NVIDIA DGX SuperPOD

NVIDIA DGX SuperPOD

NVIDIA DGX SuperPOD is a high-performance AI supercomputer combining multiple DGX servers, optimized for scalable AI model training and accelerated computing with high-speed networking and simple management.

It is also the first complete solution for AI Factory and takes care of everything from servers, network infrastructure to software. With this, and with the help of NVIDIA’s AI Enterprise tools, AI applications can be developed much faster.

Exascale

Up to 1 exaFLOP / RU

Modular

Scalable architecture

AI Factory

Turnkey AI solutions

LC/AIR

Water cooling option

Services

Full NVIDIA support

Design

Reference architecture

* Parameters may vary depending on configuration.

Hardware Software NVIDIA NGC Support

System specifications

The NVIDIA DGX SuperPOD is built on NVIDIA DGX compute servers in NVIDIA DGX H200, DGX B200, GB200 NVL72, and DGX B300*(Blackwell Ultra architecture) configurations.

At the same time, all NVIDIA DGX SuperPOD systems are delivered “turnkey” – so you are left to use the system yourself. The complete installation, connection and setup of the system is handled by dedicated technicians. Management of the entire compute cluster is managed using the intuitive NVIDIA Base Command Management software, which consolidates all management and monitoring into a single location.

Here are the parameters of the individual configurations on which the entire supercomputer is built. Since the entire system is modular and scalable, it can be configured with different numbers of DGX servers to suit your exact requirements and usage.

*It is only announced

Parametr	NVIDIA DGX H200 1128 GB	NVIDIA DGX B200 1440 GB	NVIDIA DGX GB200 NVL72
GPUs	8× NVIDIA H200 SXM5 141 GB	8× NVIDIA B200 SXM5 180 GB	72× NVIDIA GB200 186 GB*
GPU memory	1128 GB	1440 GB	36× 372 GB *
CPU	Dual Intel Xeon Platinum 8480C (112 jader)	Dual Intel Xeon Platinum 8570 2.1 GHz (112 jader)	36× NVIDIA Grace CPU (2592 jader)
Výkon (FP8 tensor operace)	32 petaFLOPS	72 petaFLOPS	720 petaFLOPS
# CUDA jader	135 168	TBA	TBA
# Tensor jader	4 224	TBA	TBA
Multi-instantce GPU	56 instancí	56 instancí	504 instancí **
RAM	2 TB	až 4TB DDR5	až 17 TB LPDDR5X
HDD	OS: 2× 1.92 TB NVMe data: 30 TB (8× 3.84 TB) NVMe	OS: 2x 1.9TB NVMe M.2 Storage: 8x 3.84TB NVMe U.2	N/A
Network	8x single-port ConnectX-7 VPI 400 Gb/s InfiniBand/ 200Gb/s Ethernet 2x dual-port ConnectX-7 VPI 400 Gb/s InfiniBand/ 200Gb/s Ethernet	4x OSFP pro 8x single-port NVIDIA ConnectX-7 VPI (400Gb/s InfiniBand/Ethernet) 2x dual-port QSFP112 NVIDIA BlueField-3 DPU (400Gb/s InfiniBand/Ethernet)	4x OSFP pro 8x single-port NVIDIA ConnectX-7 VPI (400Gb/s InfiniBand/Ethernet) 2x dual-port QSFP112 NVIDIA BlueField-3 DPU (400Gb/s InfiniBand/Ethernet) 3.6 TBps NVLink
Max. spotřeba	~10,2 kW max	~ 14.3 kW max	~ 120 kW max ***
Provedení	8U rack	10U rack	1RU
Technická specifikace	Datasheet	Datasheet	Datasheet

* Maximální konfigurace GPU paměti dle NVIDIA
** Teoretická kalkulace na základě schopností a počtu grafických čipů B200
*** Odhadovaná spotřeba na základě specifikace 1,2 kW / GB200 modul

Network infrastructure

DGX SuperPOD systems have a specially designed network infrastructure built on NVIDIA Quantum network elements to provide high-speed connectivity between all compute nodes. The entire system is built on Scalable Units(SUs), which build perfectly on the existing infrastructure of the installed SuperPOD system. Below is a table with the scaling scheme of SUs, GPU nodes and InfiniBand switches.

			InfiniBand Switch Counts			Cable Counts
SU Count	Node Count	GPU Count	Leaf	Spine	Core	Node-Leaf	Leaf-Spine	Spine-Core
4	128	1024	32	16	-	1024	1024	1024
8	256	2048	64	32	-	2048	2048	2048
16	512	4096	128	128	64	4096	4096	4096
32	1024	8192	256	256	128	8192	8192	8192
56	2048	16384	512	512	256	16384	16384	16384

Complete architecture documentation

NVIDIA
DGX H200

Fifth generation rackmount supercomputer for machine learning and artificial intelligence computing, equipped with eight Hopper architecture accelerators – Nvidia H200 141GB.

More information

NVIDIA
DGX B200

The latest sixth generation of the most powerful supercomputer exclusively designed for AI training is equipped with eight Blackwell accelerators – NVIDIA B200.

More information

NVIDIA
GB200 NVL72

The powerful and efficient DGX GB200 NVL72 computing system is the perfect solution for demanding computations and AI training. The system consists of 36 compute trays with 2x B200 and 1x NVIDIA Grace CPU.

More information

Storage

Of course, the entire AI Factory also needs powerful storage. That’s why DGX SuperPOD systems are paired with a storage solution from one of the leading providers(VAST, DDN or Pure), resulting in a complete and highly optimized system for training and inferencing AI or HPC computations for which you will always have enough storage.

