Uvation MarketplaceMarketplace

NVIDIA HGX B300
AI Server

The NVIDIA HGX B300 AI Server puts eight Blackwell Ultra GPUs into one air-cooled system. Fifth-generation NVLink at 14.4 TB/s at 2.1 TB of HBM3e memory.

8×
NVIDIA B300 Blackwell Ultra SXM
2.1 TB
HBM3e
14.4
TB/s NVLink Bandwidth

A Supercomputer Purpose-Built for AI and HPC

The NVIDIA HGX™ platform provides the foundation for the world’s most powerful AI and high-performance computing (HPC) systems. Built on NVIDIA Blackwell Ultra, HGX B300 scales training and inference efficiently.

Eight GPUs Linked in One Compact Platform

Eight Blackwell Ultra GPUs, linked at 1.8 TB/s each, fit in one 8–10 RU system, saving valuable rack space.

Massive Memory and Bandwidth for Larger Models

2.1 TB of shared HBM3e and 14.4 TB/s NVLink bandwidth let you train and serve larger models on fewer servers.

Blackwell Ultra Built for AI Reasoning

Blackwell Ultra GPUs deliver 2x the attention-layer performance of HGX B200, accelerating reasoning models, long-context inference and large-scale AI training.

Eight GPUs Working as
One Accelerator

Fifth-generation NVLink and NVSwitch connect eight Blackwell Ultra GPUs, letting them share 2.1 TB of memory as one unified domain.

NVIDIA HGX B300 baseboard with eight Blackwell Ultra SXM GPUs
8×
NVIDIA Blackwell Ultra SXM GPUs
2.1 TB
HBM3e total GPU memory
1.8 TB/s
NVLink bandwidth per GPU
14.4 TB/s
total NVLink bandwidth

A platform for models that refuse to fit the old shape.

From frontier model training to real-time inference, HGX B300 handles the most demanding AI and HPC workloads at data-center scale.

Large Language Model Training at Scale

72 petaFLOPS of sparse FP8 compute and 2.1 TB HBM3e help teams train larger models in less time.

Fast Reasoning and Long-Context AI Inference

108 petaFLOPS of dense FP4 compute and 2x attention performance over HGX B200 speed up reasoning and long-context responses.

Scientific Simulation and High-Performance Computing Workloads

10 teraFLOPS of FP64 and 600 teraFLOPS of FP32 support simulation, research and data analytics alongside AI on one platform.

Engineered to dominate

Engineered to Dominate
Every AI Workload

HGX B300 combines Blackwell Ultra compute, NVLink 5 bandwidth, and 800 Gb/s networking to deliver peak performance for AI factories.

Exploded view of the NVIDIA HGX B300 GPU assembly
144 PFLOPS
FP4 (sparse) / 108 PFLOPS (dense)
72 PFLOPS
FP8/FP6 Tensor Core (sparse)
2x
attention performance vs. HGX B200
1.6 TB/s
total networking bandwidth
Exploded view of the NVIDIA HGX B300 GPU assembly
144 PFLOPS
FP4 (sparse) / 108 PFLOPS (dense)
72 PFLOPS
FP8/FP6 Tensor Core (sparse)
2x
attention performance vs. HGX B200
1.6 TB/s
total networking bandwidth

The rack, itemized

SUBSYSTEM

SPECIFICATION / CONFIGURATION DETAILS

Processor

Not specified (standalone baseboard platform; host CPUs provided by system OEMs)

Operating System

Not specified

Chipset

NVIDIA HGX B300 8-GPU SXM baseboard

Accelerators / GPU

8× Blackwell Ultra SXM; 144 sparse / 108 dense PFLOPS FP4; 72 sparse PFLOPS FP8/FP6; 5th-Gen NVLink; 14.4 TB/s aggregate NVLink bandwidth

Memory

2.1 TB total GPU memory (HBM3e; 288 GB per GPU)

Primary Storage

Not specified

Secondary Storage

Not specified

Security

Not specified

Chassis

HGX B300 GPU baseboard; OEM chassis required

Power Supply

Not specified

Ports

800Gb/s network interfaces through ConnectX-8; physical ports OEM-dependent

Slots / PCIe Architecture

5th-generation NVIDIA NVLink / NVSwitch platform; host PCIe layout OEM-dependent

Bezel

Not specified

Form Factor

8-GPU SXM baseboard

Network Card

8 × ConnectX-8 SuperNICs (platform design)

Warranty

Not specified

NVIDIA HGX B300 GPU baseboard

Power and Networking Built
for AI Scale

~14.5 kW Max System Power

Delivers full Blackwell Ultra performance within roughly 14.5 kW at peak, letting you plan rack power with confidence.

8× 800 Gb/s ConnectX-8 SuperNICs

Each GPU gets its own 800 Gb/s SuperNIC for fast, low-latency multi-node training and inference.

2× BlueField-3 DPUs, 400 Gb/s Per Port

Two dual-port BlueField-3 DPUs offload networking, storage and security tasks, keeping CPUs and GPUs focused on AI.

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.