Deep Learning has shown remarkable results in many fields.Instant parameter adjustment is substantial for a successful deep learning model. To accelerate training process of deep learning, many studies are designed to use distributed deep learning systems with multiple GPUs.
Hardware performance and utilization efficiency of multi-GPU systems are dependent on factors such as model size and amount of data. In this whitepaper, we will analyze the multi-GPU working model to identify performance bottleneck and its corresponding solutions regarding to model settings and hardware configurations. Benchmark test and Face Swap practice are also used for verification.
Download it to learn more.