A method and system for aging test of a microcomputer host

Abstract

The present invention discloses a method and system for aging test of a microcomputer host, which relates to the technical field of computer performance testing. It includes connecting the device to be tested to the test system, performing a system standby test, and collecting initial reference data; constructing an aging prediction model and optimizing the model training strategy by combining curriculum learning and transfer learning; based on the prediction results, using AIDA64 to perform a host preheating test and a high-load test, and synchronously running the Heaven Benchmark tool to test the graphics processing ability of the computer. The present invention combines a radial basis function neural network aging prediction model, uses curriculum learning and transfer learning for model training, intelligently analyzes the degradation trends of the CPU / GPU, and combines a quantum random number generator to generate a random load pattern to simulate the load fluctuations in the actual application scenario, thereby improving the authenticity of the test. In addition, the present invention provides a BIOS optimization strategy to achieve intelligent aging prediction and optimization control, so as to improve the long-term stability and service life of the microcomputer host.

Description

Technical Field

[0001] The present invention relates to the technical field of computer performance testing, and particularly to an aging test method and system for a microcomputer host. Background Art

[0002] With the rapid development of computer hardware, microcomputer hosts have been widely used in fields such as industrial control, edge computing, smart home, and embedded systems due to their advantages of small size, high energy efficiency, and strong computing power. However, during long-term operation of computer hardware devices, due to factors such as heat accumulation, material aging, circuit loss, and storage medium fatigue, performance degradation gradually occurs, and even failures occur, affecting the stability and service life of the system. Therefore, how to effectively evaluate the aging state of a microcomputer host and optimize its operation strategy to extend its service life has become an important research direction in the field of current computer hardware engineering.

[0003] Currently, the evaluation methods for computer hardware aging in the industry mainly include accelerated aging tests, temperature stress tests, power consumption monitoring, storage life evaluation, etc. These methods generally use long-term high-load operation, temperature and humidity extreme environment tests, etc. to observe the stability and performance degradation trends of devices under harsh conditions. For example, traditional computer aging tests usually use AIDA64 FPU to test CPU performance, Heaven Benchmark to evaluate GPU rendering capabilities, Rebooter for continuous restart tests, and S3 / S4 sleep recovery tests to verify the stability of low-power modes. These test methods can simulate different working scenarios, but there are problems such as long test cycles, insufficient data analysis, and limited test coverage, making it difficult to comprehensively reflect the aging trends of microcomputer hosts in different application scenarios. In addition, most existing test methods rely on fixed test modes and cannot adapt to the aging characteristics of different devices, resulting in poor generalization ability of test results. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an aging test method and system for a microcomputer host, which solves the problem that most existing test methods rely on fixed test modes and cannot adapt to the aging characteristics of different devices, resulting in poor generalization ability of test results.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an aging test method for a microcomputer host, which includes

[0008] Connect the device under test to the test system, perform a system standby test, and collect initial reference data;

[0009] Build an aging prediction model and optimize the model training strategy by combining curriculum learning and transfer learning;

[0010] Based on the prediction results, use AIDA64 to conduct host warm-up tests and high-load tests, and run the Heaven Benchmark tool synchronously to test the graphics processing ability of the computer;

[0011] Use a quantum random number generator to generate random load patterns, alternate high and low loads of the CPU / GPU, simulate real user scenarios, conduct Rebooter and S3 / S4 sleep recovery tests and trigger anomalies;

[0012] Comprehensively analyze all test data and provide optimization suggestions.

[0013] As a preferred solution of the aging test method for the microcomputer host described in the present invention, wherein: the building of the aging prediction model and the optimization of the model training strategy by combining curriculum learning and transfer learning include the following steps:

[0014] Use a radial basis function neural network model to build an aging prediction model, define a curriculum learning strategy, and divide the curriculum learning into three stages: a basic stage, an adaptation stage, and a generalization stage for model training;

[0015] Take the radial basis function neural network model trained by curriculum learning as the starting point of transfer learning, freeze the first N layers of the radial basis function neural network model and keep them unchanged, only optimize the center points of the radial basis function neural network model in the last M layers, use a correlation alignment loss function for data feature alignment, use the gradient descent method to optimize the radial basis function neural network model, update the parameters of the radial basis function neural network model after training is completed, save the radial basis function neural network model trained by transfer learning as the final model, normalize the data in the benchmark dataset, and input the data into the radial basis function neural network model to obtain the aging trend of the computer host.

[0016] As a preferred solution of the aging test method for the microcomputer host described in the present invention, wherein: the use of AIDA64 to conduct host warm-up tests and high-load tests based on the prediction results, and synchronously run the Heaven Benchmark tool to test the graphics processing ability of the computer includes the following steps:

[0017] Calculate the benchmark performance offset ratio based on the prediction results , set a threshold A, if the offset ratio is greater than the threshold A, then pre-lower the CPU core voltage, run the AIDA64 FPU test, record the CPU temperature, power consumption, and fan speed, and calculate the temperature change rate of the CPU frequency through multi-stage calculation and the actual frequency reduction ratio ;

[0018] If the actual frequency reduction ratio is greater than the reference performance offset ratio, it indicates a large prediction error, and the BIOS adjustment mode is entered, including further reducing the CPU core voltage and increasing the fan speed;

[0019] Continue to run the AIDA64 FPU test, record the CPU frequency and calculate the CPU frequency reduction rate. Set a threshold D. If the CPU frequency reduction rate is less than the threshold D, enter the BIOS power reduction strategy, optimize voltage and power management, and synchronously run HeavenBenchmark for GPU high-load testing. Record the GPU frequency and calculate the GPU frequency reduction trend. If the GPU frequency reduction trend is less than the threshold D, adjust the GPU frequency management strategy and optimize the power consumption mode.

