Uvation MarketplaceMarketplace

NVIDIA DGX Vera Rubin NVL72 Rack-Scale
AI System

A 100% liquid-cooled, 48RU rack-scale DGX system uniting
72 NVIDIA Rubin GPUs and 36 Vera CPUs with 75 TB of fast memory.

72
Rubin GPUs
36 x
Vera CPUs
75 TB
Fast memory

One Rack, 72 Rubin GPUs,
36 Vera CPUs

DGX Vera Rubin NVL72 brings Rubin GPUs, Vera CPUs, NVLink switching and NVIDIA software together in one liquid-cooled rack for training and inference.

Nine NVLink Switches Join 72 GPUs

Nine L1 NVIDIA NVLink Switches connect all 72 Rubin GPUs across the rack with 216 TB/s of NVLink bandwidth.

75 TB of Fast Memory per Rack

20.7 TB of HBM4 GPU memory at 1,400 TB/s and 54 TB of LPDDR5X CPU memory feed large models.

Confidential Computing Across the Rack

Third-generation confidential computing protects workloads across Vera CPUs, Rubin GPUs and NVLink, from compute to interconnect.

72 Rubin GPUs Under One
NVLink Fabric

Nine L1 NVLink Switches link all 72 Rubin GPUs into one rack-scale domain with 216 TB/s of NVLink bandwidth.

NVIDIA DGX Vera Rubin NVL72 Rack-Scale AI System rack architecture
72×
NVIDIA Rubin GPUs
36×
NVIDIA Vera CPUs
20.7 TB
HBM4 GPU memory
216 TB/s
NVLink bandwidth

Workloads

From frontier model training to large-scale inference, DGX Vera Rubin NVL72 runs demanding AI workloads in a single rack.

Frontier Model Training

2,520 PFLOPS of dense NVFP4 and 1,260 PFLOPS of dense FP8/FP6 training compute help teams train larger models.

Rack-Scale AI Inference

Up to 3,600 PFLOPS of NVFP4 inference and 20.7 TB of HBM4 serve large models with high throughput.

Managed AI Infrastructure

NVIDIA Mission Control supports configuration, monitoring and operations, while NVIDIA AI Enterprise and DGX OS provide the software stack.

Engineered to dominate

Rubin Compute at Rack Scale

DGX Vera Rubin NVL72 combines Rubin GPU compute, 1,400 TB/s of
HBM4 bandwidth and 800 Gb/s ConnectX-9 networking in one
liquid-cooled 48RU rack.

Exploded view of the NVIDIA DGX Vera Rubin NVL72 Rack-Scale AI System
3,600 PFLOPS
NVFP4 inference
75 TB
HBM4 + LPDDR5X
54 TB
LPDDR5X CPU Memory
32.4 TB/s
Scale-Out Bandwidth
Exploded view of the NVIDIA DGX Vera Rubin NVL72 Rack-Scale AI System
3,600 PFLOPS
NVFP4 inference
75 TB
HBM4 + LPDDR5X
54 TB
LPDDR5X CPU Memory
32.4 TB/s
Scale-Out Bandwidth

The rack, itemized

SUBSYSTEM

SPECIFICATION / CONFIGURATION DETAILS

Processor

36 × NVIDIA Vera 88-core CPUs

Operating System

NVIDIA DGX OS; Ubuntu supported

Chipset

Compute tray with 2× NVIDIA Vera Rubin Superchips (2 Rubin GPUs + 1 Vera CPU each)

Accelerators / GPU

72× NVIDIA Rubin GPUs; 9× NVLink switch trays

Memory

75 TB total fast memory: 20.7 TB HBM4 GPU memory and 54 TB LPDDR5X CPU memory

Primary Storage

Per compute tray: 1× 2 TB E1.S NVMe OS drive

Secondary Storage

Per compute tray: 4× 7.68 TB E1.S NVMe local drives, RAID 0

Security

Third-generation confidential computing across CPU, GPU and NVLink domains

Chassis

100% liquid-cooled rack; 18 compute trays and 9 NVLink switch trays

Power Supply

4× rack power shelves; each has 6× 18.3 kW PSUs and supplies up to 110 kW

Ports

144× ConnectX-9 OSFP 800 Gb/s; 18× dual-port BlueField-4 400 Gb/s; BMC RJ45

Slots / PCIe Architecture

9 × L1 NVIDIA NVLink switches; sixth-generation NVLink

Bezel

N/A

Form Factor

48RU rack-scale DGX system

Network Card

144× ConnectX-9 VPI; 18× dual-port BlueField-4 VPI

Warranty

Three-year enterprise business-standard hardware and software support

NVIDIA DGX Vera Rubin NVL72 Rack-Scale AI System power and networking

Rack Power Shelves and 800G Fabric

4× rack power shelves

Each of four power shelves holds six 18.3 kW PSUs and supplies up to 110 kW to the rack.

144× ConnectX-9 at 800 Gb/s

144 single-port ConnectX-9 VPI adapters on OSFP provide 800 Gb/s InfiniBand or Ethernet compute networking.

18× dual-port BlueField-4 at 400 Gb/s

Eighteen dual-port BlueField-4 VPI adapters add 400 Gb/s InfiniBand or Ethernet connectivity, alongside BMC RJ45 management.

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.