Systolic Array Cross-Channel Operations in Parallel Computing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current parallel graphics processing systems face inefficiencies in performing cross channel operations due to limitations in processing parallel data streams effectively, particularly in systolic arrays used in graphics processing units (GPUs).

Innovation Solution

The implementation of systolic arrays within parallel computing machines to efficiently compute cross channel operations by optimizing data processing through specialized architectures and instruction formats, enabling simultaneous processing of multiple data points and threads.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional fixed function computational units are used in graphics processors, then the system maintains simplicity and reliability, but the ability to perform cross channel operations efficiently is limited

Engineering Contradiction:
Improvecross channel operation efficiencyVSAvoidprocessor architecture complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The processor is divided into multiple processing channels (first processing channel and second processing channel), each capable of independent operation. This segmentation allows parallel processing of different data streams while maintaining modular simplicity in each channel's design.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The processing elements in each channel are designed to perform multiple operations including cross channel operations, intra channel operations, and various computational tasks. This multi-functionality enables the same hardware structure to handle diverse workloads without requiring specialized fixed function units for each operation type.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Productivity

If parallel processing is increased using SIMD or SIMT architectures, then throughput is improved, but the complexity of coordinating multiple threads and data streams increases

Engineering Contradiction:
Improveparallel processing throughputVSAvoidthread coordination complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The parallel processing system is segmented into distinct processing channels with dedicated processing elements. Each channel can execute threads independently, reducing the coordination overhead compared to fully integrated SIMT architectures while maintaining high parallel throughput through channel-level parallelism.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Cross channel operation units act as intermediary components that facilitate data exchange and coordination between processing channels. These intermediaries simplify the coordination complexity by providing standardized interfaces and mechanisms for inter-channel communication, reducing the burden on individual processing elements.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Adaptability or versatility

If more processing elements are added to increase parallelism, then cross channel operation capability is improved, but the device complexity and resource requirements increase

Engineering Contradiction:
Improvecross channel operation capabilityVSAvoidprocessing element quantity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

Each processing element is designed with universal functionality to perform both intra channel operations and cross channel operations. This multi-functionality allows the system to achieve high cross channel operation capability without requiring separate dedicated hardware for each operation type, thereby controlling device complexity.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent merges the functionality of multiple processing elements into organized channels, where processing elements within a channel work together on related tasks. This merging approach achieves versatility in cross channel operations while reducing overall device complexity through functional integration and shared resources.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentUS20210365402A1Computing efficient cross channel operations in parallel computing machines using systolic arrays
Publication Date: 2021.11.25 INTEL CORP
  • US20210365402A1 patent drawing
  • US20210365402A1 patent drawing
  • US20210365402A1 patent drawing

AI summary

An apparatus to facilitate computing efficient cross channel operations in parallel computing machines using systolic arrays is disclosed. The apparatus includes a plurality of registers and one or more processing elements communicably coupled to the plurality of registers. The one or more processing elements include a systolic array circuit to perform cross-channel operations on source data received from a single source register of the plurality of registers, the systolic array circuit modified to receive inputs from the single source register and route elements of the single source register to multiple channels in the systolic array circuit.