NVL

NVIDIA H100 NVL 94GB

NVIDIA H100 NVL pairs two Hopper GPUs over NVLink with 94GB HBM3 each (188GB total on the board). Designed for LLM inference that needs more contiguous memory than a single 80GB H100, in a standard PCIe server chassis.

Specifications

Architecture

NVIDIA Hopper NVL

GPUs

2× H100 Tensor Core

GPU Memory

94GB HBM3 per GPU (188GB total)

NVLink Bridge

Integrated dual-GPU

Form Factor

PCIe dual-slot board

TDP

Up to 400W per GPU

Use case

Large LLM inference

Interconnect

PCIe Gen5 + NVLink