NVIDIA H200 141GB HBM3e Memory GPU with 4.8 TB per sec Bandwidth

The NVIDIA H200 NVL represents the latest evolution of the Hopper architecture, engineered specifically to address the most demanding large language model (LLM) workloads and generative AI tasks. As the first GPU to feature HBM3e memory, the H200 NVL establishes a new benchmark for data center performance, delivering unprecedented memory bandwidth and capacity to accelerate both training and inference for models with hundreds of billions of parameters.

This PCIe-based solution brings the power of NVIDIA's most advanced AI accelerators to mainstream server infrastructure, offering a seamless upgrade path for existing H100 deployments.

• Multi-Instance GPU (MIG) technology for secure partitioning into up to seven isolated instances
• Transformer Engine delivering up to 9x faster AI training for large language models
• FP8 precision support reaching up to 1,979 TFLOPS
• FP64 performance rated at 34 TFLOPS
• Optimized for rack-optimized data center environments

The NVIDIA H200 NVL is ideally suited for AI inference, HPC simulations, scientific computing applications, and the most demanding generative AI workloads requiring exceptional memory bandwidth and computational power.