Flexible GPU Partitioning for Higher Functional Unit Utilization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing data center graphics processors suffer from inefficiencies due to underutilization of functional units in partitioned partitions, leading to suboptimal resource allocation and performance.

Innovation Solution

Implement flexible partitioning of GPU resources by dynamically allocating workloads across processing clusters, allowing for efficient utilization of all functional units and optimizing performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If GPU resources are partitioned into isolated portions for multiple clients, then resource isolation and multi-tenancy are improved, but functional unit utilization deteriorates

Engineering Contradiction:

Improveresource isolationVSAvoidfunctional unit utilization

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The GPU is divided into multiple compute partitions, each with isolated resources (registers, cache, execution units), allowing multiple tenants to operate independently while maintaining resource isolation guarantees

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Compute partitions can be dynamically allocated to different tenants and workloads, allowing the same physical GPU resources to serve multiple purposes and clients through flexible partitioning and assignment mechanisms

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Ease of operation

If fixed partitioning is used for GPU resources, then resource allocation simplicity is improved, but resource utilization efficiency deteriorates

Engineering Contradiction:

Improveresource allocation simplicityVSAvoidresource utilization efficiency

Core Design Contradiction:

Ease of operationVSProductivity

Solution Approach 1:

The partitioning configuration is dynamically adjustable through runtime API calls, allowing the system to adapt partition sizes and assignments based on workload requirements while maintaining operational simplicity through standardized interfaces

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS12572392B2Flexible partitioning of GPU resources

Publication Date: 2026.03.10 INTEL CORP

AI summary

Described herein is a partitionable graphics processor having a plurality of flexibly partitioned processing resources. One embodiment provides a graphics processor comprising a plurality of processing resources configurable to be flexibly partitioned into a plurality of resource partitions and circuitry to compose multiple graphics processor device partitions from the plurality of resource partitions. The multiple graphics processor device partitions are configurable to be asymmetrically composed of different types of functional units.