Updated TOP50 (rating of CIS supercomputers): dynamics and trends

The computing power race is an ongoing process that shifts the ceiling of modern computer capabilities a little higher every day. On some days, however, the results of this background work appear in especially vivid form. Last week, an updated version of the rating of supercomputers of the CIS countries appeared on the Web, created by the joint efforts of the Lomonosov Moscow State University Research Computing Center and the RAS Interdepartmental Supercomputer Center. Under the cut you will find a brief overview of changes and general trends, created on the basis of materials from the TOP50 website .

To get started, let's go over the main events and summarize the past six months:

• – «», SberCloud ( « ») NVIDIA, .
• , . Hewlett Packard Enterprise 15 , PetaNode 1.0 Cluster, « » «», 37- .
• : « — » «-10 2» , «» Hewlett Packard Enterprise « ». , 10 ( 22 12 ), – ( 26 18 )
• 7,2 PFlop/s – 20 PFlop/s 12.8 PFlop/s 2019 . 20.9 PFlop/s 30.6 PFlop/s.
• : 50 56.84 TFlop/s Linpack – 3.3 TFlop/s.

Now dwell on some details. Naturally, those projects that entered the TOP50 for the first time cannot but arouse interest - let's look at their characteristics and capabilities.

As already mentioned, the Christofari supercomputer took the rating by storm: its performance on the Linpack test showed a result of 6.7 PFlop / s, and peak performance - 8.8 PFlop / s. If we take into account that the next Lomonosov-2 system from T-Platforms company has these figures of 2.5 PFlop / s and 4.9 PFlop / s, respectively, we can assume that Kristofari is firmly established in the leadership position. The technological base of the project is comprised of NVIDIA DGX-2 computing nodes equipped with Tesla V100 graphics accelerators. The supercomputer is designed to work with artificial intelligence algorithms from different problem areas (medical diagnostics, face recognition, voice biometry). In the TOP500 world ranking, Cristofari takes 29th place.



The first "newcomer" is the Christofari supercomputer

The following numbers are given for the UITiS NOVATEK STC computing cluster from Hewlett Packard Enterprise: 273.28 TFlop / s on the Linpack test at 496.87 Tflop / s peak performance. High power is achieved through the use of the first generation Xeon Scalable processor line. The purpose of this system makes it stand out among other participants: the supercomputer was created specifically for solving research and applied problems from the field of geophysics (extracting knowledge, processing large amounts of textual information, modeling processes and phenomena of the oil and gas industry and more).



The second of the "new" - the UITiS cluster NOVATEK STC

The supercomputer “PetaNode 1.0 Cluster” manufactured by Computer Ecosystems and TechnoCity has a performance of 79.1 Tflop / s on the Linpack test and a peak performance of 141.47 Tflop / s. A distinctive feature of the development is the use of accelerators from AMD at the rate of 10 GPUs per node. The main purpose of this supercomputer is climate modeling - a niche little mastered by the TOP50 participants.



The latest novelty is the PetaNode 1.0 Cluster supercomputer.

In general, there have been changes in the ranking due to debuts and updates. Due to the new leader of the list and two other newcomers, the positions of the remaining supercomputers from the previous TOP50 list lost from 1 to 5 points in the rating.

With regard to the technical equipment of the total mass of projects, the changes are not radical, but tangible. The Intel processor remains an indispensable attribute of the latest generation supercomputer - at the moment, there are no alternative solutions in the ranking. InfiniBand network (33 cases versus 35) and more often - 10 Gigabit Ethernet (eight cases against six) became less commonly used for communication between nodes.

It should be noted separately that two of their three newcomers to the rating (Kristofari and PetaNode 1.0 Cluster) set a clear trend for the transition, if not return, from building heterogeneous and hybrid clusters to homogeneous systems using graphics accelerators. This trend seems to be driven by criteria for energy efficiency and density.

Applications for supercomputers, for the most part, remain unchanged. The sphere of science and education remains the most relevant - it is served by 23 systems, while in applied research 10 supercomputers are used (three less than in the previous version of the top). The industry is still poorly covered by the latest technological advances - only three out of fifty machines work for its needs. In addition, the ranking featured projects designed for tasks such as seismic forecasts, IT services, cloud storage, servicing the commercial and financial sectors.

Finally, Hewlett-Packard Enterprise, RSK group of companies and T-Platforms remain the most active suppliers of supercomputers of the highest standard.
In total, their contribution to the TOP50 is 35 projects (13, 11 and 11, respectively). The remaining positions are also mainly divided between well-known players.

The next time the rating will be reviewed and updated by September 21 of this year - we are waiting and making bets.