Abstract: The cooperation of GPU and multi-core CPU can improve the efficiency of big data computing. However, users need to consider the parallel algorithm logic and collaborative computing process in the application field at the same time, which increases the programming difficulty for GPU heterogeneous collaborative computing. Therefore, based on the analysis of collaborative computing between and within GPU heterogeneous cluster nodes, the process logic is extracted, and a collaborative computing engine combining coarse and fine granularity is proposed, which automatically generates collaborative execution plans, and users only need to pay attention to the design and implementation of the algorithm in the application field. Experimental results show that the performance loss is less than 4.2% compared with the manual programming scheme. Therefore, the collaborative computing engine can be used in the development and application of GPU general computing, which can effectively improve the development efficiency of ordinary users.