Amazon EC2 P6e-GB200 UltraServers Launch: NVIDIA Grace Blackwell GPU-Based for Top AI Performance

Amazon Elastic Compute Cloud (EC2) P6e-GB200 UltraServers have been officially launched, offering the best GPU performance for AI learning and inference with NVIDIA GB200 NVL72 acceleration. The servers can connect multiple EC2 instances with dedicated high-bandwidth, low-latency interconnects for fast and efficient data processing. EC2 P6e-GB200 UltraServers can utilize up to 72 NVIDIA Blackwell GPUs in one NVLink domain, providing 360 petaflops (FP8, sparsity not applied) operation performance and 13.4TB of high-bandwidth memory (HBM3e). The servers are designed for high-computing and memory-intensive AI workloads, such as training and inference of advanced AI models, and can be used to build various applications like question answering, code generation, video and image generation, and voice recognition.

Amazon Web Services (AWS) has announced the general availability of Amazon Elastic Compute Cloud (EC2) P6e-GB200 UltraServers, designed to deliver unparalleled GPU performance for AI training and inference. These servers leverage NVIDIA GB200 NVL72 acceleration, enabling up to 72 NVIDIA Blackwell GPUs in one NVLink domain, which translates to 360 petaflops of FP8 compute and 13.4TB of high-bandwidth memory (HBM3e) [1].

The EC2 P6e-GB200 UltraServers are optimized for high-computing and memory-intensive AI workloads, such as training and inference of advanced AI models. They can be used to build various applications, including question answering, code generation, video and image generation, and voice recognition [2].

Key features of the P6e-GB200 UltraServers include:
- High GPU Performance: Up to 72 NVIDIA Blackwell GPUs in one NVLink domain, providing 360 petaflops of FP8 compute.
- High Bandwidth Memory: 13.4TB of high-bandwidth memory (HBM3e).
- Low Latency Communication: Dedicated, high-bandwidth, and low-latency interconnects for fast and efficient data processing.
- Scalability: Suitable for deploying large-scale AI models, such as those at the trillion-parameter scale.
- Liquid Cooling: Enhanced cooling systems to support high compute density in large NVLink domain architectures.

These servers are available in the Dallas Local Zone (us-east-1-dfw-2a) through EC2 Capacity Blocks for ML and can be integrated seamlessly with various AWS managed services, such as Amazon SageMaker Hyperpod and Amazon Elastic Kubernetes Services (Amazon EKS) [1].

References:
[1] https://aws.amazon.com/blogs/aws/new-amazon-ec2-p6e-gb200-ultraservers-powered-by-nvidia-grace-blackwell-gpus-for-the-highest-ai-performance/
[2] https://insidehpc.com/2025/07/aws-announces-ai-servers-with-nvidia-blackwell/
[3] https://www.benzinga.com/markets/tech/25/07/46342110/amazon-unveils-powerful-new-ai-servers-to-support-nvidias-most-advanced-chips