Prometheus ranked 103rd. in the latest TOP500 ranking

Prometheus, photo by Maciek Bernaś, KSAF AGH UST

The TOP500 list of the world’s most powerful supercomputers was officially announced on 25 June 2018 during the International SuperComputing High Performance Conference held in Frankfurt, Germany. Once again, the fastest Polish supercomputer – Prometheus of CYFRONET AGH UST Academic Computer Centre – made it to the TOP500, ranked 103rd.

Prometheus remains the fastest Polish supercomputer, built by Hewlett-Packard in accordance with design principles developed by Cyfronet experts. It consists of over 2200 HP Apollo8000 platform servers connected by superfast InfiniBand network with  56 Gb/s throughput. The supercomputer has been supplied with over 53000 computational cores (energy-efficient and high-performance Inter Haswell processors) and 282 TB RAM developed in the DDR4 technology. Additionally, Prometheus is supported by two file systems of 10PB capacity (combined) and the access speed of 180 Gb/s. The supercomputer is also accompanied by Nvidia Tesla cards with GPGPU graphics processors. Prometheus can deliver a peak performance of 2.4 PFLOPS.

– To put the speed of Prometheus into perspective: if you wanted to catch up with its amazing performance you would need to use the power of over 50000 top class PC’s in their strongest configuration, additionally connected into a superfast network and managed by dedicated software – said professor Kazimierz Wiatr, Director of ACC CYFRONET AGH UST.

Prometheus is distinguished by its state-of-the-art liquid cooling system, which makes it one of the world’s most energy-efficient computers in its class. The use of dry coolers to maintain proper liquid temperature is perfectly sufficient in our climate, a solution which is considerably cheaper than energy-intensive ice water generators and technologically advanced precision air conditioning (PAC) systems, which consume large amounts of power. Liquid cools both processors and memories using a dedicated air-tight system of heat pipes and direct contact heat transfer between the servers and the liquid flow system. The very servers remain dry – they can be removed at any time, without the risk of the liquid leaking from the cooling system. In addition, liquid-based cooling has enabled very high installation density of as many as 144 computational servers in a single cabinet.

According to ACC Cyfronet Director:
– Thanks to the innovative technology of direct liquid cooling, Prometheus is one of the world’s most energy-efficient computers in its class. Its energy efficiency is comparable  with those of the world’s largest data centres, such as Google or Facebook. What is more, liquid cooling enables extremely high installation density, which makes the computational component weighing over 40 tonnes compact enough to take up as little as 18 m2 of surface and fit into 20 cabinets. To achieve comparable computational performance, its predecessor – Zeus – would require as many as 160 cabinets, while the 1.5 TFLOPS Baribal, which preceded Zeus, was enclosed within 8. Prometheus built in the same technology would need the staggering 12000 cabinets!

The architecture of Prometheus addresses a vast array of research needs as it provides resources organised into partitions: 

  • classic computational servers with high-performance Intel Haswell processors,
  • set of servers with NVIDIAK80 graphics processors accompanied by Intel Xeon Phi 7120P and Nallatech FPGA accelerators,
  • set of servers for performing calculations in the cloud environment.

The supercomputer is used by Polish researchers to address a number of their needs, including: 

  • study results analysis,
  • numerical simulations,
  • advanced visualisations,
  • parallel big data analysis,
  • cloud services and many others.

Prometheus has been assembled in a modern computer facility run by ACC Cyfronet, which was specifically adapted for its regular operation. The proper performance of the supercomputer is secured by many important elements of supporting technical infrastructure, including guaranteed power supply system with a supplementary power generator and advanced  industrial air conditioning and gas extinguishing systems.

Polish supercomputers to have made it to the TOP500

Prometheus aside, the June ranking features three other Polish supercomputers. Below we publish the full list, with their respective computational powers: 

  • 103 – Prometheus  Academic Computer Centre CYFRONET AGH UST (2349 TFLOPS),
  • 259 – Eagle - Poznań Centre for Supercomputers and Networks (PCSS)  (1372 TFLOPS),
  • 263 – Tryton – Academic Computer Centre, Gdańsk University of Technology (1413 TFLOPS),
  • 320 – Okeanos – ICM UW Technology Centre, University of Warsaw, (1082 TFLOPS).

The world’s fastest…

...supercomputer to have been announced this year is – for the first time – the U.S.-based Summit, whose theoretical computational power exceeds 187.6 PFLOPS. Nevertheless, the TOP-500 is visibly dominated by Chinese machines (206), followed by the United States (124). The fastest European supercomputer –  Piz Daint from Switzerland – came 6th.