![]() |
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ИНХС РАН |
||
Due to the wide variety of HPC architectures and the tasks solved on them, a problem of finding the optimal architecture of supercomputers for each class of tasks arises, since, depending on the system, different tasks can be solved with varying degrees of efficiency. Thus, there is a need to compare supercomputer systems. To do this, we need to build a universal model that can objectively reflect all the characteristics of all supercomputers and systems that will be built in the future[1]. To build such a model, the structure of the description of supercomputers of the Top50 rating was taken as a basis. The current model of the description of the Top50 has been detailed. The descriptions of interconnects between nodes and in the computing group, processor, accelerator, RAM, and performance of system elements have been updated. In the future, it is planned to supplement and refine the model, taking into account the consideration of other systems from the Top50 and TOP500 ratings and its further detailing, as well as to develop a prototype of tools for sampling and comparing systems.