Data storage method and device, electronic equipment and readable storage medium

By setting up an accompanying processing module on the data transmission path to the target memory, the problem of data format conversion and packaging processing occupying core computing resources in the existing technology is solved, realizing efficient utilization of core computing resources and optimization of the data processing pipeline, thereby improving the overall throughput.

CN122285084APending Publication Date: 2026-06-26SUZHOU YIZHU INTELLIGENT TECH CO LTD

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SUZHOU YIZHU INTELLIGENT TECH CO LTD
Filing Date
2026-03-31
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

In existing technologies, the data format conversion and packaging processing tasks after model quantization consume core computing resources, leading to increased EU computing load, idle computing units, and decreased overall throughput, making it impossible to effectively utilize hardware resources.

Method used

By setting up an accompanying processing module on the data transmission path to the target memory, data format conversion and packaging processing are handled by the accompanying processing module, freeing up the burden on the core computing unit, processing data conversion and packaging tasks in parallel, and optimizing the data processing pipeline.

Benefits of technology

It improves the utilization rate of core computing resources, enables deep parallelism at the task level, optimizes the data processing pipeline, increases overall throughput, and avoids data processing bottlenecks and resource waste.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122285084A_ABST
    Figure CN122285084A_ABST
Patent Text Reader

Abstract

This application discloses a data storage method, apparatus, electronic device, and readable storage medium, relating to the field of data storage technology. The method includes: in response to a converged storage instruction, acquiring source data in a first format; on the transmission path where the source data is written to target memory, converting the source data into target data in a second format using an accompanying processing module configured on the transmission path, and packaging multiple target data sets into combined data matching the access granularity of the target memory; and writing the combined data into the target memory. This application has the effects of reducing the computational load on the EU (Entity Memory), improving the utilization rate of core computing resources, optimizing the data processing pipeline, and achieving deep task pipeline parallelism.
Need to check novelty before this filing date? Find Prior Art