[0020] As a preferred scheme of the aging test method for the microcomputer host of the present invention, wherein: the step of using a quantum random number generator to generate a random load pattern, alternating high and low loads of the CPU / GPU, simulating a real user scenario, and calculating the system stability parameters after high-load testing includes the following steps:

[0021] Extract the GPU frequency curve and the CPU temperature curve from the high-load test results;

[0022] Use a quantum random number generator to generate a non-linear CPU load pattern and a GPU load fluctuation range;

[0023] Use a quantum random number generator for non-linear mutation mode, calculate the CPU load change response time, set a threshold I. If the response time is greater than the threshold I, enter the BIOS power reduction mode;

[0024] Switch the operation of Heaven Benchmark among high load, power-saving mode, and sleep mode;

[0025] Calculate the GPU frequency volatility, set a threshold p. If the volatility is greater than the threshold p, enter the BIOS power reduction mode.

[0026] As a preferred scheme of the aging test method for the microcomputer host of the present invention, wherein: the step of performing the Rebooter and S3 / S4 sleep recovery tests and triggering exceptions includes the following steps:

[0027] Perform a standardized test environment to ensure stable external power supply, call the Rebooter automated test script, execute H consecutive restarts, and record the startup time each time. Calculate the startup time drift rate based on the maximum startup time and the minimum startup time. Set a threshold E. If the drift rate is greater than the threshold E, trigger exception detection and enter the low-power recovery test;

[0028] After the Rebooter performs x restarts, call the ACPI instruction to put the device into the S3 sleep mode, calculate the CPU response latency of the policy, set the threshold J. If the latency is greater than the threshold J, enter the BIOS adjustment. If the latency is less than or equal to the threshold J, continue to perform the S4 mode test;

[0029] Send the ACPI instruction to put the device into the S4 mode, calculate the time required to enter the sleep state, and trigger the S4 mode recovery, calculate the storage data consistency. If it is inconsistent, it means that the stored data is damaged and automatic repair is performed.

[0030] As a preferred solution of the mini - computer host aging test method of the present invention, wherein: the comprehensive analysis of all test data and providing optimization suggestions includes:

[0031] Calculate the average frequencies of the CPU and GPU at the initial time and the current time point of the test, collect the SSD read - write rate, calculate the device degradation rate based on the CPU / GPU frequencies obtained at the current time point and the running time of the device, set the threshold F. If the degradation rate exceeds the threshold F, it indicates that the device is severely aged and optimization measures need to be taken.

[0032] As a preferred solution of the mini - computer host aging test method of the present invention, wherein: the initial reference data refers to the fan speed, CPU frequency , temperature and power consumption as well as the GPU frequency , temperature and power consumption , and the randomness measure of the initial state of the device .

[0033] In a second aspect, the present invention provides a mini - computer host aging test system, including,

[0034] A data acquisition module, used to connect the device under test to the test system, perform system standby tests, and collect initial reference data;

[0035] An aging prediction module, used to construct an aging training model using a radial basis function neural network model, and improve the generalization ability of the model by combining curriculum learning and transfer learning;

[0036] A performance test module, used to perform high - load CPU / GPU tests through AIDA64 + Heaven Benchmark;

[0037] A mutation stress test module, used to generate a non - linear CPU / GPU load pattern, simulate the real - world usage scenario, calculate the load response time, and trigger the BIOS power - saving mode;

[0038] A test module for conducting continuous reboot tests on the Rebooter, calculating system stability, triggering anomaly detection, and optimizing the sleep recovery strategy;

[0039] An optimization suggestion module for comprehensively analyzing all test data and providing device optimization suggestions.

[0040] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the mini-computer host aging test method described in the first aspect of the present invention is implemented.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the mini-computer host aging test method described in the first aspect of the present invention is implemented.

[0042] The beneficial effects of the present invention are as follows: The present invention combines a radial basis function neural network aging prediction model, uses curriculum learning and transfer learning for model training, intelligently analyzes the degradation trends of the CPU / GPU, and combines a quantum random number generator to generate random load patterns to simulate load fluctuations in actual application scenarios, improving the authenticity of the test. In addition, the present invention provides BIOS optimization strategies, such as dynamically adjusting the CPU / GPU frequency, optimizing the fan strategy, and enabling the SSD TRIM operation, to achieve intelligent aging prediction and optimization control, thereby improving the long-term stability and service life of the mini-computer host. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 It is a flowchart of the mini-computer host aging test method in Embodiment 1.

[0045] Figure 2 It is a structural diagram of the mini-computer host aging test system in Embodiment 1.

[0046] Figure 3 It is a schematic flowchart of aging model training in Embodiment 1.

