Scalable Compute Fabric Dynamic Pipeline Configuration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computing systems are inflexible and power-intensive, as they consist of fixed CPU and GPU cores that remain active even when not in use, unable to dynamically reconfigure for varying workloads or power levels.
Innovation Solution
A scalable compute fabric that dynamically configures compute elements into multiple pipelines, allowing for dynamic power management and reconfiguration of CPU and GPU resources to match workload demands, enabling efficient use of resources and reducing power consumption by powering off inactive components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If fixed CPU and GPU cores are kept active, then processing capability is maintained, but power consumption increases
Solution Approach 1:
The patent implements dynamic core activation where CPU and GPU cores are activated only when needed for specific workflows. The system transitions from a static state where all cores remain active to a dynamic state where core activation is controlled based on workload requirements, resolving the contradiction between maintaining processing capability and reducing power consumption.
Solution Approach 2:
The compute fabric creates a universal processing platform where CPU and GPU cores can be dynamically allocated to different workflows based on requirements. This multi-functional approach allows the same hardware resources to serve multiple purposes, maintaining processing capability while enabling power savings when certain functions are not currently needed.
2Adaptability or versatility
If compute elements are dynamically reconfigured, then adaptability to varying workloads is improved, but system complexity increases
Solution Approach 1:
The patent segments the computing system into distinct workflows that can be independently configured and executed. By dividing the compute fabric into separable workflow units with defined inputs and outputs, the system achieves adaptability to varying workloads while managing complexity through modular organization rather than monolithic reconfiguration.
Solution Approach 2:
The compute fabric acts as an intermediary layer between the hardware resources (CPU/GPU cores) and the workflows. This intermediary structure simplifies the interface for workload adaptation by providing standardized connection points and routing mechanisms, reducing the complexity burden that would otherwise exist in directly managing core reconfiguration.
3Speed
If dedicated circuitry is fixed for specific operations, then processing speed is improved, but flexibility to handle different workloads deteriorates
Solution Approach 1:
The patent implements dynamic routing within the compute fabric that can adapt the path of data flow based on workflow requirements. This dynamic approach maintains the speed benefits of dedicated circuitry by keeping physical connections intact while providing flexibility through configurable routing paths that can be adjusted for different workload types without requiring physical reconfiguration.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method and apparatus for providing a scalable compute fabric are provided herein. The method includes determining a workflow for processing by the scalable compute fabric, wherein the workflow is based on an instruction set. A pipeline in configured dynamically for processing the workflow, and the workflow is executed using the pipeline.