NVIDIA H100 NVL – what is it?

NVIDIA H100 NVL is a high-performance GPU system configuration based on NVIDIA’s Hopper architecture, optimized for large language model workloads and generative AI with intensive inter-accelerator data exchange. In this configuration, NVIDIA H100 NVL uses the NVLink interface to interconnect GPUs into a shared compute space with high memory bandwidth, enabling efficient scaling of both inference and training for models with long context windows and demanding throughput requirements.

In practical scenarios, NVIDIA H100 NVL PCIe is used as a deployment option for integrating H100 GPUs into server platforms via the PCIe interface, providing flexibility across different infrastructure types, including cloud and enterprise data centers. Such systems are used for LLM inference, large-scale data processing, and building high-density AI services, where memory access latency, inter-GPU communication speed, and stable performance under scaling are critical.