GPU Non-Volatile Storage for Asset Loading
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Game engines face significant overhead due to disk IO limitations when transferring and loading game assets to a GPU, leading to slow load times and inefficient memory operations.
Innovation Solution
A system with a dedicated non-volatile memory for the GPU allows for direct storage and retrieval of game assets, bypassing traditional loading mechanisms by using a computer-readable storage medium to detect requests and allocate buffer resources, and transferring content via DMA when optimization criteria are met, thereby reducing reliance on host system disk IO.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If content is transferred from host system disk to GPU using traditional memory operations, then the content can be loaded into GPU memory, but the process is bound by disk IO capability and imposes considerable overhead
Solution Approach 1:
The patent extracts the non-volatile storage function from the host system and integrates it directly into the GPU device. This allows the GPU to access content directly from its own non-volatile storage without going through the host system's disk IO path, thereby eliminating the disk IO bottleneck and reducing load times.
Solution Approach 2:
The patent introduces a buffer resource as an intermediary between the non-volatile storage and the GPU processing units. This buffer serves as a staging area that allows efficient data transfer and caching, enabling the GPU to access frequently used content quickly while maintaining the ability to load new content from non-volatile storage.
2Productivity
If content is transferred through volatile memory dedicated to GPU, then the content can be accessed by GPU, but considerable overhead is imposed during memory operations
Solution Approach 1:
The patent implements preliminary action by pre-loading and caching content in the buffer resource before it is actually needed for rendering or processing. This allows the GPU to quickly access already-prepared content without performing time-consuming memory operations at the moment of need, thereby reducing overhead and improving productivity.
Solution Approach 2:
The patent changes the storage parameter from volatile memory to non-volatile storage integrated with the GPU. This fundamental parameter change allows content to be persisted without requiring continuous power and enables faster access patterns by eliminating the need to constantly reload content from the host system's volatile memory.
3Loss of time
If traditional loading mechanisms are used, then content can be loaded to GPU, but redundant content generation occurs and load times are slow
Solution Approach 1:
The patent implements discarding and recovering by allowing the buffer resource to retain cached content for repeated use while discarding the need to reload identical content from non-volatile storage. When content is accessed again, it is recovered from the buffer rather than being regenerated or reloaded, eliminating redundant operations and reducing load times.
Data Source
AI summary
Methods and systems may provide for an apparatus having a graphics processing unit (GPU) and a non-volatile memory dedicated to the GPU. If a request for content is detected, a determination may be made as to whether the non-volatile memory contains the content.