Professional HardwareMore Performance

New · Teklif ile tedarik

NVIDIA DGX H200

Brand: NVIDIA SKU: AI-NVIDIA-DGX-H200

The NVIDIA DGX H200 is an enterprise AI system that accelerates training of large models and long-context inference with eight H200 Tensor Core GPUs and 1,128 GB of total GPU memory. It combines the mature software ecosystem of the Hopper architecture with higher memory capacity than the H100, making it a strong option for organisations seeking compatibility with existing Hopper-based infrastructure.

  • Garanti ve servis koşulları teklif aşamasında üretici ve tedarik kanalına göre netleşir.
  • Delivery and installation terms are finalised at the quotation stage
  • Your card details are not stored on our site

The NVIDIA DGX H200 is an enterprise AI system that accelerates training of large models and long-context inference with eight H200 Tensor Core GPUs and 1,128 GB of total GPU memory. It combines the mature software ecosystem of the Hopper architecture with higher memory capacity than the H100, making it a strong option for organisations seeking compatibility with existing Hopper-based infrastructure.

NVIDIA DGX H200 overview

The DGX H200 is a DGX system using H200 GPUs, the increased-memory version of the Hopper architecture. Because it runs on the same DGX H100 chassis and software stack, it can be added seamlessly to Hopper-based clusters.

Hardware architecture

The system has 8 × NVIDIA H200 Tensor Core GPUs (141 GB per GPU, 1,128 GB in total). Two Intel Xeon Platinum 8480C processors provide a total of 112 cores (2.00 GHz base, 3.80 GHz max); system memory is 2 TB. According to NVIDIA, maximum system power is approximately 10.2 kW.

Use cases in AI workloads

LLM inference requiring long context windows, model fine-tuning, training with large datasets, HPC and scientific computing.

Business advantages

The Hopper architecture's long-established CUDA, NVIDIA AI Enterprise and framework support allow existing models to be moved without changes.

Scalability and infrastructure requirements

It requires approximately 10.2 kW of maximum power and rack installation; multiple systems can be scaled in a DGX BasePOD / SuperPOD architecture over InfiniBand networking.

Who is the NVIDIA DGX H200 for?

Suitable for organisations already using Hopper (H100) infrastructure that need more GPU memory in the same software environment.

Key features

  • 8 × NVIDIA H200 Tensor Core GPU
  • 1,128 GB total GPU memory (141 GB per GPU)
  • 2 × Intel Xeon Platinum 8480C, 112 cores in total
  • 2 TB system memory
  • Approximately 10.2 kW maximum power
  • Same platform and software stack as the DGX H100

Technical Specifications

MarkaNVIDIA
ModelDGX H200
Product typeRaf tipi yapay zekâ sunucusu
GPU8 × NVIDIA H200 Tensor Core GPU
Total GPU memory1,128 GB (141 GB per GPU)
Processor2 × Intel Xeon Platinum 8480C, 112 cores in total, 2.00 GHz base / 3.80 GHz max
System memory2 TB
Maximum system power~10.2 kW (standard configuration)
GPU interconnectNVLink, NVSwitch
SoftwareNVIDIA DGX OS, NVIDIA AI Enterprise
Storage and networkingProvided with the current NVIDIA configuration at the quotation stage

Procurement and corporate quotation process

The NVIDIA DGX H200 is supplied by special order. Once quantity, network and storage configuration, installation and NVIDIA support options are determined, a corporate quotation including price and delivery time is prepared. Stating your intended use and delivery country in the quote form speeds up the process.

Similar Products