NVIDIA H200 8-GPU AI Server | 141GB HBM3E
| Brand | NVIDIA |
| Model | H200 (Hopper) |
| Configuration | 8x H200 141GB HBM3E / dual Xeon or EPYC / DDR5 |
| Warranty | 3 months - 3 years |
Overview
The NVIDIA H200 is the upgraded Hopper data-center GPU, raising memory from 80GB on H100 to 141GB HBM3E and bandwidth to 4.8TB/s, with about 1979 TFLOPS FP8 (sparse). For teams already running H100, the main advantage is a familiar architecture with nearly double the memory: larger models and longer context without rewriting model code. It remains one of the most cost-effective general-purpose platforms for both training and inference.
Key specifications
| Architecture | NVIDIA Hopper |
| Memory | 141GB HBM3E |
| Memory bandwidth | 4.8TB/s |
| FP8 compute | ~1979 TFLOPS (sparse) |
| Interconnect | NVLink at 900GB/s class |
| Form factor | SXM module / 8-GPU server |
| GPU power | ~700W |
8-GPU H200 reference configuration
An 8-GPU H200 system provides about 1.1TB of total GPU memory and roughly 15.8 PFLOPS FP8 (sparse). It can be paired with dual Intel Xeon or AMD EPYC processors, DDR5 memory and NVMe storage, and connected into multi-node clusters over InfiniBand.
| GPU | 8x NVIDIA H200 141GB HBM3E |
| Total GPU memory | ~1.1TB |
| FP8 compute | ~15.8 PFLOPS (sparse) |
| CPU | Dual Intel Xeon / AMD EPYC |
| Networking | InfiniBand / RoCE optional |
Typical workloads
Large-model training, fine-tuning and alignment
Multi-GPU inference for 70B-400B parameter models
Long-context inference, RAG and document intelligence
University research, HPC and enterprise AI platforms
Supply and delivery
H200 systems are supplied from stock or build-to-order. Standard 8-GPU SXM servers are usually available quickly, while custom CPU, memory, storage and network configurations are scheduled to order. Burn-in, GPU stress testing and memory checks are completed before shipment, and complete servers, modules or individual GPUs are all available.
Warranty and service
New complete servers carry manufacturer warranty; refurbished modules are covered for 3 to 12 months. We provide rack installation, driver and CUDA setup, NCCL fabric tuning and maintenance support. Pricing follows the market, so please request a current quote.
FAQ
Is H200 compatible with H100, and can an existing cluster be upgraded?
What switches and cables are needed to network an 8-GPU H200 cluster?
For H200 server or module pricing, please send an inquiry.