[0047] Figure 4 It is a schematic flowchart of load testing in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification.

[0049] In the following description, numerous specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0050] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other with other embodiments.

[0051] Example 1, referring to Figures 1-4 , which is the first embodiment of the present invention. This embodiment provides a method for aging testing a microcomputer host, including the following steps:

[0052] S1. Connect the device under test to the test system, perform a system standby test, and collect initial reference data.

[0053] Specifically, select a standard regulated power supply (DC 12V ± 0.5V) to ensure a stable input voltage. Connect a power sensor, measure and record the input voltage V, and calculate the standby power consumption:

[0054] ,

[0055] In the formula, is the standby power consumption, V is the input voltage, and I is the standby current;

[0056] Connect a current sensor to the host power input terminal, configure the sampling frequency (1 kHz), record the initial current , calculate its average value. If deviates from the normal value by more than 10%, check the power supply system of the device under test.

[0057] Read the SSD SMART data, record the health status, perform a 10-minute standby test, and record the fan speed, CPU frequency , temperature and power consumption as well as the GPU frequency , temperature and power consumption ;

[0058] The permutation entropy (PE) method is used to calculate the system state distribution of the device under test in the standby state. The permutation entropy (PE) is used as a measure of the randomness of the initial state, and the calculation formula is as follows:

[0059] ;

[0060] In the formula, is the reference permutation entropy, is the probability of the j-th system state occurring, and k is the number of possible states of the system;

[0061] The system state in this test environment refers to the performance of the CPU, GPU, and other key hardware components within a specific time window, including but not limited to the following indicators:

[0062] CPU state, GPU state, storage device state, cooling system state, and memory state;

[0063] Integrate the reference data set based on the collected data:

[0064] .

[0065] Select a standard regulated power supply (DC 12V±0.5V) to ensure stable input voltage, avoid affecting the test results due to power voltage fluctuations, ensure a stable test environment, maintain the stability of the internal power supply of the host, prevent mismeasurement caused by abnormal power supply, use it as a test reference to make the data comparability between different test devices higher, connect a power sensor, measure and record the input voltage, calculate the standby power consumption, accurately measure the standby power consumption of the host, evaluate the power consumption of the device in different operating modes, monitor the state of the power supply system. If the power consumption increases abnormally, it may indicate hardware abnormalities or the standby mode is not correctly activated. After calculating the power consumption, the BIOS adjustment strategy can be further combined to optimize the power management and improve the energy efficiency of the host. Connect a current sensor, configure the sampling frequency (1kHz), record the initial current, and high-frequency sampling (1kHz) improves the test accuracy to ensure that small current fluctuations are detected. Read the SSD SMART data, and recording the health status can detect the degree of SSD degradation in advance, such as key parameters such as write life, I / O rate, and temperature. Perform a 10-minute standby test, record the fan speed, CPU frequency, temperature, power consumption, and GPU frequency, which can identify whether the fan strategy needs to be adjusted, analyze the standby power consumption performance of the device, compare with the high-load test data, and quantify the aging trend. The permutation entropy (PE) method is used to calculate the randomness of the system state to provide a tool for quantitatively evaluating the stability of the device's standby state and helping to identify the degree of system fluctuations.

[0066] S2. Build an aging prediction model and optimize the model training strategy by combining curriculum learning and transfer learning;

[0067] Specifically, constructing an aging prediction model and optimizing the model training strategy by combining curriculum learning and transfer learning includes the following steps:

[0068] Use a radial basis function neural network model to construct an aging prediction model, including an input layer, a hidden layer, and an output layer. Define a curriculum learning strategy, and divide curriculum learning into three stages: a basic stage, an adaptation stage, and a generalization stage;

[0069] The basic stage refers to the first 10% of the test duration, which is used for initial training. Low-complexity data is used to reduce model error and ensure stability;

[0070] The adaptation stage refers to the 10%-50% of the test duration, which is used to increase data complexity and improve the model's adaptability to the aging trend;

[0071] The generalization stage refers to more than 50% of the test duration, which is used to optimize the model's generalization ability, make it adapt to different aging modes, and optimize the BIOS resource management strategy;

[0072] Collect historical device data and perform normalization processing. Use the aging trend curve of historical data as the initial approximate prediction value. Calculate the training error using the weighted mean square error in the basic stage to reduce the impact of outliers on the model during the early training process. Use 90% real data and 10% predicted data for training to prevent unstable initial training effects caused by over-reliance on predicted values, and set the learning rate to 0.01 to reduce gradient fluctuations and improve convergence stability. Train the RBFNN to initially learn the data distribution characteristics and provide a basis for subsequent training stages;

[0073] In the adaptation stage, collect more complex aging characteristics, including storage I / O rate and thread management exception data (thread starvation, thread deadlock, and thread leak). Use 80% real data and 20% predicted data for training to improve the model's adaptability to long-term operation. Use the Huber loss to handle outliers, which is used to reduce the impact of sudden abnormal data on model convergence during the training process. Decay the learning rate to 0.005 to prevent training oscillations of the model in the adaptation stage;

