NVIDIA H200 NVL Tensor Core GPU

The NVIDIA H200 NVL is the PCIe form of the Hopper H200, designed for air-cooled enterprise servers that cannot accept SXM baseboards. Each dual-slot card carries 141 GB of HBM3e with 4.8 TB/s of bandwidth, roughly 1.5 times the memory of the H100 NVL, and two or four cards can be linked with NVLink bridges at 900 GB/s per GPU. Peak FP8 Tensor Core throughput is 3,341 TFLOPS with sparsity, the card can be partitioned into up to seven MIG instances of 16.5 GB, and board power is configurable up to 600 W.

The card is aimed at enterprises that run inference and fine-tuning in standard 2U to 4U rack servers from Dell, HPE, Lenovo, Supermicro or Cisco rather than in dedicated GPU chassis. Common deployments pair four H200 NVL cards in one server for 70B-class models, or spread single cards across virtualised hosts. NVIDIA includes a five-year NVIDIA AI Enterprise subscription with the H200 NVL, which covers NIM microservices, Triton Inference Server and the supported software stack.

InTechnologies FZCO supplies H200 NVL cards and pre-configured servers to companies in Dubai, across the UAE and in neighbouring markets. Our licensed partners issue the contract and invoice for GPU hardware, and InTechnologies looks after sourcing, consolidation in Dubai and DDP or DDU delivery. H200 NVL is subject to US export regulations, so end-user verification and a destination check are required before we confirm availability.