跳转到主要内容

Accelerating Your Discoveries

Supermicro 以从零开始构建 HPC 解决方案为荣。从可定制的设计到实施,Supermicro 的专职团队将根据您的要求提供优化的即插即用服务。灵活应对 HPC 最困难的课题,同时也简化了超级计算机的部署。

Supermicro - Accelerating Your Discoveries with General Purpose Compute Intensive

General Purpose Compute Intensive

Supermicro in Scientific Research Supercomputer

Scientific Research Supercomputer

Supermicro building HPC solutions-AI/ML/DL GPU Intensive

AI/ML/DL GPU Intensive

What is High Performance Computing?

High Performance Computing (HPC) is a computational domain that aims to solve complex problems through parallel processing. An HPC system of today consists of hundreds to thousands of CPUs connected through a high speed network. A large HPC cluster of servers can be assigned to work on a single large problem simultaneously or several smaller problems. Many HPC applications are currently written to simultaneously take advantage of tens, hundreds, and thousands of cores, resulting in orders of magnitude lower time to solution. Supermicro designs a range of high performance computing solutions to fit every need. Supermicro's HPC solutions are expertly designed, built, and tested by Supermicro professionals. Liquid cooling is available at rack scale for environments where liquid cooling is required.

Many fields can use HPC technologies, including:
  • Engineering - designing and optimizing new physical products, including automobiles, planes, structures, and consumer appliances.
  • Scientific Research - basic and applied research to model climate change, more accurate weather forecasts, galaxy and star formation, and weapons modernization.
  • Finance - fast decisions based on very low latency calculations to make trading decisions
  • Defense - create battlefield scenarios and produce optimized weapons (engineering)
  • Healthcare - design new drugs, create personalized care, and recognize disease causes and possible contributors to these diseases.

The Supermicro HPC Solutions portfolio consists of HPC servers, HPC storage solutions, HPC networking, and life cycle monitoring software. Together, these technologies all contribute towards an entire high performance computing solution that addresses the needs of HPC users.

HPC 解决方案

Supermicro 参考体系架构旨在满足您独特的 HPC 需求。我们的优势包括广泛的构建块,从主板设计到系统配置,完全集成的机架和液体冷却系统。我们专注于为客户提供量身定制的解决方案。

企业 HPC 参考设计

此多功能参考体系结构是为通用计算密集型工作负载而构建的,非常适合于企业HPC用例,如数字制造、金融风险管理、医学研究、石油和天然气勘探等。

  • 服务器: SuperBlade® SBI-420P-1C2N, 每个机架最多80个节点
  • 多达 6400 个处理器核(Intel® Xeon®), 320TB DDR4-3200,每个机架 160 个 2.5 英寸驱动器托架
  • 使用内置 IB HDR 和 25G 交换机模块减少布线
  • 热插拔节点、交换机模块、PSU 和风扇模块具有良好的可维护性
  • 高度可扩展,在 non-blocking IB 结构中,每个群集最多 10 个机架和 800 个计算节点

科学研究超级计算机参考设计

经过验证的高性能计算机设计为许多科学研究项目而建造,这种超级计算机设计在天气模拟、核反应物理模拟、基因测序、地球和空间发现等方面具有高度的可扩展性。

  • 服务器: 如 16 TwinPro® SYS-220TP-HTTRBigTwin® AS -2124BT-HTR, 每个机架 64 个节点
  • 支持 AMD EPYC™ 高达 8192 核 (BigTwin)或每个机架配备 Intel Xeon 5120 核(TwinPro)
  • 可选 DLC 液体冷却,40KW CDU 至 80KW
  • 极易扩展,在 2:1 OPA 结构中,每个群集最多 30 个机架和 1920 个节点
  • TwinPro 的每节点一个 socket 设计平衡了每种性能的成本,BigTwin 的每节点双 socket 设计实现了不折不扣的性能。

