Smart storage device, computing system, and method of operating computing system
By utilizing the cache consistency mechanism of accelerator and nonvolatile memory in intelligent storage devices, the problems of low bandwidth and long delay between the host device and the memory device are solved, and efficient artificial intelligence model learning and training are achieved.
CN120295954APending Publication Date: 2025-07-11SAMSUNG ELECTRONICS CO LTD
Patent Information
- Application Number
- CN202411105077.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-01-11
- Filing Date
- 2024-08-13
- Publication Date
- 2025-07-11
AI Technical Summary
Technical Problem
In the prior art, the connection bandwidth between the host device and the memory device is low and the delay is long, resulting in memory sharing and consistency problems, increasing the amount of data movement and input/output traffic when learning the artificial intelligence model.
Method used
By adopting accelerator and nonvolatile memory in intelligent storage devices, and using cache consistency mechanisms, the repeated movement of data between the host device and the intelligent storage device is reduced, and the learning and training of the model is realized.
Benefits of technology
Improves the bandwidth and delay of data processing, reduces input/output traffic, and improves the learning efficiency of artificial intelligence models.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN120295954A_ABST
Abstract
A smart storage device, a computing system, and a method of operation of the computing system are provided. The smart storage device includes an accelerator connected to the host device through a smart interface and including a first model distributed from the host device, and a non-volatile memory including learning data for learning the first model, the accelerator learns a first model by using learning data based on a first weight and a first offset set in advance, provides a first output value output by inputting the learning data into the first model to the host device, and outputs a second output value to the host device based on a second weight and a second offset calculated by referencing the first output value from the host device. The first model is learned by using the learning data.
Need to check novelty before this filing date? Find Prior Art
Citation Information
Patent Citations
Compositions based on polyamide and glass fibers and their use in the medical field
KR1020240004622A