VAST

VAST offers unified storage with NVMe and QLC flash, optimized for AI, HPC and big data. It leverages advanced compression and deduplication.

More information

DDN

DDN specializes in high-performance storage for HPC, AI and enterprise. It offers scalable solutions with support for parallel data access.

More information

PURE

Pure provides all-flash storage with an emphasis on simplicity, performance and low latency. It is popular for enterprise environments and cloud services.

More information

DGX SuperPOD – AI Factory

The NVIDIA DGX SuperPOD offers a comprehensive AI data center as a turnkey solution for companies building an “AI Factory” that seamlessly delivers industry-leading computing power, software tools, expertise, 24/7 support and continuous innovation. With a choice of computing options, it enables any enterprise to integrate AI into their business and create innovative applications without having to struggle with the complexity of the platform.

Software

Of course, the NVIDIA DGX SuperPOD comes with an extensive software package that takes care of everything from management to automation and facilitating the deployment of AI or HPC applications. Everything is unified under the NVIDIA Base Command platform, which greatly simplifies management and control of the entire system and allows everything to be done from a single location. Considering that the entire SuperPOD system can consist of up to 32 NVIDIA DGX servers, this management approach can save a lot of time.

At the same time, NVIDIA AI Enterprise software helps you deploy and train AI models by automating time-consuming tasks from preparing your training data to inferencing the finished AI model, so you can accelerate your AI journey with NVIDIA NIM microservices and NVIDIA NeMo automated data processing tools.

NVIDIA AI Enterprise

NVIDIA AI Enterprise is a cloud-native software platform that facilitates the development and deployment of production generative AI applications. It provides a suite of tools such as NVIDIA NIM™ microservices for efficient inferencing and NVIDIA NeMo™ for data processing and model fitting.

The platform supports deployment in a variety of environments-from the cloud to data centers to edge devices and workstations. NVIDIA AI Enterprise provides enterprise-grade security, stability, and technical support, easing the transition from prototype to production.

More about AI Enterprise

NVIDIA Base Command Manager

NVIDIA Base Command Manager is a comprehensive software solution for managing AI and HPC clusters. It enables rapid deployment and end-to-end management of heterogeneous clusters in data centers, at the edge of the network, and in multi- and hybrid cloud environments. It automates provisioning and administration of clusters from a few nodes to hundreds of thousands, supports GPU-accelerated systems, and enables orchestration with Kubernetes. It integrates with NVIDIA platforms, including DGX systems, BlueField DPUs, and NVIDIA Spectrum switches, and is part of the NVIDIA AI Enterprise software suite. It provides tools for infrastructure monitoring, workload management, and Enterprise-level security and stability.

More about Base Command

NVIDIA GPU Cloud (NGC)

NVIDIA GPU Cloud (NGC) represents the repository of the most used frameworks for machine learning and deep learning applications, HPC applications, or NVIDIA GPU cards accelerated visualization. Deploying these applications is a question minutes — copying a link of the appropriate Docker image from the NGC repository, moving it on the DGX system, and downloading and running the Docker container. The individual development environments – versions of all included libraries and frameworks, settings of environment parameters – are updated and optimized by NVIDIA for deployment on DGX systems.

NVIDIA NGC Catalogue

Key benefits of NVIDIA support

The strength of the NVIDIA solution is to support the entire system. Hardware support (in case of failure of any of the components) is a matter of course. Software support for the entire environment is critical if something does not work as intended. The customer has hundreds of developers ready to help. Support is part of NVIDIA DGX purchase. It is available for 3 – 5 years and can be further extended at the end of that period.

Repair by a technician at the customer’s site

You get professional service right at your doorstep. A technician will arrive on-site, perform a diagnosis, and make the repair quickly and efficiently.

Remote diagnostics

Quick remote assistance. Our specialists will connect to your device and perform diagnostics and repairs without the need for a personal visit.

Extended software support

Provides longer access to updates, bug fixes and software enhancements. Your device stays up-to-date and protected even after warranty.

Latest updates and upgrades

Get access to the latest software and firmware releases that improve performance, security and add new device features.

Direct contact to NVIDIA support

Don’t hesitate to communicate with the NVIDIA team of experts. Quick answers, technical assistance and advice tailored to your needs.

Private NCC storage and software

Secure and private storage for your software and data. Quick access to download tools and documentation from within the NVIDIA environment.

Installation

The entire DGX SuperPOD system is delivered “turnkey” – you as a customer will be handed a fully functional and connected system ready for use. The entire installation is arranged by specially trained technicians using specialized equipment for the installation of DGX systems. In addition, of course, the DGX SuperPOD has full NVIDIA support, including a dedicated NVIDIA assistant for your system.

NVIDIA price action

NVIDIA offers special programs for DGX systems and Tesla accelerators for EDU organizations or start-up companies. Thanks to the international collaboration between NVIDIA and IBM Global Financing, preferential financing in the form of operating leases is available for DGX models.

Get a quote

M Computers
represents NVIDIA
on the Czech market in the Enterprise
area of computing
accelerators and AI systems.

Last year, it was the first company in Central and Eastern Europe to receive the highest ELITE PARTNER status, as well as two NVIDIA AI Innovator and NVIDIA AI Champion awards.

NVIDIA Deep Learning Institute

The NVIDIA Deep Learning Institute (DLI) offers both online and hands-on training for developers, data scientists, and researchers looking to solve challenging problems with deep learning and accelerated computing.

More information

Testing

Are you interested in the NVIDIA DGX SuperPOD system and would you like to deploy it at your site?
Feel free to contact us via the contact form or contact our NVIDIA product specialist directly.

Kamila Jeřábková

NVIDIA, AI solutions

+420 734 161 516

kamila.jerabkova@mcomputers.cz