there any involvement there from your site?
Prof. Dieter Kranzlmüller:The support is always part of our equation. We buy machines and we also establish the relevant infrastructure to support processes and all these kinds of things. That is part of the budget we actually have. That is a very important part and we are grateful that the ministries understand that. From our statutes it is also clear that we only support science at large which means that we cannot give cycles to industry. All we can do is, we can do scientific projects where industry is possibly involved. That is a different kind of approach we have than what EuroHPC wants to do. As far as the long tail of science is concerned, I believe with the wide range of applications we are supporting, we are covering a large part of this long tail of science, even if we don't go down to those that need smaller computing capabilities. At the far end, at the top end, I think our system is the one that does a good chop in that.
Primeur Magazine:Do you expect that in Germany things will change in HPC or are the plans now firmly established and will you continue in the next decade in the same way?
Prof. Dieter Kranzlmüller:The current funding cycle is until 2025 but the ministry has already understood that we need to think about the future. I think you have been there two years ago when Professor Lucas from the ministry announced the Smart Scaling Initiative which is the foundation for a strategy on how to try to zone into the future. I'm happy that our ministry understands that and gives us this task to do but of course, again, it is something we need to work out. We need to work on the details. We need to see how we get there and how we go on after 2025.
Primeur Magazine:Just as a closing remark, is it true that your centre has been featuring in a German television series?
Prof. Dieter Kranzlmüller:Yes, that was just recently. It is called the "Crime Story". "Tatort" is the German name. We got quite some feedback. I think it is also good that people say how these systems look like.
Primeur Magazine:Thank you very much for this interview.