The invention provides a GPU-CPU mixed source allocation system and method under a virtual cloud environment. The system comprises an injection module and an allocator. The injection module sets a hook and limits the resource occupation of progresses through a sleep manner. The allocator calls a resource allocation algorithm to obtain a resource allocation target value and transmits the resource allocation target value to the injection module. The GPU-CPU mixed source allocation system and method has the advantages that an FEA algorithm and a resource allocation frame are provided to efficiently perform the dynamic resource allocation of multiple heterogeneous mixed resources; by the resource allocation, resource allocation fairness is improved, and resource allocation efficiency is guaranteed at the same time; the fairness and efficiency of resource allocation can be evidently improved when multiple CPU-GPU multiple mixed resource demand tasks operate; the fairness of the tasks using the allocation algorithm is maximally increased by 45%; meanwhile, task operation efficiency and quality are guaranteed through the constrain conditions of the FEA algorithm.