Search form
Забыли свой пароль?

Официальный интернет-портал Президента Республики Беларусь

Национальная академия наук Беларуси


Грид-сеть высокопроизводительных вычислительных ресурсов Национальной академии наук Беларуси   

Портал рейтинговой оценки качества оказания услуг организациями Республики Беларусь   

БИСРС (Белорусская интегрированная сервисно-расчетная система и биометрические документы)   

Supercomputer systems of hybrid architecture

Supercomputer systems of hybrid architecture

Developer company: The United Institute for Informatics Problems of the National Academy of Sciences of Belarus

Contact persons:
N. N. Paramonov, 284-21-41,

Product brief description:
Supercomputer systems of hybrid architecture offer the following architectural solutions:
1. Tera-scale capacity range.
2. Hybrid meta-cluster architecture with homogeneous computational nodes or cluster architecture with hybrid computing nodes with coprocessors.
3. Computational nodes work on the basis of classic multi-core processors with x86-64 architecture and/or specialized processors (coprocessors) of Cell type, graphics processor- accelerator (GPU), etc.
4. The basic design and technological solutions:

  • Form factors are: 1U, 2U, blade-servers;
  • Posts are: 19 " and up to 44 U high;
  • Air cooling system.

Three models have been made:

  • The "SKIF UIIP" cluster. It represents meta-cluster consisting of two parts: the blade technologies-based cluster (blade-cluster) and the Cell technology-based cluster (Cell-cluster). The Cell-cluster of "SKIF UIIP" supercomputer has 8 dual processor computational nodes based on PowerXCell 8i 3.2Ghz processors. Cell-cluster peak performance is: Rpeak = 1638 Gflops (double); Rpeak = 3276 Gflops (single). Computational nodes of blade-cluster are made on the basis of T-blades dual-processor servers developed by the «T-Platform» Russian company and have 4 Intel XEON E5472 3.0GHz core processors. The total number of computational clusters: 50; the total number of cores: 400. Peak performance of blade-cluster of "SKIF UIIP" supercomputer is 5.0 Tflops. Actual efficiency of "SKIF UIIP" supercomputer prototype according to Linpack test for 52 units (including two control units) on the blade-servers is 4027 Gflops. The Ueff cluster efficiency indicator is 80, 67%. The "SKIF UIIP" cluster takes 47th place in the 14th edition (from 29.03.2011) of Top50 list of CIS the most powerful supercomputers.
  • The "SKIF-GPU” cluster. The "SKIF-GPU» cluster was created in 2010. Its computational nodes and other modules are located in two 19”

44 U-height posts of original design.The "SKIF-GPU» cluster has 34 computational and 2 control nodes. Each node contains of two
4-core Nehalem Intel Xeon X5570 2,93 GHz processors with х86-64 architecture and GPU NVIDIA GeForce GTX 295 graphic engine. The cluster peak performance cluster for graphic engines if making calculations with floating-point of single precision is not less than 60.0 TFLOPS; with that of double precision - not less than 5.0 teraflops. Actual efficiency of
"SKIF-GPU” supercomputer prototype  according to the Linpack test for 36 units without graphic engines is 3025 Gflops. The Ueff cluster efficiency is 89,62% of the peak 3375.36 Gflops. The "SKIF-GPU " cluster takes 45th place in the 13th edition (from 21.09.2010) of Top50 list.

  • The "SKIF-GRID" cluster. The cluster uses advanced technical solutions based on the 12-core AMD Opteron processors with x86-64 architecture and graphic processors-accelerators. "SKIF-GRID" cluster computational nodes and other modules are located in one 19 "44 U-height post of original design. "SKIF-GRID" cluster’s efficiency is the highest among all the Belarusian "SKIF" models. Its peak efficiency (with 36 computational nodes and 2 control units) without taking into account the acceleration through graphic engines is 8.0 teraflops. The Linpack test efficiency indicator is 82,15%. The "SKIF-GRID" cluster takes 40th place in the 14th edition of the Top50 list (29.03.2011).

    Main advantages:
  • high performance with high efficiency;
  • high packing density, small size;
  • low power consumption in comparison with other supercomputers of the same class


Peak efficiency (Tflops)

Ueff  %

Power consumption Kw


8 ,276 (single)
6,638 (double)




63 ,0(single)
8,0 (double)




8,0 (double)



Field of application:
The solution of science-intensive and recourses-intensive tasks in various branches of science. The creation of technological bases for the development of science intensive technologies and their subsequent implementation into different sectors of the economy, such as research; mechanical engineering; oil and gas sector; climate and ecology; environmental management; nanotechnology; biotechnology and pharmaceuticals.

Stages of development:  
The product is in operation.

Forms of cooperation:
1. Delivery, adjustment and commissioning are available on the customer's request. Efficiency indicators are available on the customer’s request.
2. Access to the resources can be provided under commercial contracts.

Supercomputer systems of hybrid architecture

Back to the list