Professional HardwareMore Performance

New · Teklif ile tedarik

NVIDIA DGX GB200

Brand: NVIDIA SKU: AI-NVIDIA-DGX-GB200

The NVIDIA DGX GB200 is a rack-scale AI system that combines 72 Blackwell GPUs and 36 Grace CPUs in a single NVLink domain. Designed for large language model training and real-time trillion-parameter inference, it requires liquid cooling and high-power infrastructure, so it is planned together with a data centre project and offered by corporate quotation.

  • 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 GB200 is a rack-scale AI system that combines 72 Blackwell GPUs and 36 Grace CPUs in a single NVLink domain. Designed for large language model training and real-time trillion-parameter inference, it requires liquid cooling and high-power infrastructure, so it is planned together with a data centre project and offered by corporate quotation.

NVIDIA DGX GB200 overview

The DGX GB200 is a DGX-branded rack system based on the GB200 NVL72 architecture and is the Grace Blackwell-generation building block of NVIDIA DGX SuperPOD.

Hardware architecture

Each rack contains 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs, forming a 72-GPU NVLink domain. According to NVIDIA's rack data, total GPU memory is approximately 13.4 TB. The system is liquid-cooled; detailed network, power and storage configuration is determined in the project according to NVIDIA's DGX GB rack-scale system guide.

Use cases in AI workloads

LLM training, low-latency inference on trillion-parameter models, large-scale recommendation systems and scientific AI.

Business advantages

The DGX GB200 comes with NVIDIA's rack-level validated software and management tools, standardising installation and operation.

Scalability and infrastructure requirements

It requires liquid cooling, high power per rack and a dedicated data centre design; third-party sources put rack power at approximately 120–132 kW (NVIDIA's official figure could not be verified in this work). Multi-rack installations follow the DGX SuperPOD architecture.

Who is the NVIDIA DGX GB200 for?

Suitable for cloud providers building rack-scale AI infrastructure, telecom and financial institutions, and national-scale research centres.

Key features

  • 72 × NVIDIA Blackwell GPU, 36 × NVIDIA Grace CPU
  • Single 72-GPU NVLink domain
  • ~13.4 TB GPU memory per rack
  • Liquid-cooled rack-scale design
  • DGX SuperPOD building block

Technical Specifications

MarkaNVIDIA
ModelDGX GB200 (NVL72 rack-scale system)
Product typeRack-scale liquid-cooled AI system
MimariNVIDIA GB200 NVL72 (Grace Blackwell)
GPU72 × NVIDIA Blackwell
CPU36 × NVIDIA Grace
GPU memory (rack)~13,4 TB
CoolingLiquid cooling
Form factorRack-scale system
Rack powerCould not be verified with the manufacturer in this work; determined in the project design
SoftwareNVIDIA DGX software stack

Procurement and corporate quotation process

The NVIDIA DGX GB200 is supplied on a project basis: first your data centre's power, cooling and network infrastructure are assessed together, then a corporate quotation covering the number of racks, installation and support scope is prepared. Price and delivery schedule are finalised in the quotation. Stating your intended use and delivery country in the quote form speeds up the process.

Similar Products