The Supercomputer Centre is to be upgraded

As a result of the developments implemented in several phases during the past two decades – the performance of the first supercomputer installed on 21 March 2001 corresponds to a that of better-quality smartphone today (60 gigaflops) – and of the upgrades, KIFU now operates an HPC infrastructure with a total performance of almost 0.5 petaflops (420,000 gigaflops). In the past 20 years, there has been a more than 7000-fold increase in capacities. 6 years ago, this was considered extremely good, but now it cannot satisfy the constantly increasing demand anymore; therefore, KIFU is expanding its infrastructure ten times with the new supercomputer having a performance of 0.5 petaflops.

The latest Hungarian supercomputer Komondor is scheduled to arrive in the end of 2021; however, the building which will accommodate the equipment at the campus of the University of Debrecen and its surroundings need significant investments to be prepared for this. The preparation of KIFU’s Supercomputer Centre will start in June 2021 because electricity, engineering, and architectural developments are all necessary.

In 2014, Ybl Prize- (2013) and Pro Architectura Prize- (2009) winning architect Marcel Ferencz specifically designed a two-storey building with a footprint of 400 m2 to host supercomputers, and its facade features electric circuits. As regards size, Komondor is not larger than an older machine given that not only performance but also performance density is on a steady rise; therefore, the new machine with ten-fold performance can be hosted by the same footprint size as its predecessor. However, the associated infrastructure requires space and unique technical and engineering solutions.

One of the most significant challenges is satisfying the electricity needs because our supercomputers are typically under maximum usage. For a stable and safe operation in the long run, high-voltage power supply independent of the university network because the electric input of the machine room may increase to as high as 1.3 MW/hours, which is equal to the electricity needs of Jászfényszaru, a town with a population of 5600. Of course, the final consumption rate and the final configuration will depend on external temperatures and on the usage rates. This level of performance already includes a significant standby to ensure that potential technological extensions may be implemented without a major remake even as regards cooling.

“Green” technologies are of increasing importance, and the largest supercomputers in the world have also been ranked by energy efficiency since July 2013. It is worth mentioning that the 552-petaflops supercomputer LUMI being constructed in Finland under the aegis of the EuroHPC will be 100% renewable-based, and will also provide 20% of the district heating of the region.

Environmental considerations were also taken into account by the experts during the renewal of KIFU’s Supercomputer Centre.

The huge cooling needs of Komondor will be satisfied using direct hot water cooling with a technology applied only by a few in Hungary. The 7-ton cooling unit to be installed on the roof will cool the circulating cooling liquid to a temperature of 32°C, as required by the HPC equipment, via circulating the outdoor air and – if necessary – using water spray technology. Instead of cooling down the waste heat of the hot water emitted by the HPC at a temperature of more than 45°C, the University of Debrecen can make direct use of it. This has a double benefit: reducing the energy costs of the Debrecen Sports Pool, but also the cooling costs of the supercomputer by draining the heat, in addition it is also an extremely environmentally friendly, forward-looking solution. Komondor’s cooling will be provided by a circular system which enables the shut-off of malfunctioning sections during potential repairs and maintenance, and the availability of the supercomputer may thus remain uninterrupted. For the cooling of traditional components, the cold water systems of the machine room remain in place.

In the same way, ensuring an uninterrupted operation is also a high priority due to a need to maximise availability. Power outages should be immediately remedied by switching to battery supply for not only the machine itself but also for the associated infrastructure, including the cooling system. If in turn, the room-sized batteries run down, a diesel engine-operated electric generator is switched on. In case the power supply cannot be restored even during the time thus gained, the supercomputer enters the emergency operation phase, followed by a gradual shutdown. Instead of the previous container-based arrangement, the power supply infrastructure will now be installed inside the new group of buildings.

Altogether, the preparations for the installation of this a supercomputer, which provides ten times the previous capacities, are progressing according to schedule, and Komondor can take its expected place – around No. 60 – on the top500.org ranking already in 2022.