Uvation MarketplaceMarketplace

NVIDIA Vera Rubin NVL4 Module with 4 Rubin GPUs and 2 Vera CPUs

The NVIDIA Vera Rubin NVL4 Module combines four Rubin GPUs and two Vera CPUs with NVLink-C2C and sixth-generation NVLink. Designed for liquid-cooled NVIDIA MGX servers, it accelerates AI-for-Science, scientific simulations, and inference workloads.

4×
NVIDIA Rubin GPUs
2×
NVIDIA Vera CPUs
14.4 TB/s
NVIDIA NVLink

Unified CPU-GPU Compute for Advanced AI

Bring GPU acceleration and CPU processing together in a tightly interconnected module built for demanding scientific and AI workloads

High-Density GPU Compute

Four Rubin GPUs provide accelerated processing for complex scientific and AI tasks.

Integrated CPU Processing

Two Vera CPUs support the compute demands of data-intensive workloads.

High-Speed Chip Interconnect

NVLink and NVLink-C2C connect the GPU and CPU components.

A Closely Connected AI
Compute Domain

The NVL4 architecture links four Rubin GPUs through a second-generation NVLink bridge and connects them to two Vera CPUs over NVLink-C2C, supporting coordinated processing across the module.

NVIDIA Vera Rubin NVL4 Module with 4 Rubin GPUs and 2 Vera CPUs architecture
4×
Rubin GPUs
2×
Vera CPUs
3.6 TB/s
NVLink Gen 6 connector bandwidth
Compatible
with liquid-cooled NVIDIA MGX servers

Workloads

Built for workloads that demand substantial parallel compute, close CPU-GPU coordination, and fast execution of complex AI and scientific tasks.

Scientific Simulation

176 custom NVIDIA Olympus cores across dual Vera CPUs and 14.4 TB/s of NVLink bandwidth enable coordinated processing and extreme data throughput for complex HPC modeling and multi-physics research

Post-Training

Up to 200 petaFLOPS of sparse NVFP4 compute and 14.4 TB/s of NVLink domain bandwidth across the module accelerate alignment, fine-tuning, and reinforcement learning for domain-specific models

High-Throughput AI Inference

Up to 200 petaFLOPS of NVFP4 compute, 1,161 tokens/sec aggregate decode throughput, and 1.88M tokens/sec aggregate prefill throughput accelerate real-time reasoning and large-scale model serving across the module

Engineered to dominate

Engineered for Accelerated
Scientific Computing

NVL4 combines Rubin GPU acceleration, Vera CPU processing, and high-speed interconnects in a module designed for demanding AI workloads.

Exploded view of the NVIDIA Vera Rubin NVL4 Module with 4 Rubin GPUs and 2 Vera CPUs
200 PFLOPS
NVFP4 Tensor Core Inference Performance (Dense)
1.1 TB
HBM4 GPU Memory (88 TB/s Bandwidth)
Up to 8×
performance for AI-for-Science inference versus Hopper
176
Custom NVIDIA Olympus Cores across Dual Vera CPUs
Exploded view of the NVIDIA Vera Rubin NVL4 Module with 4 Rubin GPUs and 2 Vera CPUs
200 PFLOPS
NVFP4 Tensor Core Inference Performance (Dense)
1.1 TB
HBM4 GPU Memory (88 TB/s Bandwidth)
Up to 8×
performance for AI-for-Science inference versus Hopper
176
Custom NVIDIA Olympus Cores across Dual Vera CPUs

The rack, itemized

SUBSYSTEM

SPECIFICATION / CONFIGURATION DETAILS

Processor

2 × NVIDIA Vera 88-core CPUs

Operating System

N/A

Chipset

4× Rubin GPUs and 2× Vera CPUs connected over NVLink-C2C

Accelerators / GPU

4× NVIDIA Rubin GPUs; sixth-generation NVLink via second-generation NVLink bridge

Memory

Up to 4.1 TB fast-access memory: 1.1 TB HBM4 and up to 3 TB LPDDR5X

Primary Storage

N/A

Secondary Storage

N/A

Security

N/A

Chassis

Vera Rubin NVL4 compute module; OEM chassis required

Power Supply

N/A

Ports

N/A

Slots / PCIe Architecture

6th-generation NVLink bridge; NVLink-C2C CPU interconnect

Bezel

N/A

Form Factor

Module for liquid-cooled NVIDIA MGX modular servers—not a complete server

Network Card

N/A

Warranty

N/A

NVIDIA Vera Rubin NVL4 Module with 4 Rubin GPUs and 2 Vera CPUs power and networking

Power Fabric

2 MW TDP per Scalable Unit

Comprising 8 DGX Vera Rubin NVL72 liquid-cooled racks

4 power shelves

per DGX Vera Rubin NVL72 rack

Up to 110 kW per power shelf

using 6 × 18.3 kW power supply units

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.