单核性能提升55%,阿里云发布基于第四代AMD EPYC处理器的弹性计算实例

简介: 11月11日,阿里云宣布,将推出新一代基于AMD EPYC处理器的计算实例规格族g8ae。该款计算实例基于阿里云自研CIPU与磐久服务器,搭载AMD公司最新发布的第四代AMD EPYC(霄龙)处理器,单核性能相比上一代g7a实例提升55%以上;网络带宽升级至2 x 100G,提升100%,标配阿里云自研eRDMA大规模加速能力;全面搭载NVMe,支持共享盘;支持虚拟可信能力vTPM与安全加密虚拟化SEV等特性。

11月11日,阿里云宣布,将推出新一代基于AMD EPYC处理器的计算实例规格族g8ae。该款计算实例基于阿里云自研CIPU与磐久服务器,搭载AMD公司最新发布的第四代AMD EPYC(霄龙)处理器,单核性能相比上一代g7a实例提升55%以上;网络带宽升级至2 x 100G,提升100%,标配阿里云自研eRDMA大规模加速能力;全面搭载NVMe,支持共享盘;支持虚拟可信能力vTPM与安全加密虚拟化SEV等特性。
On November 11th, Alibaba Cloud announced that they will launch a new generation of Compute Instance Type Family, g8ae, based on AMD EPYC Processors. This compute instance is based on Alibaba Cloud self-developed CIPU and Panjiu (磐久) Servers. It is equipped with the 4th Gen AMD EPYC processors recently launched by AMD. Compared with the previous generation of g7a instance, the single-core performance has improved by more than 55%. The network bandwidth is upgraded to 2 x 100G, increased by 100%, and the large-scale acceleration capability of eRDMA developed by Alibaba Cloud is standard. Fully equipped with NVMe and support for shared disks, Virtualizing the Trusted Platform Module (vTPM) and Secure Encrypted Virtualization (SEV).

阿里云弹性计算产品专家表示,g8ae实例将为游戏、电商、视频直播、内容平台、金融、高校科研机构,以及芯片设计等行业客户提供优质高效的算力服务,满足其在AI推理与训练、深度学习、高性能数据库等应用场景下日益增长的性能需求。
Alibaba Cloud elastic computing product experts said that the g8ae instance is designed to offer high-quality and high-efficient computing services to the customers in gaming, e-commerce, livestream, content platforms, finance, university research institutions, chip design and other industries, meeting the increasing performance requirement of the applications scenarios in AI reasoning and training, deep learning, high-performance database.

性能的巨大跃升,离不开基于“Zen 4”核心架构的第四代AMD EPYC处理器。它可以为严苛的工作负载提供极具突破性的创新和性能。第四代AMD EPYC处理器的CPU拥有多达96个核心并支持高达6TB内存,兼容全新的SP5插槽,并加入对高性能DDR5 DIMM和第五代高速PCIe I/O的支持,可满足当今对AI和ML应用程序的需求,以及对加速器、GPU、FPGA等日益增加的使用要求。
The huge jump in performance is due to the 4th Gen AMD EPYC processors based on the "Zen 4" core architecture. It provides radical innovations and performance for demanding workloads. The 4th Gen AMD EPYC processors have up to 96 cores and support up to 6TB of memory, compatible with new SP5 socket, and add the support of the high-performance DDR5 DIMMs and the fast PCIe Gen 5 I/O, which enables the demands of today’s AI and ML applications and the increasing use of accelerators, GPUs, FPGAs and more.

AMD全球副总裁、中国区企业与商用事业部总经理刘宏兵表示:“随着’Zen 4’核心的推出,我们正在从7nm制程工艺迈向5nm制程工艺节点。所有这些创新使我们可以用低功耗提供令人叹服的卓越性能。我们的第四代AMD EPYC处理器可以为阿里云g8ae实例提供业界领先的性能,帮助客户更快地实现业务成果。AMD将与阿里云携手,用创新的产品持续为客户创造价值。”
Liu Hongbing, AMD Corp VP and General Manager of Enterprise and Commercial Business in China, said that “with the ‘Zen 4’ core, we are moving from 7nm to 5nm process node. All of these innovations allow us to deliver impressive performance with low power consumption. Our 4th Gen AMD EPYC processors deliver industry-leading performance for Alibaba Cloud g8ae instances, helping customers achieve business results faster. AMD will join hands with Alibaba Cloud to continuously create value for customers with innovative products.”

