Data Stream Accelerator Dynamic Parallelism to Reduce Zero Filling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data stream architecture-based accelerators face issues with fixed parallelism that lead to increased data transmission bandwidth, storage requirements, and power consumption due to zero filling, which reduces computing resource utilization.
Innovation Solution
A data stream architecture-based accelerator with a storage unit, read-write address generation unit, and computing unit that allows for configurable read-write and computing parallelism, optimizing data access by dynamically determining optimal parallelism based on data size and reducing unnecessary data transmission and storage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If fixed parallelism is used in conventional data stream architecture-based accelerator, then computing resource utilization is improved, but data transmission bandwidth increases due to zero filling
Solution Approach 1:
The patent implements dynamic parallelism configuration where the read-write parallelism and computing parallelism can be adjusted based on the actual data size and processing requirements. The controller determines optimal parallelism values dynamically rather than using fixed parallelism, allowing the system to adapt to different workloads and avoid zero-filling operations that waste bandwidth.
Solution Approach 2:
The system changes the parallelism parameter (read-write parallelism and computing parallelism) based on the characteristics of the data being processed. By adjusting these parameters dynamically, the system optimizes the balance between computing resource utilization and data transmission bandwidth requirements, eliminating the need for zero-filling when data size does not align with fixed parallelism values.
2Productivity
If fixed parallelism is used in conventional data stream architecture-based accelerator, then computing resource utilization is improved, but storage requirements increase due to zero filling
Solution Approach 1:
The patent implements dynamic parallelism configuration where the read-write parallelism and computing parallelism can be adjusted based on the actual data size and processing requirements. The controller determines optimal parallelism values dynamically rather than using fixed parallelism, allowing the system to adapt to different workloads and avoid zero-filling operations that waste storage space.
Solution Approach 2:
The system changes the parallelism parameter (read-write parallelism and computing parallelism) based on the characteristics of the data being processed. By adjusting these parameters dynamically, the system optimizes the balance between computing resource utilization and storage requirements, eliminating the need to store filled zero values.
3Productivity
If fixed parallelism is used in conventional data stream architecture-based accelerator, then computing resource utilization is improved, but processing time increases due to zero filling operations
Solution Approach 1:
The patent implements dynamic parallelism configuration where the read-write parallelism and computing parallelism can be adjusted based on the actual data size and processing requirements. The controller determines optimal parallelism values dynamically rather than using fixed parallelism, allowing the system to adapt to different workloads and avoid zero-filling operations that extend processing time.
Solution Approach 2:
The system changes the parallelism parameter (read-write parallelism and computing parallelism) based on the characteristics of the data being processed. By adjusting these parameters dynamically, the system optimizes the balance between computing resource utilization and processing time, eliminating unnecessary zero-filling operations that prolong execution.
4Productivity
If fixed parallelism is used in conventional data stream architecture-based accelerator, then computing resource utilization is improved, but power consumption increases due to storage and computation for filled zero
Solution Approach 1:
The patent implements dynamic parallelism configuration where the read-write parallelism and computing parallelism can be adjusted based on the actual data size and processing requirements. The controller determines optimal parallelism values dynamically rather than using fixed parallelism, allowing the system to adapt to different workloads and avoid zero-filling operations that consume unnecessary power for storage and computation.
Solution Approach 2:
The system changes the parallelism parameter (read-write parallelism and computing parallelism) based on the characteristics of the data being processed. By adjusting these parameters dynamically, the system optimizes the balance between computing resource utilization and power consumption, eliminating unnecessary computations and storage operations on filled zero values that waste energy.
Data Source
AI summary
A data stream architecture-based accelerator includes a storage unit, a read-write address generation unit and a computing unit. The storage unit includes a plurality of banks. The read-write address generation unit is used for generating storage unit read-write addresses according to a preset read-write parallelism, determining target banks in the storage unit according to the storage unit read-write addresses and reading to-be-processed data from the target banks for operations in the computing unit. The computing unit includes a plurality of data paths and is configured to determine target data paths according to a preset computing parallelism so that the target data paths can perform operations on the to-be-processed data to obtain processed data, and then store the processed data into the target banks according to the storage unit read-write addresses.