Fixed-Point ML Computing Architecture for Faster Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
General purpose processors are inefficient in processing large loads, leading to delays in machine learning tasks such as training artificial intelligence models, due to their limited operating speed.
Innovation Solution
A computation device and method that includes a controller unit, operation unit, and conversion unit, utilizing a primary processing circuit and secondary processing circuits to perform machine learning computations by converting input data into fixed-point data and executing operations efficiently, with a DMA unit and storage unit for data management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If general purpose processors are used to process information, then the processing can be performed with flexible software programs, but the operating speed is limited and efficiency is low when handling large loads
Solution Approach 1:
The processor is divided into multiple processing units (e.g., vector processing units, matrix processing units) that can operate in parallel. Each unit handles specific types of computations, allowing the system to process large loads efficiently while maintaining software flexibility through configurable unit activation.
Solution Approach 2:
The architecture transitions from scalar processing to vector and matrix processing by adding dimensional capabilities. This allows simultaneous processing of multiple data elements across different dimensions (time, space, data types), dramatically increasing throughput while preserving programming flexibility through high-level language support.
2Productivity
If general purpose processors operate at high speed, then processing efficiency improves, but power consumption increases
Solution Approach 1:
Different processing units are optimized for specific task types (vector, matrix, scalar operations). The system activates only the necessary units for each computation, concentrating processing power where needed rather than uniformly powering all units, thus achieving high speed for specific tasks with reduced overall power consumption.
Solution Approach 2:
The processor dynamically adjusts operating parameters such as clock frequency and voltage based on computational requirements. For intensive computations, higher frequencies are applied to active units; for lighter loads or idle units, frequencies are reduced, optimizing the balance between processing speed and power consumption.
Data Source
AI summary
The present disclosure provides a computation device. The computation device is configured to perform a machine learning computation, and includes an operation unit, a controller unit, and a conversion unit. The storage unit is configured to obtain input data and a computation instruction. The controller unit is configured to extract and parse the computation instruction from the storage unit to obtain one or more operation instructions, and to send the one or more operation instructions and the input data to the operation unit. The operation unit is configured to perform operations on the input data according to one or more operation instructions to obtain a computation result of the computation instruction. In the examples of the present disclosure, the input data involved in machine learning computations is represented by fixed-point data, thereby improving the processing speed and efficiency of training operations.