METHOD FOR GPU MEMORY MANAGEMENT FOR DEEP NEURAL NETWORK AND COMPUTING DEVICE FOR PERFORMING SAME

摘要

Embodiments disclosed in the present specification relate to a method for GPU memory management for a deep neural network and a computing device for performing the same and, more particularly to, a GPU memory management method, which observes the performance of a deep neural network of a GPU to reduce the size of data to be input to a GPU memory, thereby boosting the performance of the GPU, and a computing device for performing the same. An embodiment provides a method for GPU memory management for a deep neural network, performed by a computing device including a GPU and a CPU, the method comprising the steps of: generating a GPU memory management schedule on the basis of the GPU's processing of a unit operation included in the deep neural network; and transferring, on the basis of the schedule, data required for the performance of the deep neural network, between the GPU memory and the CPU memory.