Heterogeneous Neural Network Frequency Scaling for Targeted Execution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural network computation systems face challenges in reducing energy consumption while maintaining target execution times due to the use of heterogeneous computing devices, as existing frequency control methods often fail to consider the specific power consumption and execution time requirements of these devices.
Innovation Solution
A neural network computing system that dynamically adjusts the operating frequencies of heterogeneous computing devices based on feedback control mechanisms, using a frequency level combination table to minimize energy consumption while ensuring target performance levels are met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If heterogeneous computing devices are used to perform neural network computation, then computation capability is improved, but power consumption increases
Solution Approach 1:
The patent implements dynamic frequency adjustment for heterogeneous computing devices based on real-time workload analysis. The system continuously monitors computation requirements and dynamically scales the operating frequency of each device type (CPU, GPU, NPU) to match the actual computational demand, avoiding static high-frequency operation that wastes energy during low-utilization periods
Solution Approach 2:
The system changes operational parameters by adjusting the frequency levels of heterogeneous computing devices according to the specific characteristics of neural network workloads. Different device types are assigned optimal frequency ranges based on their architectural characteristics and the computational patterns required, transforming fixed-frequency operation into adaptive parameter control
2Use of energy by moving object
If frequency levels of heterogeneous computing devices are adjusted to reduce power consumption, then energy efficiency is improved, but execution time may exceed target
Solution Approach 1:
The patent implements a feedback control mechanism that monitors actual execution time and power consumption, then adjusts frequency levels in subsequent operations. The system compares measured performance against target values and uses this feedback to optimize frequency selection, ensuring that energy efficiency improvements do not compromise execution time requirements
Solution Approach 2:
The system performs preliminary analysis of the neural network workload characteristics before execution begins. By pre-calculating the computational requirements and estimating optimal frequency levels for each device type, the system can proactively configure frequency settings that balance energy efficiency with execution time constraints, avoiding reactive adjustments that may miss performance targets
3Device complexity
If existing frequency control methods are used, then system simplicity is maintained, but they fail to consider specific power consumption and execution time requirements of heterogeneous devices
Solution Approach 1:
The patent segments the frequency control mechanism into device-specific control units that independently manage CPU, GPU, and NPU frequency levels. Each segment handles the specific power consumption and performance characteristics of its assigned device type, allowing tailored optimization without requiring complete system redesign or excessive complexity
Data Source
AI summary
A neural network computing system includes a processor including heterogeneous computing devices, a memory, a memory controller that controls the memory, and a system bus that communicates between the processor and the memory controller. The processor generates a frequency level combination table representing frequency level combinations of hardware devices that include the heterogeneous computing devices, the memory controller, and the system bus, each of the frequency level combinations maximizing a decreasing amount of execution time relative to an increasing amount of energy consumption when a performance level of the neural network model is increased by one level, determines, a target performance level, selects a frequency level combination among the frequency level combinations according to the target performance level, and controls, based on a selected frequency level combination, the hardware devices while the neural network model is executed.