This article announces the general availability of Amazon EC2 P5e instances powered by NVIDIA H200 Tensor Core GPUs, designed for high-performance deep learning, generative AI, and HPC workloads.

Specifically, the article covers:

• Overview of P5e instances with 8 NVIDIA H200 GPUs, 1128 GB of GPU memory, and other high-performance specifications
• Upcoming P5en instances with improved CPU-GPU bandwidth and lower network latency for ML workloads
• Key benefits of P5e instances for large language model (LLM) inference, including higher memory bandwidth, GPU memory capacity, and support for larger batch sizes
• Performance and cost improvements for deploying large LLMs like Meta Llama 3.1 70B and 405B on P5e instances compared to P5 instances
• Suitability of P5e instances for memory-intensive HPC applications
• Getting started with P5e instances using AWS Deep Learning AMIs and AWS Deep Learning Containers
• Conclusion: P5e instances now available in US East (Ohio) region for ML and HPC workloads