Posted On: May 23, 2019
Starting today, Amazon EC2 P3dn instances are available in the Asia Pacific (Tokyo) AWS region. P3dn instances were first introduced in December of 2018 and feature eight NVIDIA Tesla V100 GPUs with 32GB of GPU memory each, 100 Gbps networking, 96 Intel Skylake-based vCPUs, and 768 GB of system memory that expands the P3 portfolio on the top end for faster distributed machine learning (ML) training workloads.
P3dn instances come in one size, P3dn.24xlarge, and offer twice as much GPU memory as previous P3 instances to enable ML training on higher resolution images and can accommodate larger batch sizes for more accurate training results and faster throughput. New Intel Skylake CPUs also offer Intel’s AVX-512 instruction set to help accelerate pre-processing of images. Finally, the 100 Gbps networking provides better scaling efficiency when scaling out workloads across multiple instances. These instances are great for machine learning training along with HPC jobs like 3D rendering, video transcoding, and financial risk modeling.
With this regional expansion, these instances are available in the US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and Europe (Ireland) AWS regions. P3dn instances can be purchased as On-Demand, Reserved or Spot instances.