Consumer Internet Industry HPC Cluster Introduction

Feb 21, 2023

Description

What High-Performence Applications and Needs Are There in Consumer Internet?

The Consumer Internet industry provides internet-based services and products to consumers, including services such as social media, online shopping, streaming media, and other web-based services. Thus consumer Internet companies’ needs for high-performance computing (HPC) are to accelerate business and provide end-users with a better experience. Here are some general uses of the high-performance computing network in the consumer Internet field.

Social Media
Personalized Content Push
Personalized Advertising Services

E-commerce
Customized Shopping Experience
Predictive Fraud Prevention Transactions

Delivery Services
AI Calculates Optimal Paths
AI-Operated Delivery Methods

Video Conferencing
AI Corrects Images
Translate Caption in Real-time

What Software Stacks Are Typically Used in Consumer Internet HPC Cluster?

Except for software stacks including Linux, Apache, MySQL, and PHP (LAMP), as well as other web development frameworks such as Ruby on Rails and Node.js, consumer Internet HPC clusters also choose some distributed computing frameworks and AI-based specialized software such as NVIDIA Riva, NVIDIA Merlin to manage large datasets and process data in parallel. The following software stacks are typically used in consumer Internet to build high-performance general-purpose AI workload clusters.

  • NVIDIA Riva Conversational AI
  • NVIDIA Merlin Recommendation System
  • NVIDIA Rapids Accelerated Data
  • NVIDIA Maxine AI Conference
  • Deep Learning Framework
  • CUDA-X Accelerated Library

What Hardware Are Typically Used in Consumer Internet HPC Cluster?

Consumer Internet HPC clusters data center infrastructure includes servers, storage systems, InfiniBand networking interconnect devices (switches, NICs, cables and transceivers), and other specialized hardware such as GPUs and FPGAs. Popular brands such as Intel, AMD, NVIDIA, and HPE are widely used for HPC computing and storage equipment, while Mellanox was the most commonly chosen, and often the only option, for HPC optical connectivity assemblies. The following NVIDIA high-performance GPU for servers, and Mellanox ConnectX VPI InfiniBand NICs are commonly used to build high-performance GPU clusters for consumer Internet applications.

Need AI-Based Computing Ability?
Recommended GPU Model: NVIDIA A100/A30/A40/A10.

Need Low Latency Access to Storage, and Supports for GPUDirect RDMA?
Use NVIDIA Mellanox InfiniBand NICs.

Need Connectivity Assemblies to Build Low-Latency and High-Performance InfiniBand Cluster?
Use 200Gb/s HDR InfiniBand active optical cables/passive copper cables/transceivers.