Data transmission method and apparatus, electronic device, storage medium, and program product
By determining the target communication path between GPUs and utilizing the shared cache of intermediate GPUs for data transmission, the problem of low data transmission rate caused by insufficient bandwidth between CPUs is solved, thus improving data transmission efficiency.
CN122152731APending Publication Date: 2026-06-05DAWNING INT INFORMATION IND CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DAWNING INT INFORMATION IND CO LTD
- Filing Date
- 2024-12-03
- Publication Date
- 2026-06-05
AI Technical Summary
Technical Problem
When two central processing units (CPUs) are interconnected using an inter-chip global memory interconnect (XGMI) bus, the data transfer rate is low, resulting in insufficient bandwidth.
Method used
By acquiring the GPU topology map, the target communication path is determined to be the path through which the first GPU communicates with the second GPU via the first target switch, the intermediate target GPU, and the second target switch, avoiding data transmission between CPUs and utilizing the shared cache of the intermediate target GPU for data transmission.
Benefits of technology
This increases the data transfer rate and avoids traffic between GPUs passing through the CPU, thus allowing the limited bandwidth between CPUs to be used by more services.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122152731A_ABST
Abstract
The application provides a data transmission method and device, electronic equipment, storage medium and program product. The method comprises the following steps: acquiring a GPU topology graph; the GPU topology graph comprises a plurality of communication paths; the communication path is a path for communication between a first GPU and a second GPU; a target communication path is determined from the plurality of communication paths; wherein the target communication path is a path for communication between the first GPU and the second GPU through a target first switch, a target intermediate GPU and a target second switch; when there is target data to be transmitted, the target data is transmitted from the first GPU to the second GPU through the target communication path. The method of the application improves the data transmission rate.
Need to check novelty before this filing date? Find Prior Art