HGX H800 算力服务

支持全新的多实例GPU特性，可将单颗GPU最多分割成7个GPU实例，每个GPU实例可运行不同应用，极大提升了GPU的资源利用率。可同时部署深度学习模型训练、深度学习推理、高性能计算、数据分析等多种应用，性能增强的同时灵活性也得到了前所未有的提升。可支持多达2颗英特尔®至强®第三代可扩展家族处理器，单颗多达40个内核，最大功率270W，可支持16根英特尔®傲腾®持久内存PMem 200系列，最大可支持内存容量高达12TB。通过25块2.5英寸硬盘（可支持多达12块NVMe）的支持能力，用户可获得最高200TB的存储空间。高速计算网络的加持让用户可以部署高性能的AI算力集群，加速人工智能在企业的落地与应用。Supports the brand-new multi-instance GPU feature, which can split a single GPU into up to 7 GPU instances. Each instance can run distinct applications, greatly improving the resource utilization rate of GPUs. It enables simultaneous deployment of multiple applications including deep learning model training, deep learning inference, high-performance computing, and data analysis, delivering unprecedented flexibility enhancements while boosting performance. It supports up to two 3rd Generation Intel® Xeon® Scalable family processors, each with up to 40 cores and a maximum power of 270W. It supports up to 16 Intel® Optane™ Persistent Memory PMem 200 Series, with a maximum supported memory capacity of 12TB. With support for 25 2.5-inch hard drives (up to 12 NVMe drives supported), users can achieve a maximum storage capacity of 200TB. Backed by high-speed computing networks, users can deploy high-performance AI computing clusters, accelerating the deployment and application of artificial intelligence in enterprises.

搜集汇总

背景与挑战

背景概述

HGX H800算力服务支持将单GPU分割为7个独立实例运行不同应用，配备高性能至强处理器、12TB内存和200TB存储空间，适用于AI训练、推理及高性能计算等场景。其高速网络架构可构建AI算力集群，显著提升资源利用率和计算灵活性。