Configurable GPU Pipeline for Systolic and Normal Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current graphics processing units (GPUs) are inadequate for processing large amounts of data required for machine learning and related technologies, limiting their compute capabilities for certain applications.
Innovation Solution
A graphics processing unit (GPU) is communicatively coupled to host processor cores to accelerate graphics and machine-learning operations, featuring a configurable pipeline for systolic output, Min/Max operations, and a pseudo-random number generator, enhancing calculation capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional GPU architecture is used, then graphics processing capability is maintained, but compute capability for machine learning and large data processing is inadequate
Solution Approach 1:
The patent implements a unified processing architecture where the same computational units handle both traditional graphics processing tasks and machine learning computations. The processing clusters can dynamically switch between rendering operations and data processing operations, allowing a single GPU to serve multiple purposes without requiring separate specialized hardware for each function type.
Solution Approach 2:
The processor employs dynamic configuration of processing pipelines, where the same hardware resources can be reconfigured on-the-fly to handle different types of computations. The architecture allows runtime adjustment of processing modes, enabling seamless transition between graphics rendering and machine learning workloads based on the current task requirements.
2Productivity
If standard processing functions are implemented, then device complexity is reduced, but processing efficiency for specific calculations is limited
Solution Approach 1:
The processor is divided into multiple independent processing clusters, each capable of handling specific computational tasks. Within each cluster, functional units are segmented to handle different operation types (arithmetic, logic, data movement). This segmentation allows parallel execution of multiple operations simultaneously, improving overall processing efficiency while keeping each individual unit relatively simple.
Solution Approach 2:
The patent combines multiple specialized functions into integrated processing units. For example, arithmetic logic units are merged with data movement capabilities and control logic into unified processing elements. This merging reduces the number of separate components needed while maintaining high processing efficiency through coordinated operation of integrated functions.
Data Source
AI summary
Enhanced processor functions for calculation are described. An example of an apparatus includes one or more processors including one or more processing resources and a memory to store data, the data including data for compute operations. A processing resource of the one or more processing resources includes a configurable pipeline for calculation operations, and wherein the configurable pipeline may be utilized to perform both a normal instruction for a calculation in a certain precision and a systolic instruction for a calculation in a certain precision.


