This white paper describes the architecture, design details, and configuration for an AI reference design based on Intel® products. The reference design was developed for the China National AI Innovation Competition.
1.1 Platform Introduction
The China AI Open Innovation Platform was built on Intel AI hardware platforms and comprehensive solutions, including the following aspects:
AI Hardware Infrastructure
We constructed an AI hardware infrastructure consisting of high performance servers with Intel® Xeon®Platinum processors and Intel®Xeon Phi™ technology, Intel® OmniPath Architecture (Intel® OPA) 100Gb high speed network interconnection devices, Intel SSD high speed flash storage devices, and security devices, to meet the needs of massive data access, storage, and intelligent processing, and to satisfy requirements for safe and reliable operations.
AI Cloud Platform
By adopting open source cloud platform architecture, including KVM*, Docker*, OpenStack*, Kubernetes*, CEPH*, and GlusterFS* distributed storage architecture, we performed virtualization for hardware facilities to form a virtual-level resource pool system. Each application system was provided with basic IT resources on demand, including computing capability, storage capacity and networking capability to quickly adapt to dynamically changing business requirements and achieve resilient resource allocation capabilities.
Using AI development frameworks optimized for Intel platforms, including Intel® Optimization for Caffe*, TensorFlow*, and others, we achieved high performance, high scalable and distributed AI training frameworks. The training frameworks used AI libraries and tools optimized for Intel platforms, including Intel® MKL, Intel® MKL-DNN, and BigDL, to maximize the performance of hardware platforms
A unified Web interface provided central management for all the hardware resources, including virtual machines, resource pools, networks, storages, containers, bare metals, AI development frameworks, and others, which then allocated various types of software and hardware resources for AI development.
AI Development Suite Environment
We integrated AI suites, including Intel FPGA development boards, Intel®Movidius™ compute sticks, desktop computers, peripherals, and others, to build an AI development suite environment. The environment enabled the development teams to develop AI inference sides, mobile and low-power scenarios, and display their results. The environment also provided front-end equipment support for the residing platform teams.
China's AI platform open innovation platform functional architecture