日前，随着深度学习、高性能计算、大模型训练等技术的保驾护航，通用人工智能时代即将到来。各个厂商也都在紧锣密鼓的布局，如英伟达前段时间发布GH 200包含 36 个 NVLink开关，将256个GH200 Grace Hopper 芯片和144TB 的共享内存连接成一个单元。除此之外，英伟达A100、A800、H100、V100也在大模型训练中广受欢迎。AMD MI300X 其内存远超120GB的英伟达GPU芯片H100，高达192GB。
6月22日，英特尔 (lntel) 宣布，美国能源部阿贡国家实验室已完成新一代超级计算机"Aurora"的安装工作。这台超级计算机基于英特尔的CPU和GPU，预计在今年晚些时候上线，将提供超过2 exaflops的FP64浮点性能，超越美国能源部橡树岭国家实验室的"Frontier”，有望成为全球第一台理论峰值性能超过2 exaflops的超级计算机。
Aurora超级计算机是英特尔、惠普(HPE) 和美国能源部(DOE)的合作项目，旨在充分发挥高性能计算(HPC)在模拟、数据分析和人工智能(AI)领域的潜力。该系统由10624个刀片服务器组成，每个刀片由两个英特尔Xeon Max系列CPU (至强Max 9480) 和六个英特尔Max系列GPU组成。
Recently, with the protection of deep learning, high-performance computing, large model training and other technologies, the era of general artificial intelligence is coming. Various manufacturers are also working hard on the layout, such as Nvidia released some time ago that the GH200 contains 36 NVLink switches, connecting 256 GH200 Grace Hopper chips and 144TB of shared memory into a unit. In addition, the NVIDIA A100, A800, H100, and V100 are also popular in large model training. The AMD MI300X has far more memory than the 120GB Nvidia GPU chip H100, up to 192GB.
On June 22, Intel (lntel) announced that the U.S. Department of Energy's Argonne National Laboratory has completed the installation of its next-generation supercomputer, Aurora. The supercomputer, which is based on Intel cpus and Gpus and is expected to come online later this year, will deliver more than 2 exaflops of FP64 floating-point performance, surpassing the U.S. Department of Energy's Oak Ridge National Laboratory's "Frontier "and is expected to become the world's first supercomputer with a theoretical peak performance of more than 2 exaflops.
The Aurora Supercomputer is a collaboration between Intel, Hewlett-Packard (HPE) and the U.S. Department of Energy (DOE) to realize the full potential of high-performance computing (HPC) in the fields of simulation, data analytics and artificial intelligence (AI). The system consists of 10,624 blade servers, each made up of two Intel Xeon Max series cpus (Xeon Max 9480) and six Intel Max series Gpus.
