Amazon Web Services has announced the general availability of Amazon Elastic Compute Cloud (Amazon EC2) P5en instances, which are equipped with NVIDIA H200 Tensor Core GPUs and custom fourth-generation Intel Xeon Scalable processors. P5en instances aim to enhance performance for machine learning training and inference workloads by providing significantly faster memory bandwidth and improved throughput between CPU and GPU.

The new instances support up to 3200 Gbps of network bandwidth through the Elastic Fabric Adapter (EFA) version 3, resulting in a latency improvement of up to 35% compared to the previous P5 generation. This enhancement benefits collective communication performance for distributed training tasks, particularly in deep learning, generative AI, and high-performance computing (HPC) applications.

Previously, on September 9, AWS launched Amazon EC2 P5e instances, which are powered by eight NVIDIA H200 GPUs and third-generation AMD EPYC processors. P5e instances deliver similar network performance with EFA version 2.

With the introduction of P5en instances, users can expect better overall efficiency in GPU-accelerated applications by reducing inference and network latency while also enhancing local storage performance and Amazon Elastic Block Store (EBS) bandwidth.

To utilize Amazon EC2 P5en instances, users can access them in the US East (Ohio), US West (Oregon), and Asia Pacific (Tokyo) regions. The instances are available through EC2 Capacity Blocks for ML, On Demand, and Savings Plan purchase options.

To reserve EC2 Capacity Blocks, customers should select the Capacity Reservations option in the Amazon EC2 console of the available AWS regions. By choosing Purchase Capacity Blocks for ML and specifying the required capacity, users can reserve instances for a range of durations from 1 to 28 days.

Following the reservation process, users can launch P5en instances using the AWS Management Console, AWS Command Line Interface (CLI), or AWS SDKs. A sample AWS CLI command is provided for users looking to run multiple P5en instances optimized for EFAv3 benefits.

For containerized ML applications, AWS provides Deep Learning AMIs and AWS Deep Learning Containers that are compatible with P5en instances.

Amazon EC2 P5en instances are now available across the specified AWS regions. For details regarding pricing and to explore the new instances, visit the Amazon EC2 pricing page.

Users are encouraged to test the P5en instances via the Amazon EC2 console and may provide feedback through AWS re:Post or local AWS support contacts.