[0074] Thread starvation is measured by the thread waiting time. If the waiting time is too long, it means that the thread is in a long-term starvation state. Thread deadlock is measured by the thread mutex waiting time. If the waiting time increases continuously for a long time, there may be a deadlock risk. Thread leak is measured by the total number of active threads. If the total number continues to increase while the CPU load does not increase significantly, there may be a thread leak. The specific judgment criteria are determined according to the actual application scenario and the experience of professionals;

[0075] In the generalization stage, collect system call exception data (invalid parameters, unavailable resources, timeouts, and interruptions) and process management status data (new status, ready status, running status, blocked status, and terminated status), use 70% real data and 30% predicted data for training to improve the generalization ability of the model, use the L1+L2 combined loss function for regularization to prevent the model from overfitting, and reduce the learning rate to 0.001 to ensure the final convergence of the model;

[0076] Take the radial basis function neural network model trained through curriculum learning as the starting point of transfer learning. Freeze the first N layers of the radial basis function neural network model and keep them unchanged, only optimize the last M layers. Use KL divergence to evaluate the difference between the new device data distribution and the old device data. Set the judgment threshold P through the cross-validation method. If the difference is greater than the threshold P, it means that the new device data distribution is different from the old device data, and data normalization processing is required (through Z-score normalization and applying the batch normalization method). If the difference is less than or equal to the threshold P, directly enter the transfer learning training without additional data preprocessing;

[0077] During the transfer learning training process, only optimize the center points of the radial basis function neural network model, use the correlation alignment loss function for data feature alignment to reduce the difference in data distribution between different devices, use the gradient descent method to optimize the radial basis function neural network model, update the parameters of the radial basis function neural network model after training is completed, and save the radial basis function neural network model trained by transfer learning as the final model;

[0078] After normalizing the data in the benchmark dataset, input the data into the radial basis function neural network model to obtain the aging trend of the computer host.

[0079] The basic stage (the first 10% of the test duration) improves the robustness of the initial model to ensure that subsequent training will not be interfered by early prediction errors. The adaptation stage (10%-50% of the test duration) enhances the model's adaptability to complex aging patterns and improves the prediction accuracy for long-running devices. The generalization stage (more than 50% of the test duration) enhances the model's adaptability on different devices and improves the generalization effect of aging prediction. Transfer learning (TL) optimizes the adaptability of new devices, reduces the distribution difference between new device data and old device data, improves the generalization ability of the model, and makes aging prediction more accurate. During the transfer learning process, use the correlation alignment loss function (CORAL) to reduce the difference in data distribution between devices, improve the consistency of the model on different devices, and reduce the transfer error caused by the data distribution difference.

[0080] S3. Based on the prediction results, use AIDA64 to conduct host warm-up tests and high-load tests, and synchronously run the HeavenBenchmark tool to test the graphics processing ability of the computer;

[0081] Specifically, based on the prediction results, use AIDA64 to conduct host warm-up tests and high-load tests, and synchronously run the Heaven Benchmark tool. The steps to test the graphics processing capabilities of the computer are as follows:

[0082] Calculate the baseline performance offset ratio based on the prediction results :

[0083] ,

[0084] wherein is the predicted downclocking trend, is the initial CPU frequency;

[0085] Set the threshold A based on the calculation method of historical data quantiles. If the offset ratio is greater than the threshold A, then pre-lower the CPU core voltage;

[0086] Run the AIDA64 FPU test. According to the strategy of increasing the load by 10% every 5 minutes until 50% load and lasting for 30 minutes, record the CPU temperature, power consumption, and fan speed, and calculate the temperature change rate of the CPU frequency through multi-step calculations and the actual downclocking ratio :

[0087] ;

[0088] wherein is the CPU frequency change amount, is the CPU temperature change amount;

[0089] ;

[0090] wherein is the initial CPU frequency, is the CPU frequency attenuation amount;

[0091] If the actual downclocking ratio is greater than the baseline performance offset ratio, it indicates a large prediction error, and enter the BIOS adjustment mode, including further lowering the CPU core voltage and increasing the fan speed;

[0092] Continue to run the AIDA64 FPU test, increase the CPU load to 100% and maintain it for 3.5 hours, record the CPU frequency and calculate the CPU downclocking rate. Set the threshold D based on industry experience. If the CPU downclocking rate is less than the threshold D, then enter the BIOS power consumption reduction strategy to optimize voltage and power management;

[0093] Run Heaven Benchmark synchronously for GPU high - load testing (4K rendering + maximum anti - aliasing + maximum lighting rendering) for 3.5 hours;

[0094] Record the GPU frequency and calculate the GPU frequency - down trend :

[0095] ;

[0096] Wherein, is the actual operating frequency of the GPU at the current moment, is the benchmark frequency of the GPU at the beginning of the test, and t is the running time;

[0097] If the GPU frequency - down trend is less than the threshold D, then adjust the GPU frequency management strategy to optimize the power consumption mode.

[0098] Based on the prediction results, optimize the intelligent aging test. Conduct CPU and GPU load tests through AIDA64 and Heaven Benchmark, and conduct dynamic analysis in combination with the benchmark performance offset ratio, CPU frequency change rate, and GPU frequency decay rate. At the same time, optimize the CPU / GPU operating status through BIOS adjustment strategies, improve the heat dissipation efficiency, reduce power consumption, and improve the long - term stability of the device. This method is more intelligent, adaptable, and has better power consumption optimization ability compared with the traditional fixed - mode aging test.

