GPU Memory Control via Fixed Variable Area Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In graphic processing units, the limited image memory leads to overhead during task switching, causing degradation in image quality such as video discontinuity or dropped frames, especially in real-time applications like games, due to the need to transfer and rewrite image data.
Innovation Solution
The memory is divided into fixed and variable areas, allowing efficient sharing by allocating space and time-divided memory spaces to tasks, where data for real-time tasks is stored in a fixed area to avoid re-reading and data for non-real-time tasks is stored in a variable area, reducing switching costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If image memory is shared by multiple tasks, then memory utilization is improved, but task switching overhead increases causing video discontinuity and dropped frames
Solution Approach 1:
The image memory is divided into multiple independent task-specific memory areas, each allocated to a specific task. This segmentation allows simultaneous access by multiple tasks without requiring data transfer during task switching, thereby reducing task switching overhead while maintaining memory sharing capability.
Solution Approach 2:
Memory areas are pre-allocated to tasks before task switching occurs. This preliminary allocation ensures that when task switching happens, data is already in the appropriate memory area, eliminating the need for time-consuming data transfer operations and preventing video discontinuity.
2Productivity
If image data is rewritten in limited image memory, then new task data can be stored, but image quality degrades due to rewriting time
Solution Approach 1:
By segmenting image memory into task-specific areas, each task has dedicated space that does not require rewriting when tasks switch. This eliminates image quality degradation caused by rewriting while allowing new task data to be stored in previously unused task-specific areas.
Solution Approach 2:
The patent transitions from temporal memory allocation (overwriting data over time) to spatial memory allocation (dedicated areas for each task). This dimensional change allows multiple tasks to coexist in the limited image memory without rewriting, maintaining image quality while enabling task data storage.
3Adaptability or versatility
If data transfer is performed before task switching, then memory can be reused, but system overhead increases
Solution Approach 1:
The memory is segmented into fixed task-specific areas that are allocated once and reused throughout task execution. This eliminates the need for complex data transfer operations before task switching, reducing system overhead while maintaining memory reuse capability through persistent allocation.
Solution Approach 2:
Each task-specific memory area serves multiple purposes: it stores data for its designated task, can be quickly switched to another task, and maintains data persistence across task switches. This multi-functionality achieves memory reuse without requiring complex data transfer protocols.
Data Source
AI summary
A graphic memory is space-divided into a first area and a second area. In the first area, a task corresponding to a predetermined application is executed regardless of which task is processed by a main processor. A switchable area is divided in time so that data related to tasks corresponding to a plurality of applications is sequentially stored in the switchable area in accordance with task switching in the main processor.