Development of DT-CNN Emulator Based on GPGPU

The computation model of the DT-CNN is classified into two types. One is called the synchronous model. The other is the asynchronous model. In recent years, the graphics processing unit (GPU) is getting a lot more attention. Because the GPU has many processor cores, it appears that the GPU accelerates the computation of the synchronous model. In this paper, for evaluating computational performance of the GPU, we compare processing times of the synchronous model using the GPU with that of the synchronous / asynchronous model using the CPU. As a result, the processing time of the synchronous model using the GPU is much faster than that of the synchronous model using the CPU. On the other hand, there is a little difference between the time of the synchronous model with the GPU and that of the asynchronous model with the CPU.