[0099] S4. Use a quantum random number generator to generate a random load pattern, alternate high and low loads of the CPU / GPU, simulate real - user scenarios, conduct Rebooter and S3 / S4 sleep recovery tests and trigger exceptions;

[0100] Specifically, using a quantum random number generator to generate a random load pattern, alternate high and low loads of the CPU / GPU, simulate real - user scenarios, and calculate the system stability parameters after high - load testing, including the following steps:

[0101] Extract the GPU frequency curve and CPU temperature curve from the high - load test results;

[0102] Use a quantum random number generator to generate a non - linear CPU load pattern and GPU load fluctuation range;

[0103] The generation of the non-linear CPU load pattern means connecting a quantum random number generator based on single-photon detection in the test system to ensure its stable output of a random number stream, setting the sampling rate of the quantum random number generator so that it can provide random number data in real time, usually set above 1 Mbps (megabits per second) to ensure that the generation speed of random numbers meets the CPU load regulation requirements, running the self-check program of the quantum random number generator to verify whether its output meets the requirements of uniform randomness, using standard NIST randomness tests (such as single-bit frequency test, run test, block frequency test, etc.) to ensure that the generated random number sequence is qualified, obtaining the original random bit stream generated by the quantum random number generator through the API or driver, using a bit conversion algorithm to convert the binary random sequence into a floating-point random sequence W, taking 16 bits as a data block and converting it into a floating-point number between 0 and 1, and using a linear transformation to convert the data into the range of [-1, 1] so that the change of CPU load can fluctuate up or down;

[0104] Due to the possible presence of noise in the quantum measurement process, perform uniform filtering on the random numbers to ensure that the random number distribution is close to the standard uniform distribution, and use a sliding window smoothing algorithm to preprocess W so that there are no extreme deviations in the random load changes;

[0105] Extract the operating parameters of the CPU in the high-load state, including the maximum load power consumption and the reference load power consumption, use the random numbers generated by the quantum random number generator to regulate the CPU load. If the random number is greater than 0, increase the CPU load, and the amplitude is controlled by the quantum random number generator. If the random number is less than 0, decrease the CPU load, and the amplitude is also controlled by the quantum random number generator. Record the CPU load change range, and adjust the range of the quantum random number generator according to the load change range. For example, if the load change range is greater than 30% (the load change is too large), reduce the load change range of the quantum random number generator. If the load change range is less than 10%, expand the load range of the quantum random number generator;

[0106] The generation of the GPU load fluctuation range includes reading the GPU frequency and power consumption data during the operation of Heaven Benchmark and calculating the GPU power consumption volatility recorded by Heaven Benchmark;

[0107] Use the quantum random number generator to generate a set of true random numbers to control the switching of the GPU load pattern:

[0108] The GPU load during the operation of Heaven Benchmark is usually relatively stable. The random numbers generated by the quantum random number generator can simulate the uncertainty of the GPU load and improve the authenticity of the test;

[0109] Use the random numbers generated by a quantum random number generator to regulate the GPU load, record the range of GPU load changes, and perform sudden adjustments to the GPU load using Heaven Benchmark in the following modes:

[0110] High-performance mode (full GPU frequency): Simulate high-load scenarios such as gaming and deep learning training;

[0111] Power-saving mode (GPU frequency reduced by 30%): Simulate scenarios such as normal web browsing and video playback;

[0112] Sleep mode (GPU enters the lowest power consumption state): Simulate long periods of idle situations;

[0113] Generate random loads through a quantum random number generator, control the switching time of Heaven Benchmark between different modes, and ensure that the GPU load can truly reflect the changes in burst computing tasks;

[0114] Generate random numbers through a quantum random number generator, control the sudden change process of the GPU load, and simulate the sudden changes in GPU load during tasks such as game rendering and deep learning inference;

[0115] Use the actual GPU load data recorded by Heaven Benchmark as a reference parameter to ensure reasonable changes in the GPU load;

[0116] If the change range of the GPU load is too large (exceeding 25%), adjust the calculation formula of the quantum random number generator to make the GPU load more stable;

[0117] If the change range of the GPU load is too small (less than 10%), increase the fluctuation range generated by the quantum random number generator to ensure that the changes in the GPU load are more in line with the actual application scenarios;

[0118] Compare the GPU load fluctuations generated by the quantum random number generator with the GPU frequency and power consumption data recorded by Heaven Benchmark to ensure that sudden changes in the GPU load do not cause overheating or abnormal crashes. If the sudden change in the GPU load exceeds the device tolerance threshold (for example, the temperature is higher than 85°C), enter the fan speed increase mode to reduce the GPU load and optimize the thermal management;

[0119] Use the non-linear mutation mode of the quantum random number generator to perform load fluctuations once every 20 seconds (the load adjustment range is 30% - 100%, and the maximum instantaneous change amplitude ), calculate the CPU load change response time, dynamically set the threshold I based on the sliding window statistics. If the response time is greater than the threshold I, then enter the BIOS power-saving mode;

[0120] Switch the operation of Heaven Benchmark among high-performance mode, power-saving mode, and sleep mode;

[0121] Calculate the GPU frequency volatility, set a threshold p by statistically calculating the standard deviation of the volatility, and enter the BIOS power-saving mode if the volatility is greater than the threshold p.

