GPU Virtualization via Hardware Display Planes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current solutions for GPU virtualization, particularly in data centers and cloud environments, do not optimally support alternative models of GPU virtualization, limiting their effectiveness in processing graphics and machine-learning operations.
Innovation Solution
A computing system architecture that integrates a graphics processing unit (GPU) with host/processor cores, enabling efficient processing of graphics and machine-learning operations through a high-speed interconnect, and allowing for parallel processing techniques such as SIMT architectures to maximize parallelism in the graphics pipeline.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If GPU virtualization is implemented using existing data center solutions, then remote rack server operation is supported, but alternative virtualization models are not optimally supported
Solution Approach 1:
The GPU system is designed to support multiple virtualization models (data center remote rack, local display, and hybrid configurations) through a universal architecture that can adapt to different operational modes. The system provides multi-functionality by enabling the same GPU hardware to serve various virtualization scenarios without requiring dedicated hardware for each model.
2Ease of operation
If GPU output is collected and sent over network connection, then remote rack server operation is enabled, but local display processing efficiency is reduced
Solution Approach 1:
The system dynamically adapts its output path based on operational requirements. The GPU can switch between sending output over network connections for remote operations and direct local display connections for performance-critical applications. This dynamic configuration allows the system to optimize for either ease of remote operation or graphics processing efficiency depending on the current workload and user needs.
Data Source
AI summary
Described herein is a partitional graphics processor including a display controller including hardware display virtualization. One embodiment provides a graphics processor comprising a system interface including a first virtual interface and a second virtual interface, a render engine to perform graphics rendering operations, and a display engine including hardware display virtualization. The render engine is configured to perform a first rendering operation in response to a command received via the first virtual interface and a second rendering operation in response to a command received via the second virtual interface. The display engine configured to present output of the first rendering operation via a first physical display plane that is associated with the first virtual interface and present output of the second rendering operation via a second physical display plane that is associated with the second virtual interface.