Comрuting Caрability Evaluation Method of Embedded Intelligent Comрuter

Computing capability evaluation is one of the research hotspots in the field of embedded intelligent computing. It is an important means to test the execution speed and performance of intelligent computers when performing inference tasks. The rationality of its evaluation results directly affects the optimization and improvement direction of embedded intelligent computers. Due to the various optimization schemes for neural form computing and machine learning accelerators, the evaluation of embedded intelligent computers is more difficult than that of general-purpose computers. Benchmark testing is currently a commonly used evaluation method, but on resource-limited embedded devices, the reuse ability of benchmark test sets and evaluation indicators is limited, making it difficult to adapt to the diversified configuration of embedded intelligent systems. The computational intensity of the neural network models in the test set has a certain randomness, which cannot fully explore the computing potential of the device under test, and the evaluation indicators are not unified, making it difficult to compare and analyze the computing capabilities of different embedded intelligent computers. To solve the problem that embedded intelligent devices have different configurations and cannot be performance tested through fixed models, this paper proposes a neural network model generation algorithm based on neural evolution algorithm to generate neural network models that can characterize the intelligent computing capabilities of embedded intelligent devices, and inversely infer the intelligent computing capabilities of embedded intelligent computers based on the complexity of the model. Firstly, based on the Roofline theory model, the advantages of integrating computing potential mining, resource adaptation, and unified evaluation indicators are used to propose a computing capability evaluation framework that can adapt to various embedded intelligent computers, and its rationality is analyzed. Secondly, a neural network model generation algorithm with the goal of generating the maximum complexity model is proposed. By using the neural evolution algorithm, the computational intensity of the generated model approaches the upper limit of the computational intensity of the embedded intelligent computer, fully exploring the computing potential of the device under test, and making the evaluation results more objective. Then, using a fixed upper machine as a reference, the generated neural network model is cross-operated between the device under test and the upper machine. The floating-point operation per second during the two inference task executions is used as the calculation factor to provide a general formula for computing capability evaluation, which can realize the comparative analysis of computing capabilities of different embedded intelligent computers. Finally, in the Mindspore-cpu, Tensorflow-cpu, and Mindspore-ascend310 frameworks, Huawei Atlas200 is evaluated. Compared with the five neural network models commonly used in benchmark tests, the evaluation results of the neural network models generated in this paper are more reasonable, proving that the intelligent computing capability of two DaVinci cores is 42.37 times that of eight Cortex-A55 cores. The experimental results are within the theoretical expected range, indicating that the model generation method based on neural evolution algorithm proposed in this paper exhibits stable evaluation effects under different computing frameworks. In summary, the computing capability evaluation method for embedded intelligent computers proposed in this paper mainly addresses issues such as resource adaptation, computing potential mining, neural model structural changes, and unified performance evaluation indicators during the evaluation process. It can improve the accuracy of computing capability evaluation for embedded intelligent computers and shorten the product evaluation cycle.