NVL – what is it?

NVL is a designation for high-performance GPU system configurations in NVIDIA’s portfolio that use the NVLink interface to interconnect multiple GPUs into a single compute cluster with high memory bandwidth. Architecturally, NVL implies tight GPU integration using high-speed interconnects, enabling efficient scaling of compute workloads for artificial intelligence tasks, including training and inference of large language models.

Typical configurations based on the NVL approach include solutions such as NVIDIA H100 NVL, NVIDIA H200 NVL, and other server platform variants with dual- or multi-GPU interconnects via NVLink. In these configurations, multiple accelerators operate as a single logical system with unified memory and minimal data exchange latency. In practical scenarios, NVL is used to build high-density AI infrastructure where memory access latency and inter-GPU communication speed are critical. Such configurations are deployed in data centers, HPC systems, and cloud environments for large-scale model training, ensuring efficient resource utilization and stable performance when working with large volumes of data.