[0122] Traditional aging tests usually adopt a fixed CPU load curve, such as linear increase or constant full load. However, the random load patterns generated by quantum random number generators can more accurately simulate the non-linear load fluctuations in actual application scenarios, such as sudden computing tasks and switches between high-load tasks. Since the load fluctuations generated by quantum random number generators are non-linear and have strong randomness, they can avoid the pattern dependence that may occur in fixed load tests, expand the scope of application of the tests, and make the test results more representative. By using a quantum random number generator to control the CPU load fluctuations and dynamically adjusting the amplitude of the fluctuations generated by the quantum random number generator according to the load change range, it is possible to ensure that the CPU load test is closer to the real operating environment and improve the accuracy of aging prediction. Heaven Benchmark usually runs with a constant load, while in actual application scenarios, the GPU load fluctuates greatly. For example, in tasks such as games and AI computing, the GPU load will change suddenly. The present invention uses the random numbers generated by a quantum random number generator to control the GPU load pattern and simulate the sudden change of the GPU load, making the test more in line with the actual usage situation. By using a quantum random number generator to control Heaven Benchmark to run in different modes (high-performance mode, power-saving mode, sleep mode), it is possible to ensure that the GPU load test not only covers high-load tasks but also includes low-power modes, making the test more comprehensive. If the GPU load fluctuation amplitude is too large, it may cause the GPU temperature to rise above the safety threshold (for example, 85°C). The present invention uses a quantum random number generator to control the GPU load, making the GPU load fluctuation conform to the set range and avoiding GPU overheating, thereby improving the long-term operation stability of the device. In many computing tasks, the CPU and GPU need to work together, such as games, video processing, and AI training. The present invention uses a quantum random number generator to control the CPU / GPU load, enabling its load pattern to be dynamically adjusted and more in line with the actual application scenario. By using a quantum random number generator to conduct CPU / GPU load switching tests and calculating the CPU load change response time, it is ensured that the CPU / GPU load switching will not cause system instability or crashes, improving the reliability of the test. Heaven Benchmark provides GPU load benchmark test data. The present invention uses a quantum random number generator to control the GPU load pattern, making the test more stable and without excessive load fluctuations. By using the GPU frequency and power consumption data recorded by Heaven Benchmark, the calculation formula of the quantum random number generator can be optimized, making the GPU load change more in line with the actual application scenario and improving the test accuracy.

[0123] Further, in order to ensure that the device can normally enter the low-power mode after high-load operation and maintain data integrity at startup, the Rebooter and S3 / S4 sleep recovery tests are conducted and abnormal triggering includes the following steps:

[0124] Conduct a standardized test environment to ensure the stability of the external power supply. Invoke the Rebooter automated test script, perform H consecutive restarts, and record the startup time for each restart. Calculate the startup time drift rate based on the maximum and minimum startup times. Set the threshold E through statistical analysis of historical data. If the drift rate is greater than the threshold E, trigger anomaly detection and enter the low-power recovery test;

[0125] After the Rebooter performs 10 restarts, invoke the ACPI (Advanced Configuration and Power Interface) instruction to put the device into the S3 sleep mode. Record the total time to enter sleep. After waiting for 180 seconds, simulate the normal user usage scenario and then press the power button or the keyboard / mouse to wake up the device. Record the time required to wake up. Run a timed calculation task (such as CPU instruction loop calculation) and calculate the CPU response latency of the policy. Set the threshold J through experiments. If the latency is greater than the threshold J, it indicates that the CPU has a delay after waking up, which may be caused by power management problems, and enter the BIOS adjustment. If the latency is less than or equal to the threshold J, it means that the CPU response time is normal, and continue to perform the S4 (hibernation) mode test;

[0126] Send the ACPI instruction to put the device into the S4 mode. The device shuts down all active processes, writes the data in the memory to the storage device, and completely shuts down the power. Calculate the time required to enter hibernation. After the device is completely powered off for 30 minutes, press the power button to trigger the S4 mode recovery, and calculate the storage data consistency. If it is inconsistent, it means that the stored data is damaged, and perform automatic repair.

[0127] The present invention introduces the Rebooter restart test, S3 / S4 sleep recovery test, and combines an automated anomaly detection mechanism, significantly improving the aging test accuracy of the microcomputer host. The traditional method only monitors the single restart time. This solution calculates the drift rate through multiple consecutive tests, enabling earlier detection of startup delays caused by storage device degradation or BIOS configuration problems, providing data support for device optimization. The traditional low-power mode test only detects whether the device can recover. This solution further quantifies key parameters such as CPU response time and storage data consistency to ensure that both power management and data integrity meet industrial standards. The traditional method only detects anomalies but does not provide optimization suggestions. When this solution detects abnormal startup time drift rate, excessive CPU response latency, or damaged storage data, it can automatically adjust the BIOS settings, optimize the device power management strategy, and improve the long-term stability of the device. If the traditional method finds damaged hibernation data, it usually requires manual repair. This solution combines the TRIM operation of the SSD, data verification, and error detection mechanism to automatically repair when damaged storage data is found, ensuring data integrity after the device recovers and improving system reliability.

