Integrated GPU Direct System Memory Access Eliminating Data Copy

Solution Overview

Problem

In systems with integrated graphics processors (IGPs), existing methods require data copying from system memory to GPU memory for processing, which is inefficient due to separate memory access and external bus limitations.

Innovation Solution

A method where the GPU directly accesses system memory without copying data when it is an IGP, utilizing shared memory spaces and APIs to determine and utilize the integrated GPU capabilities for processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If data is copied from system memory to GPU memory for processing, then compatibility with discrete GPUs is maintained, but processing efficiency deteriorates due to additional copy operations

Engineering Contradiction:

Improveprocessing efficiencyVSAvoiddata copy time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent merges the system memory and GPU memory into a unified memory space when an integrated graphics processor is detected. This eliminates the need for data copying between separate memory spaces, as both the CPU and IGP can access the same memory region directly. The memory manager detects the IGP configuration and configures the memory architecture to allow direct access, thereby resolving the technical contradiction by eliminating the harmful data copy operation entirely.

Inventive Principle:

Principle #5Merging (Combining)

2Ease of operation

If data copying is performed for GPU processing, then memory access compatibility is maintained, but device complexity increases due to memory management overhead

Engineering Contradiction:

Improvememory access simplicityVSAvoidmemory management complexity

Core Design Contradiction:

Ease of operationVSDevice complexity

Solution Approach 1:

The patent implements a dynamic memory management approach where the system automatically detects whether an integrated graphics processor is present and adjusts the memory access configuration accordingly. When an IGP is detected, the system switches to a unified memory mode with direct access; when a discrete GPU is detected, it switches to the traditional copy mode. This dynamic adaptation simplifies memory access for IGP systems while maintaining compatibility with discrete GPUs, resolving the contradiction between ease of operation and device complexity.

Inventive Principle:

Principle #15Dynamics

3Speed

If separate memory spaces are used for system memory and GPU memory, then discrete GPU compatibility is maintained, but processing speed deteriorates due to external bus access

Engineering Contradiction:

Improveprocessing speedVSAvoidmemory architecture

Core Design Contradiction:

SpeedVSDevice complexity

Solution Approach 1:

The patent merges the system memory and GPU memory into a unified memory space when an integrated graphics processor is detected. This eliminates the need for data copying between separate memory spaces, as both the CPU and IGP can access the same memory region directly. The memory manager detects the IGP configuration and configures the memory architecture to allow direct access, thereby resolving the technical contradiction by eliminating the harmful data copy operation entirely.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS8760452B2Integrated graphics processor data copy elimination method and apparatus when using system memory

Publication Date: 2014.06.24 ADVANCED MICRO DEVICES INC

AI summary

A method and apparatus for processing data in a system comprising a central processing unit (CPU), a system memory, and a graphics processing unit (GPU) includes determining whether the GPU is an integrated graphics processor (IGP). Based upon a determination that the GPU is an IGP, data stored in the system memory is accessed by the GPU without copying the data to a memory on the GPU. Processing is performed on the data in the GPU.