Home > News content

1300 trillion performance Huawei self-research Kunpeng Super calculation in Shenyang Pneumatic Engineering Institute

via:博客园     time:2019/10/13 14:11:21     readed:192

Author: Xianrui

According to local media in Shenyang,The 1300 trillion high performance computing platform independently developed by Huawei Company landed in Shenyang.In the Aeronautical Industry Pneumatic Institute (hereinafter referred to asIt uses the Kuipeng 920 processor developed by Huawei.It will be used in the design, research and verification of computational fluid dynamics in pneumatic institutes, which is expected to greatly enhance the independent innovation ability of the existing aviation systems in China.

Fluid mechanics needs enormous computing power. The introduction of 1000 trillion times over-computing platform in this institute meets its needs in wind tunnel test, computational fluid mechanics engineering calculation, test data analysis, research and calculation of pre-research topics, and jointly develops Huawei in the development of computational fluid mechanics software for aviation independent research and development, and the upgrading of high-performance computing technology. For new research, both sides will jointly promote the incubation of independent computational fluid dynamics software.

Although the report did not mention the specific model of Huawei's over-calculation,However, this super-calculation uses the Kunpeng 920 processor developed by Huawei.With 64 cores, 2.6GH main frequency and 8-channel DDR4 memory, compared with the competing memory bandwidth increased by 46%, total I/O bandwidth increased by 66%, and network bandwidth increased by 4 times. At the same time, it includes two 100G RoCE ports, supporting PCIe Gen4 and CCIX.

In the SPECint benchmark, the Dapeng 920 score is more than 930 points, nearly 25% higher than the industry benchmark, while the power consumption is reduced by 30%.

The performance of 13 billion times is 130 million times FLOPS, which can rank around 300 bits in the global TOP 500 overrun, but most of the previous overruns use X86 or Power architecture. In the ARM overrun, 13 billion times are already very high level. At the end of last year, the Astra overruns built by HPE for Sandia National Laboratory in the United States also use the ARM processor. 1500 trillion times, entering TOP 500 for the first time.

Compared with X86 ecosystem, ARM lacks ecosystem in the field of high-performance universal computing. Hu Houkun, vice chairman of Huawei, also announced Huawei's general processor business strategy at the previous Huawei Full Connection Conference. According to him, in the field of general computing, Huawei invested in Kuipeng series to build competitive general computing processors.

On this basis,Huawei is continually investing in key technologies and products such as board cards, servers, operating systems, databases and compilers, hoping to break through the whole ecological chain through strong investment.Let our partners be more confident and firmly participate in building Kunpeng industry.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments