AWS EC2 P6-B200

AWS EC2 P6-B200 instances with NVIDIA B200 GPUs are now generally available for high-performance AI, ML, and HPC workloads

These instances are ideal for large-scale distributed AI training, multimodal inference, and HPC applications like drug discovery and climate modeling

P6-B200 instances deliver up to 2x the AI training and inference performance of previous P5en instances

Each instance features 8 NVIDIA B200 GPUs (1440 GB HBM3e GPU memory), 2 TiB system memory, 30 TB local NVMe SSD storage, and 5th Gen Intel Xeon Scalable CPUs

They offer up to 125% more GPU TFLOPs, 27% more GPU memory, and 60% more GPU memory bandwidth than P5en

High-speed networking includes 1800 GB/s GPU-to-GPU NVLink and 3.2 Tbps EFAv4 for efficient distributed training

AWS Nitro System provides enhanced security, stability, and live firmware updates for reduced downtime

Amazon FSx for Lustre and Amazon S3 integration enable high-throughput, scalable storage for large datasets

EC2 UltraClusters allow petabit-scale networking, scaling to tens of thousands of GPUs for hyperscale AI workloads

EC2 Capacity Blocks for ML let users reserve P6-B200 capacity for 1–182 days, with upfront payment and flexible scheduling