Research on Testing Methods for China-Made Heterogeneous Computing Power Based on Parameter Surface Network

With the increasing demand for domestic heterogeneous computing power platforms in fields such as artificial intelligence and big data, traditional performance evaluation methods struggle to effectively reflect the performance of multiple computing units working collaboratively. This study proposes a testing method based on the Parameter Surface Network (PSN), which constructs a multi-dimensional parameter space to evaluate the performance of computing units in collaborative operations. The method integrates dynamic scheduling and hardware resource optimization, utilizing optimization algorithms to simulate performance responses under different hardware configurations, thereby providing more accurate performance evaluations. Experimental results show that the PSN method effectively optimizes resource scheduling, load balancing, and data transmission efficiency, offering theoretical support and practical guidance for performance testing and optimization of domestic heterogeneous computing power platforms.