AWS Unveils P4d GPU Instances Powered by Nvidia A100 Chips
AWS introduces its new P4d instances featuring eight Nvidia A100 GPUs, promising up to 2.5 times faster deep learning performance at a 60% cost reduction for model training.
AWS announces the launch of its new P4d GPU instances, marking a decade of GPU-powered cloud computing. These next-generation instances feature Intel Cascade Lake processors and eight Nvidia A100 Tensor Core GPUs that connect via Nvidia’s NVLink interface. The new setup delivers up to 2.5 times the deep learning performance of the previous generation while reducing the cost of training comparable models by about 60 percent.
Each p4d.24xlarge instance boasts impressive hardware specifications, including 320 GB of high-bandwidth GPU memory, 400 Gbps networking, 96 CPU cores, 1.1 TB of system memory, and 8 TB of SSD storage. This immense power comes at a premium on-demand price of $32.77 per hour, though customers commit to reserved instances to bring costs down to under $20 per hour.
For organizations with extreme computational needs, AWS offers the ability to link over 4,000 GPUs into an EC2 UltraCluster, effectively creating a supercomputer-scale environment. Enterprise clients like Toyota Research Institute, GE Healthcare, and Aon are already testing these robust instances to accelerate their complex machine learning workloads and push the boundaries of artificial intelligence research.