[0128] S5. Comprehensively analyze all test data and provide optimization suggestions;

[0129] Specifically, comprehensively analyze all test data, evaluate the device lifespan, and provide optimization suggestions including:

[0130] Calculate the average frequencies of the CPU and GPU at the initial time and the current time point of the test, collect the SSD read / write rate, calculate the device degradation rate based on the CPU / GPU frequencies obtained at the current time point and the device running time, set a threshold F by fitting the degradation curve to historical device failure data. If the degradation rate exceeds the threshold F, it indicates that the device is severely aged and optimization measures need to be taken. The optimization measures include reducing the power consumption of the CPU and GPU, reducing overheating and frequency reduction under high load conditions, adjusting the fan strategy, improving the heat dissipation efficiency, preventing the device from accelerating aging due to long-term high temperature, optimizing the storage management, improving the SSD read / write efficiency, avoiding I / O degradation affecting the system performance, intelligently adjusting the BIOS parameters, adaptively adjusting the frequency and voltage to make the device run more stably.

[0131] By real-time monitoring the CPU / GPU frequencies and the SSD read / write rate, the downward trend of the device performance can be accurately identified, rather than relying solely on long-term test data, ensuring the real-time and reliability of the test results. Calculating the degradation rate in combination with the device running time makes the test method applicable to different usage scenarios, whether it is a long-term high-load server or a light office device. By adjusting the voltage and frequency of the CPU and GPU, it can adaptively adjust the power consumption in different load modes, reduce the hardware aging rate caused by continuous high-load operation, avoid the CPU and GPU entering the thermal throttling state, and improve the stability of computing tasks. By optimizing the fan speed curve, the fan speed is increased in advance when the CPU / GPU temperature reaches a specific threshold, reducing the impact of excessive temperature on the device hardware. By using the TRIM command to optimize the garbage collection mechanism of the storage medium, reducing the performance degradation of the SSD caused by the write amplification effect of the storage unit. By intelligently adjusting the BIOS settings, the CPU and GPU automatically adjust the voltage, power consumption, and heat dissipation strategy in different load modes, optimizing the energy efficiency ratio of the hardware.

[0132] This embodiment also provides a mini computer host aging test system, including:

[0133] A data acquisition module, used to connect the device under test to the test system, perform system standby tests, and collect initial reference data;

[0134] An aging prediction module, used to construct an aging training model using a radial basis function neural network model, and improve the generalization ability of the model by combining curriculum learning and transfer learning;

[0135] Performance testing module, used to perform high-load CPU / GPU testing through AIDA64 + Heaven Benchmark;

[0136] Mutation stress testing module, used to generate a non-linear CPU / GPU load pattern, simulate real usage scenarios, calculate the load response time, and trigger the BIOS power-saving mode;

[0137] Testing module, used to perform continuous reboot testing of Rebooter, calculate system stability, trigger anomaly detection, and optimize the sleep recovery strategy;

[0138] Optimization suggestion module, used to comprehensively analyze all test data and provide device optimization suggestions.

[0139] This embodiment also provides a computer device, applicable to the case of the aging test method for a microcomputer host, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the aging test method for a microcomputer host as proposed in the above embodiment.

[0140] This computer device can be a terminal. This computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the computer device housing, or an external keyboard, a touchpad, or a mouse, etc.

[0141] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for realizing the aging test of a microcomputer host as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, a magnetic disk or an optical disc.

[0142] In summary, the present invention combines a radial basis function neural network aging prediction model, uses curriculum learning and transfer learning for model training, intelligently analyzes the degradation trend of the CPU / GPU, and combines a quantum random number generator to generate a random load pattern to simulate the load fluctuations in the actual application scenario, thereby improving the authenticity of the test. In addition, the present invention provides BIOS optimization strategies, such as dynamically adjusting the CPU / GPU frequency, optimizing the fan strategy, and enabling the SSD TRIM operation, to achieve intelligent aging prediction and optimized control, thereby improving the long-term stability and service life of the microcomputer host.

[0143] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A microcomputer host aging test method, characterized in that: include, Connect the device under test to the test system, perform system standby testing, and collect initial benchmark data; Build an aging prediction model and optimize the model training strategy by combining curriculum learning and transfer learning; Based on the prediction results, AIDA64 was used to perform host warm-up tests and high-load tests, and the HeavenBenchmark tool was run simultaneously to test the computer's graphics processing capabilities; Use the quantum random number generator to generate random load patterns, alternate between high and low CPU / GPU loads, simulate real user scenarios, perform Rebooter and S3 / S4 sleep recovery tests, and trigger exceptions; Comprehensively analyze all test data and provide optimization suggestions; The method of constructing an aging prediction model and optimizing the model training strategy by combining curriculum learning and transfer learning includes the following steps: Use radial basis function neural network model to build aging prediction model, define curriculum learning strategy, and divide curriculum learning into three stages: basic stage, adaptation stage and generalization stage for model training; The radial basis function neural network model trained through course learning is used as the starting point of transfer learning. The first N layers of the radial basis function neural network model are frozen and remain unchanged. Only the center point of the radial basis function neural network model in the last M layers is optimized. The correlation alignment loss function is used to align data features. The radial basis function neural network model is optimized using the gradient descent method. After the training is completed, the radial basis function neural network model parameters are updated. The radial basis function neural network model trained through transfer learning is saved as the final model. After normalizing the data in the benchmark data set, the data is input into the radial basis function neural network model to obtain the aging trend of the computer host.

