计算资源服务

国家超级计算太原中心是经科技部认定的全国14个超算中心之一，主机“太行一号”双精度浮点运算能力达300P，即每秒运算30亿亿次，人工智能半精度运算能力达800P，拥有7200个异构计算节点和1600个同构计算节点，采用14nm二代国产X86CPU和7nm二代DCU深度计算单元异构架构设计建设，CPU核心总数332800个，DCU卡28800块，存储容量300PB，高速计算网络单向通信带宽200Gb/s，采用全新一代全浸没相变液冷技术，核心机房PUE值突破性降至1.04，做到低能耗高性能运行。

National Supercomputing Center Taiyuan is one of the 14 national supercomputing centers recognized by the Ministry of Science and Technology of China. The host system "Taihang No.1" features a double-precision floating-point computing performance of 300 petaflops, equivalent to 30 billion billion operations per second, and an AI-oriented half-precision floating-point computing performance of 800 petaflops. It is equipped with 7200 heterogeneous computing nodes and 1600 homogeneous computing nodes, and is constructed with a heterogeneous architecture adopting second-generation domestic 14nm X86 CPUs and second-generation 7nm DCU deep computing units. The total number of CPU cores is 332,800, and there are 28,800 DCU cards. Its total storage capacity reaches 300 petabytes, with the unidirectional communication bandwidth of its high-speed computing network standing at 200 Gb/s. It adopts a new-generation fully submerged phase-change liquid cooling technology, and the Power Usage Effectiveness (PUE) value of its core machine room has been drastically reduced to 1.04, achieving low-energy-consumption and high-performance operation.

搜集汇总

背景与挑战

背景概述

国家超级计算太原中心配备'太行一号'主机，具备300P双精度浮点运算能力和800P人工智能半精度运算能力，采用国产异构架构设计，拥有数万计算节点和300PB存储容量，并通过液冷技术实现1.04的超低PUE值。