Uvation MarketplaceMarketplace

NVIDIA GB300 NVL72 by HPE

The NVIDIA GB300 NVL72 by HPE integrates high-speed GPU interconnects, BlueField-3 DPUs, and rack-level power distribution to support large-scale AI computing.

72 NVIDIA B300 GPUs
GPUs per rack
36 NVIDIA Grace
Processors per rack
48U Rack-Scale Architecture
Form Factor

Rack-Scale Compute for Large AI Workloads

A fully integrated rack architecture combines GPU compute, CPU memory, high-speed networking, and rack-level power distribution in one system.

High-Density GPU Compute

72 NVIDIA B300 GPUs provide the compute foundation for demanding AI training and inference workloads.

High-Bandwidth Memory Architecture

Up to 20 TB of HBM3E capacity with 569 TB/s bandwidth supports data-intensive GPU workloads.

Integrated Rack Power

Eight 1U, 33 kW DC power shelves distribute power through a 50V DC bus bar

Unified NVLink GPU Architecture

The rack integrates 72 NVIDIA B300 GPUs with nine NVLink Switch trays, creating a high-density architecture for large-scale AI computing.

Rack-scale NVLink GPU architecture in a darkened data hall
72 NVIDIA B300
GPUs per Rack
36 NVIDIA Grace
Processors per Rack
Up to 20 TB HBM3E
Total GPU Memory
569 TB/s HBM3E
Bandwidth

Workloads

Designed for compute-intensive AI workloads that require high GPU density, high-bandwidth memory, and integrated rack-scale infrastructure.

Large-Scale AI Model Training

Connects 72 NVIDIA Blackwell Ultra GPUs via NVLink, accelerating training and fine-tuning for massive models over one trillion parameters.

AI Inference

Unifies 72 GPUs with high-speed ConnectX-8 networking, eliminating data bottlenecks to power real-time reasoning and ultra-fast inference.

Data-Intensive AI Computing

Combines 20 TB HBM3E memory and 569 TB/s bandwidth with Grace CPUs to handle massive datasets and extreme computing workloads.

Engineered to dominate

Rack-Scale AI with
Unified GPU Connectivity

Designed for compute-intensive AI workloads that require high GPU density, high-bandwidth memory, and integrated rack-scale infrastructure.

Exploded view of the NVIDIA GB300 NVL72 rack
72× NVIDIA Blackwell Ultra
GPUs
36× NVIDIA
Grace CPUs
Up to 20 TB HBM3E
total GPU memory
130 TB/s aggregate
NVLink bandwidth
Exploded view of the NVIDIA GB300 NVL72 rack
72× NVIDIA Blackwell Ultra
GPUs
36× NVIDIA
Grace CPUs
Up to 20 TB HBM3E
total GPU memory
130 TB/s aggregate
NVLink bandwidth

The rack, itemized.

SUBSYSTEM

SPECIFICATION / CONFIGURATION DETAILS

Processor

36× NVIDIA Grace CPUs per rack

Operating System

Ubuntu 24.04

Chipset

ASPEED 2600 management controller, based on NVIDIA OpenBMC.

Accelerators / GPU

72× NVIDIA B300 GPUs

Memory

18.432 TB installed / 17.28 TB usable CPU memory Up to 20 TB HBM3E with 569 TB/s bandwidth

Primary Storage

1 × 1.92TB M.2 SSD

Secondary Storage

8 × E1.S NVMe SSDs

Security

Secure Boot, Hardware Root of Trust, TPM, Signed Firmware, Secure Firmware Updates

Chassis

48U rack; 18 compute trays; 9 NVLink Switch trays

Power Supply

8x 1U 33 kW (6x 5.5 kW PSUs) DC Power Shelves with 50V DC Bus Bar

Ports

FIO (RJ45 1 GbE port, USB 3.0, USB 2.0 UART, Mini Display Port); BlueField-3 DPUs (RJ45 BF3 BMC port, 2x QSFP); ConnectX-8 IO Board (2x OSFP)

Slots / PCIe Architecture

1× BlueField-3 DPU + 2× ConnectX-8 I/O modules

Bezel

N/A

Form Factor

48U rack-scale system

Network Card

Per compute tray: 2 × NVIDIA ConnectX-8, 800 Gbps I/O modules and 1 × NVIDIA BlueField-3 B3240 DPU

Warranty

3/3/3 warranty - three years parts, three years labor, and three years onsite support

HPE rack with 50V DC power fabric

Power Fabric

8 × 1U DC power shelves

33 kW per power shelf

50V DC bus bar

Nuclear-Ready, Plug and Play AI Factories

Bypass grid constraints with self-contained, high-density inference ready infrastructure directly adjacent to clean, base load nuclear power.

Global Deployment Engine

Any Infra, Anywhere in the world.

Global Shipping and Deployment Capability treats advanced infrastructure as mobile, modular freight.

Packaged complex computing nodes, cooling manifolds, and power distribution units into certified, standard form factors.

360° Service Ecosystem

Unifying Legacy with Next-Gen Compute.

Vertically integrated service ecosystem that serves as an operational bridge between legacy environments and next-generation, AI-native infrastructure.

Finance and Leasing

Capital-as-a-Service for High-Density Compute

Convert upfront capital expenditures (CapEx) into predictable operating expenses (OpEx).

Uvation allows enterprises and sovereign entities to deploy multi-megawatt AI factories and advanced GPU clusters without liquidity constraint.