Amazon cloud services (AWS) launched a new P4 instance based on GPU today, ten years after the company’s first GPU cluster instance was launched. Specifically, the P4 instance uses the Intel cascade Lake processor + 8-channel NVIDIA 100 tensorflow core GPU. AWS promises that the deep learning performance provided by the new instance is 2.5 times higher than that of the previous generation, and the cost-effectiveness of model training also has a 60% advantage. < / P > < p > the price of the AWS P4 GPU instance is $32.77/hour for on-demand use. If it is used for a long time, the average annual cost will be & lt; 20 dollars / hour (average $11.57/hour in three years). < / P > < p > in extreme cases, users can choose to combine 4000 + GPUs into an EC2 ultracluster cluster. However, considering the cost of actual calculation, only a few enterprise customers can afford it, such as Toyota Research Institute, GE Healthcare and AON, which participated in the early test. Global Tech