Unified Virtual Memory for Heterogeneous System Architecture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current GPU technology configures host and GPU memories as separate entities, leading to decreased processing speed and complex configuration due to the need for data transfer between them.
Innovation Solution
Implementing a heterogeneous system architecture with unified virtual memory that allows the host and HSA device to share memory, enabling parallel processing across multiple processor cores and offloading host operations onto the device.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If separate GPU memory (GDDR) is used instead of sharing main memory, then device memory capacity is improved, but data transfer complexity and time increase
Solution Approach 1:
The patent merges the host memory and device memory into a unified memory space, allowing both CPU and GPU to access the same physical memory without data transfer. This eliminates the need for separate GDDR memory while removing the associated data transfer overhead between host and device.
Solution Approach 2:
The unified memory architecture introduces a memory management unit that acts as an intermediary, handling address translation and memory allocation for both host and device simultaneously. This mediator enables direct memory sharing without requiring traditional data transfer protocols.
2Adaptability or versatility
If separate GPU memory is used, then device independence is maintained, but system configuration complexity increases
Solution Approach 1:
The unified memory architecture creates a universal memory interface that serves both host and device functions simultaneously. This single memory system performs multiple roles - serving as both host memory and device memory - thereby eliminating the need for separate interface configurations while maintaining device independence through virtual memory addressing.
3Stability of the object's composition
If data transfer between separate memories is required, then memory separation is maintained, but processing speed decreases
Solution Approach 1:
The patent combines separate memory spaces into a unified memory architecture where host and device share the same physical memory resources. This merging eliminates data transfer operations entirely, as both CPU and GPU can directly access the same memory locations, thereby maintaining memory stability while dramatically improving processing speed.
Data Source
AI summary
Disclosed herein is a heterogeneous system based on unified virtual memory. The heterogeneous system based on unified virtual memory may include a host for compiling a kernel program, which is source code of a user application, in a binary form and delivering the compiled kernel program to a heterogenous system architecture device, the heterogenous system architecture device for processing operation of the kernel program delivered from the host in parallel using two or more different types of processing elements, and unified virtual memory shared between the host and the heterogenous system architecture device.