🏗️
72-GPU NVLink Domain
All 72 Blackwell GPUs communicate at 130 TB/s via the largest NVLink Switch System ever built — enabling trillion-parameter models to run as if on a single massive GPU.
UK’s trusted IT infrastructure partner since 2003
72 GPUs
Blackwell per rack
30× Faster
LLM inference vs H100
130TB/s NVLink
Rack bandwidth
25× Efficiency
Better than H100/watt
✅ Authorised Reseller
📦 Free UK Delivery
🔒 Full Warranty
📞 24/7 Support
Download Official Datasheet
PDF · Official manufacturer document
NVIDIA DGXSKU: NVIDIA-GB200-NVL72
NVIDIA GB200 NVL72 — Rack-Scale Grace Blackwell AI Supercomputer
The Exascale Computer in a Single Rack.
NVIDIA GB200 NVL72 connects 72 Blackwell GPUs and 36 Grace CPUs in a single fully liquid-cooled rack — delivering 130 TB/s NVLink bandwidth, 30× faster LLM inference vs H100, and 720 PFLOPS FP8 training in the most powerful AI rack system ever built.
- 72 NVIDIA Blackwell GPUs + 36 Grace CPUs in one liquid-cooled rack
- 130 TB/s NVLink Switch System — all 72 GPUs act as one massive GPU
- 30× faster real-time trillion-parameter LLM inference vs NVIDIA H100
- 4× faster LLM training at scale vs NVIDIA H100
- 25× better energy efficiency vs H100 air-cooled infrastructure
- 13.4 TB total HBM3e GPU memory — 576 TB/s aggregate bandwidth
- 720 PFLOPS FP8 / 1,440 PFLOPS FP4 per full NVL72 rack
- Foundation of DGX SuperPOD — the world's most advanced AI factory
Get Pricing — Speak to Our Team
Request a Quote
Competitive pricing · Response within 4 hours
Key Features
Everything you need in one device
⚡
30× Real-Time LLM Inference
Delivers sub-50ms token latency for trillion-parameter language models — a performance level impossible for any prior infrastructure to achieve in real time.
🌊
Full Liquid Cooling
Direct liquid cooling increases compute density, eliminates air constraints, and achieves 25× better performance per watt vs H100 air-cooled systems.
🧬
Grace CPU + Blackwell GPU Unity
NVLink-C2C at 900 GB/s creates coherent CPU-GPU memory with 17 TB unified address space — eliminating PCIe bottleneck entirely for data-intensive AI workloads.
📊
720 PFLOPS FP8 per Rack
720 PFLOPS sparse FP8 training per NVL72 rack — equivalent to over 100 DGX H100 nodes concentrated in a single, manageable rack-scale system.
🌐
Quantum-X800 InfiniBand Scale-Out
Scale multiple NVL72 racks via NVIDIA Quantum-X800 InfiniBand or Spectrum-X800 Ethernet for petaflop AI factory clusters managed by Mission Control.
🏭
DGX SuperPOD Foundation
GB200 NVL72 is the core unit of DGX SuperPOD — the world's most advanced AI factory, used by leading research organisations, pharma firms, and global enterprises.
🔬
18× Database Acceleration
NVLink-C2C and Blackwell's decompression engines accelerate critical database queries 18× vs CPU — combining AI inference and enterprise analytics in one system.
About the NVIDIA GB200 NVL72
NVIDIA GB200 NVL72 is not an incremental GPU server improvement — it is a fundamental reimagining of enterprise AI infrastructure. Connecting 72 Blackwell GPUs and 36 Grace CPUs in a single fully liquid-cooled rack with 130 TB/s NVLink bandwidth makes all 72 GPUs operate as a single massive GPU — eliminating the communication bottlenecks that fragment performance in traditional multi-node clusters.
The performance implications are extraordinary: 30× faster real-time LLM inference compared to H100 infrastructure — enabling trillion-parameter model responses at sub-50ms latency. 4× faster training throughput at scale. And 25× less energy per unit of AI output versus H100 air-cooled systems. This is what makes GB200 NVL72 the foundation of the world's most advanced AI factories.
Grace CPUs and Blackwell GPUs connect via NVLink-C2C creating a coherent 17 TB unified memory space — eliminating the PCIe bottleneck entirely. GB200 NVL72 is deployed as the core building block of DGX SuperPOD, powering AI factories at leading research institutions, pharmaceutical companies, automotive manufacturers, and financial services firms globally — at 8 of the top 10 global telcos, 7 of the top 10 pharmas, and all 10 top universities.
Technical Specifications
📄 Download Datasheet (PDF)Use Cases
Ideal for a wide range of deployments
🧠
Trillion-Parameter LLM Inference
30× faster real-time inference enables trillion-parameter models to serve production traffic at sub-50ms latency — the infrastructure behind next-generation AI products.
🏋️
Foundation Model Training
Train 1.8T parameter MoE models 4× faster than H100 clusters with dramatically lower energy cost — the platform behind the world's most advanced AI labs.
🔬
Scientific Discovery at Scale
Climate modelling, drug discovery, particle physics — GB200 NVL72 provides exascale computing density for the world's most demanding science.
🏭
AI Factory Core Infrastructure
Deploy one NVL72 rack as an AI Centre of Excellence or scale to dozens of racks for national-scale AI infrastructure via DGX SuperPOD.
FAQ
Frequently Asked Questions
Why Servnet
Why buy from Servnet?
Trusted UK IT reseller since 2003 — supplying businesses of all sizes with genuine, competitively priced technology.
🏆
Authorised UK Reseller
Servnet is an authorised IT reseller with direct access to leading product lines at competitive pricing.
💬
Expert Pre-Sales Advice
Our engineers will help you select the right product and advise on compatibility with your existing infrastructure.
🚚
Fast UK Delivery
Most products ship from UK stock. Express delivery available for urgent deployments.
🔒
Genuine, Warranted Kit
All products are genuine, brand-new, with full manufacturer warranty and RMA support.
📋
Configuration Services
We can pre-configure your hardware before shipping — reducing on-site deployment time and cost.
📞
24/7 Support
Post-sale support from our certified engineers. We're here if anything goes wrong.