AI,深度学习 HPC 参考设计

我们的可扩展 AI+HPC 即插即用设计具有最新的突破性 AI 功能,能够释放前所未有的科学发现速度,让企业专注于构建改变游戏规则的 AI 应用程序,实现自动驾驶、自然语言处理等。

  • 服务器: 例如 5 个 4U 8-GPU 系统, SYS-420GP-TNARAS -4124GO-NART, 带 NVIDIA® HGX A100 和 NVLink®
  • 多达 80 个 NVIDIA A100 GPU,每个机架具有 40 PB 的 AI 性能
  • 每个机架最多支持 40TB DDR4-3200 和 30 个热插拔 Gen4 NVMe 驱动程序
  • 高速 GPU 互连,每个节点 8 个 200G IB HDR+2 个 100G 以太网端口
  • 高度可扩展性,每个群集最多 100 个节点和 800 个 A100 GPU,采用非阻塞 IB HDR 结构,可选择 D2C 液体冷却

机架集成和部署服务

一系列渐进的步骤,以确保技术和产品满足客户独特的 HPC 要求,并顺利部署。

1. Design

  • Application Analysis
  • Power Budget
  • BOM Creation
  • Rack Layout

2. Assembly

  • Node Assembly
  • Rack & Stack
  • Cabling & Labeling
  • Networking & Power

3. Configuration

  • BIOS Setting Check
  • Firmware Management
  • Switch & IP Addresses
  • OS & Customer Image

4. Testing

  • Multi-vendor Interoperability
  • Full Rack Burn-in & QA
  • Full Rack Test Report
  • Performance Report

5. Logistics

  • Asset Labeling & Docs
  • Crating
  • Air-Ride Trucking
  • White Glove Services

成功案例

Supermicro 可以定制HPC解决方案,以满足各种工作负载:计算密集型、高吞吐量 GPU,或用于不同行业的高容量存储应用程序。Supermicro HPC 解决方案可以与各种开源平台和商业应用程序捆绑在一起,并取得了成功。

University of Tsukuba, “Pegasus”

The Center for Computational Sciences at the University of Tsukuba is a multidisciplinary center that partners with many organizations on basic research in several areas. As part of an ongoing project to deliver state-of-the-art computing capabilities to a wide range of researchers, the University of Tsukuba, with NEC (as the prime contractor), has created a supercomputer that will address the computing and storage needs using Supermicro SuperBlade. The system is among the first in the world to use NVIDIA H100 GPUs and 4th Gen Intel® Xeon® Scalable processors.

大阪大学 "SQUID"

大阪大学与 NEC 公司和 Supermicro 紧密合作,设计并实现了一种基于 Supermicro SuperBlade® 组合的超级计算类系统和4U 8-GPU NVIDIA NVlink® 系统,采用第 3 代 Intel Xeon 可扩展处理器构建。SQUID 系统是一个 27 机架群集,包含1520 个 SuperBlade,超过 120000 个内核,42 个 4U 8 GPU NVIDIA NVLink 系统节点,并由 Supermicro 机架集成服务团队进行了集成和测试。

IMS "Molecular Simulator"

日本分子科学研究所(IMS)已经安装了两个名为“高性能分子模拟器”的集群,其 Supermicro 服务器通过 Intel Omni-Path 架构互连。分子模拟器通过 40588 核 Intel Xeon 处理器和 216768GB 内存,运行分子动力学的大规模并行计算和许多量子化学要求更高的串行操作,实现了 1.8petaflops Linpack 和 3.1petaflops 理论峰值性能。

LLNL “Corona”

Lawrence Livermore National Laboratory(LLNL)选择 Supermicro 为科学家提供额外的计算能力和资源,帮助他们找到治愈 COVID-19 的方法。Supermicro 的 4U 8 GPU 服务器为这个即插即用集群配备了近 1000 个 AMD Radeon Instinct™ 图形加速器,结合AMD的 state-of-the-art EPYC™ CPU 扩展集群,为高级计算工作负载提供高达 11Petaflops 的计算能力。