GPU cluster connection method and device, switch and storage medium

By using Leaf switches to interconnect server clusters within a GPU cluster, the connection rules and communication routes within and between clusters are optimized, solving the problems of high communication latency and low training efficiency in existing GPU cluster technologies, and realizing an efficient large-scale GPU cluster networking architecture.

CN120849336APending Publication Date: 2025-10-28CHINA UNITED NETWORK COMM GRP CO LTD +1
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202510962123.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-07-11
Publication Date
2025-10-28

AI Technical Summary

Technical Problem

Existing GPU cluster networking architectures suffer from high communication latency and low training efficiency in large-scale parametric model training. In particular, the two-layer fat tree architecture increases network costs and communication hops, leading to load imbalance and network congestion.

Method used

A novel GPU cluster connection method is adopted to interconnect Leaf switches with the same sequence number from different server clusters, thereby expanding the scale of the GPU cluster. By optimizing the connection rules and communication routes within and between clusters, communication latency is reduced and training efficiency is improved.

Benefits of technology

Without increasing network layers, the GPU cluster size was expanded, communication latency was reduced, model training efficiency was improved, and network costs were saved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120849336A_ABST
    Figure CN120849336A_ABST
Patent Text Reader

Abstract

The invention provides a GPU cluster connection method and device, a switch and a storage medium, relates to the technical field of communication, and is used for solving the problems that a GPU cluster of a single-layer fat tree architecture is relatively small in scale and a GPU cluster of a two-layer fat tree architecture is relatively low in efficiency in related technologies. The GPU cluster connection method comprises the steps that Leaf switches with the same serial number of different server clusters are interconnected, and each server cluster comprises a plurality of Leaf switches and a plurality of servers; each server comprises a plurality of graphic processing units (GPUs); and each switch is in communication connection with the GPU in the server cluster. According to the method, a GPU cluster with a larger scale and higher efficiency can be constructed in a transverse expansion mode.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Cited By

  • Computer system, server optimization method, electronic device and storage medium

    CN121217598A

  • Intelligent computing cluster networking topology method and device, electronic equipment and storage medium

    CN121357190A

  • Artificial intelligence cluster networking topology method and device, electronic equipment and storage medium

    CN121357190B