随着越来越多的企业迁移上云,基础设施的安全保障和数据安全也持续受到关注。g8ae实例全量搭载安全芯片作为硬件可信根,实现服务器的可信启动,确保零篡改;虚拟化层面,支持虚拟可信能力,提供实例启动过程核心组件的校验能力;不仅如此,g8ae搭载的第四代AMD EPYC处理器也配备了完整的“Security by Design”体系,包括一套先进的安全功能集,可以帮助保护其最宝贵的数据资产。g8ae为新金融、大型互联网等业务场景提供了更高安全等级数据保护能力和云上可信运行环境。
As more and more enterprises move to the cloud, infrastructure security and data security continue to be a concern. The g8ae instance is fully equipped with the security chip as the hardware trusted root to realize the trusted startup of the server and ensure zero tampering. In terms of virtualization, the capability of Virtualizing the Trusted Platform Module is supported to verify core components during instance startup. Not only that, the g8ae instance's 4th Gen AMD EPYC processor also comes with a complete "Security by Design" system, including a set of advanced security features to help protect its most valuable data assets. The g8ae instance provides higher level security data protection capabilities and a trusted operating environment on the cloud for business scenarios such as new finance and large-scale Internet.

通过业界领先的能力,阿里云弹性计算云服务器的计算、存储、网络能力将持续升级,让客户持续享受最新技术红利,支撑数字经济的高质量发展。
Through industry leading capabilities, the computing, storage, and network capabilities of Alibaba Cloud elastic computing cloud servers will continue to be upgraded, allowing customers to continue to enjoy the latest technology dividends, and supporting the high-quality development of the digital economy.

相关文章
|
8月前
|
弹性计算 数据库 云计算
阿里云服务器如何购买价格更便宜?领取优惠券便宜购买图文教程参考
本文重点介绍2026年阿里云推出的多元优惠券体系。包括学生专享300元无门槛券、企业迁移算力补贴、新用户满减券(满100减10元)、165元券包(满50减15至满350减100)及系统赠券等。通过实操案例展示阿里云目前优惠券的领券路径与使用流程,如经济型e实例2核8G配置通过150元券实现757.39元至607.39元的减免,计算型c9i实例4核8G配置通过150元券实现3147.56元至2997.56元的优惠。
阿里云服务器如何购买价格更便宜?领取优惠券便宜购买图文教程参考
|
3月前
|
人工智能 安全 前端开发
来不及学 AI 就不用学了?——从 AI 到 Agent,再到 Harness Engineering 与 Loop Engineering :一条被产品化加速的技术演进线
本文是一份AI工程演进图谱,以「概念地图」形式梳理2022–2026年LLM→Agent→Harness→Loop的演进脉络,厘清OpenClaw、MCP、Skills等核心概念的关系,破除术语迷雾,强调Harness Engineering才是当前最值得投入的系统性工程能力。(239字)
477 1
|
机器学习/深度学习 人工智能 API
大模型推理服务全景图
国内大模型推理需求激增,性能提升的主战场将从训练转移到推理。
3943 144
|
SQL XML Java
【mybatis】第二篇:@Select注解中加入字段判断
【mybatis】第二篇:@Select注解中加入字段判断
|
人工智能 监控 测试技术
阿里云磐久服务器稳定性实践之路
阿里云服务器质量智能管理体系聚焦自研服务器硬件层面的极致优化,应对高并发交付、短稳定性周期、早问题发现和快修复四大挑战。通过“三个重构”(质量标准、开发流程、交付模式)、“六个归一”(架构、硬件、软件、测试、部件、制造)策略,实现芯片、整机和云同步发布,确保快速稳定上量。此外,全场景测试体系与智能预警、分析、修复系统协同工作,保障服务器在萌芽阶段发现问题并及时解决,提升整体质量水平。未来,阿里云将继续深化大数据驱动的质量管理,推动服务器行业硬件质量的持续进步。
|
机器学习/深度学习 算法 PyTorch
10招立竿见影的PyTorch性能优化技巧,让模型训练速度翻倍
本文系统总结了PyTorch性能调优的关键技术,涵盖混合精度训练、PyTorch 2.0编译功能、推理模式优化、Channels-Last内存格式、图优化与变换、cuDNN基准测试、内存使用优化等多个方面。通过实证测试,文章详细分析了各技术的实现细节、优势及适用场景,如混合精度训练可显著提升计算效率和内存利用率,torch.compile()能自动优化代码生成以加速模型运行。此外,还探讨了推理模式的选择、卷积操作优化及模型构建的最佳实践。这些方法结合良好的编码习惯,有助于开发者构建高效、可扩展的深度学习应用。
1324 3
10招立竿见影的PyTorch性能优化技巧,让模型训练速度翻倍
|
Java Shell 应用服务中间件
Mac系统下配置环境变量:Javajdk、maven、tomcat 环境变量配置及对应配置文件
这篇文章介绍了如何在Mac系统下配置Java JDK、Maven和Tomcat的环境变量,包括配置文件的选择、解决环境变量在zsh shell中无效的问题、查看和设置系统环境变量的方法,以及JDK和Maven的下载、配置和测试步骤。
7410 1
Mac系统下配置环境变量:Javajdk、maven、tomcat 环境变量配置及对应配置文件
|
存储 人工智能 算法
阿里云AI基础设施升级亮相,模型算力利用率提升超20%
阿里云AI基础设施升级亮相,模型算力利用率提升超20%
1697 18

热门文章

最新文章