2. The microcomputer host aging test method according to claim 1, characterized in that: The method of using AIDA64 to perform host warm-up test and high-load test based on the prediction results, and running the HeavenBenchmark tool simultaneously to test the computer's graphics processing capability includes the following steps: Calculate the benchmark performance deviation ratio R based on the prediction results o , set threshold A. If the offset ratio is greater than threshold A, lower the CPU core voltage in advance, run AIDA64FPU test, record CPU temperature, power consumption and fan speed, and calculate the temperature change rate α and actual frequency reduction ratio β of CPU frequency through multiple steps; If the actual frequency reduction ratio is greater than the baseline performance deviation ratio, it means that the prediction error is large, and the BIOS adjustment mode is entered, including further reducing the CPU core voltage and increasing the fan speed; Continue to run the AIDA64 FPU test, record the CPU frequency and calculate the CPU frequency reduction rate, set the threshold D, if the CPU frequency reduction rate is less than the threshold D, enter the BIOS power reduction strategy, optimize the voltage and power consumption management, and simultaneously run the Heaven Benchmark for GPU high load test, record the GPU frequency and calculate the GPU frequency reduction trend, if the GPU frequency reduction trend is less than the threshold D, adjust the GPU frequency management strategy and optimize the power consumption mode.

3. The microcomputer host aging test method as claimed in claim 2, characterized in that: The method of using a quantum random number generator to generate a random load pattern, alternating high and low CPU / GPU loads, simulating real user scenarios, and calculating system stability parameters after high load testing includes the following steps: Extract GPU frequency curve and CPU temperature curve from high load test results; Use quantum random number generator to generate nonlinear CPU load patterns and GPU load fluctuation ranges; Use the nonlinear mutation mode of the quantum random number generator to calculate the response time of the CPU load change and set the threshold I. If the response time is greater than the threshold I, the BIOS power reduction mode is entered; Switch Heaven Benchmark between high load, power saving mode and sleep mode; Calculate the GPU frequency fluctuation rate and set a threshold p. If the fluctuation rate is greater than the threshold p, enter the BIOS power reduction mode.

4. The microcomputer host aging test method as claimed in claim 3, characterized in that: The Rebooter and S3 / S4 sleep recovery tests and abnormal triggering include the following steps: Conduct a standardized test environment to ensure that the external power supply is stable. Call the Rebooter automated test script to perform H consecutive restarts and record the startup time of each restart. Calculate the startup time drift rate based on the maximum startup time and the minimum startup time, set the threshold E, and if the drift rate is greater than the threshold E, trigger anomaly detection and enter the low power recovery test. After the rebooter executes x reboots, it calls the ACPI command to put the device into S3 sleep mode, calculates the policy CPU response delay, sets the threshold J, and if the delay is greater than the threshold J, enters the BIOS adjustment; if the delay is less than or equal to the threshold J, continues to perform the S4 mode test; Send ACPI instructions to put the device into S4 mode, calculate the time required to enter sleep mode, trigger S4 mode recovery, calculate the consistency of storage data, if inconsistent, it means that the storage data is damaged, and perform automatic repair.

5. The microcomputer host aging test method as claimed in claim 4, characterized in that: The comprehensive analysis of all test data and the provision of optimization suggestions include: Calculate the average frequency of the CPU and GPU at the beginning of the test and at the current time point, collect the SSD read and write rates, calculate the device degradation rate based on the CPU / GPU frequency obtained at the current time point and the device's running time, and set a threshold F. If the degradation rate exceeds the threshold F, it indicates that the device is seriously aged and optimization measures need to be taken.

6. The microcomputer host aging test method according to claim 5, characterized in that: The initial benchmark data refers to the fan speed, CPU frequency A c , temperature B c and power consumption C c And GPU frequency A G , temperature B G and power consumption C G , and the randomness measure PE0 of the device's initial state.

7. A microcomputer host aging test system, based on the microcomputer host aging test method according to any one of claims 1 to 6, characterized in that: include, A data acquisition module is used to connect the device under test to the test system, perform system standby testing, and collect initial benchmark data; Aging prediction module, which is used to build an aging training model using a radial basis function neural network model, and improve the generalization ability of the model by combining curriculum learning and transfer learning; Performance testing module, used for CPU / GPU high load testing through AIDA64+Heaven Benchmark; Mutation stress test module, used to generate nonlinear CPU / GPU load patterns, simulate real usage scenarios, calculate load response time, and trigger BIOS power reduction mode; The test module is used to perform the Rebooter continuous restart test, calculate the system stability, trigger anomaly detection, and optimize the sleep recovery strategy; The optimization suggestion module is used to comprehensively analyze all test data and provide equipment optimization suggestions.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the microcomputer host aging test method described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the microcomputer host aging test method described in any one of claims 1 to 6 are implemented.

Images