AWS Introduces P4d Cloud Instances with NVIDIA A100 GPUs for HPC
Amazon Web Services releases its new EC2 P4d instances featuring NVIDIA A100 Tensor Core GPUs to handle demanding machine learning and high-performance computing workloads. The offering significantly reduces model training times through advanced hyper-scale cloud clusters.
Amazon Web Services makes its next-generation EC2 P4d instances available to all customers, featuring NVIDIA's powerful A100 Tensor Core GPUs. These new instances target advanced cloud applications that require massive computing power, including natural language processing, seismic analysis, genomics research, and object detection.
The P4d instances deliver up to 400 Gb/s of instance networking and support both Elastic Fabric Adapter and NVIDIA GPU Direct to efficiently scale high-performance computing and multi-node machine learning training. AWS deploys these instances within EC2 UltraClusters, which act as cloud supercomputers containing over four thousand A100 GPUs, petabit-scale networking, and low-latency storage.
Compared to previous instances using FP32 precision, the P4d platform reduces machine learning training times by up to three times with FP16 and six times with TF32. This update continues AWS's decade-long history of expanding its GPU cloud offerings and establishes the P4d as its most cost-effective GPU platform for both training and inference.