Artemis II Rack

OVERVIEW

Redefining what is possible in AI performance.

 

The Inventec Artemis II rack redefines what is possible in AI performance, merging extreme compute power with fluid scalability. Featuring the NVIDIA GB300 NVL72 system and NVIDIA® MGX™ platform, it integrates 36 NVIDIA Grace™ CPUs and 72 Blackwell GPUs into a single, massive-scale compute engine through NVIDIA NVLink™ technology, delivering the speed and capacity to handle tomorrow’s most demanding AI challenges.

To further optimize data movement in large-scale deployments, the system incorporates two I/O boards, each carrying two NVIDIA ConnectX®-8 SuperNICs, enabling high-bandwidth, low-latency back-end networking that keeps pace with the GPUs’ performance. A fully integrated liquid cooling solution spans the entire rack, maintaining thermal stability and peak efficiency even under continuous, high-intensity AI training and inference workloads. With this architecture, organizations gain the confidence, performance, and scalability to push the boundaries of AI innovation.

36 Grace™ CPUs72 Blackwell GPUs2 I/O boards · 4× ConnectX®-8Rack-wide liquid cooling

136kW

Max rack power for GB300 NVL72

72

Blackwell GPUs into a single, massive-scale compute engine through NVIDIA NVLink™ technology

48RU

NVIDIA MGX™ rack

6 × 33kW

Power shelves, N+N feed redundancy

KEY FEATURES

Built for the most demanding AI workloads

Four design pillars give Artemis II the density, cooling and serviceability needed for NVIDIA GB300 NVL72 deployments.

01

NVIDIA MGX™ Design

  • Complies with NVIDIA® MGX™ 1U architecture

02

High-Power Density

  • Up to 136kW for NVIDIA GB300 NVL72 building blocks
  • Power busbar rated 1400A
  • 6× 33kW power shelves for N+N feed redundancy

03

Liquid Cooling

  • Rack-mounted, sidecar or row-based cooling
  • Inventec self-developed rack-mounted CDU, tailored to customer requirements
  • Leakage detection solution
  • 100% or hybrid liquid-cooled compute and switch trays

04

MGX Rack Integration

  • 48RU NVIDIA® MGX™ rack
  • 1GbE OOB management switch
  • Blind-mate design for NVLink™ cable cartridge and manifold for easy service

RACK CONFIGURATION

Artemis II Rack Overview

18 compute trays and 9 NVLink™ switch trays form a complete GB300 NVL72 building block, with power shelves at both ends of the rack.

FRONT VIEW

Artemis II Rack front view
 
 
 
3×1G GbE Management Switch
 
 
 
3×1RU 33kW Power Shelf
 
 
 
10×1U NVIDIA Compute Tray
 
 
 
9×1U NVIDIA NVLink™ Switch Tray
 
 
 
8×1U NVIDIA Compute Tray
 
 
 
3×1RU 33kW Power Shelf

BACK VIEW

Artemis II Rack back view
 
 
 
Power Shelf · 60A single AC input each
 
 
 
Blind-Mate Manifold
 
 
 
NVIDIA NVLink™ Cable Cartridge

DIVERSE LIQUID COOLING SOLUTIONS

Choose the cooling that fits your data center

Every option includes leakage detection, and compute and switch trays can be 100% or hybrid liquid-cooled.

L2A · Liquid-to-Air

Sidecar

A sidecar heat exchanger next to the rack returns heat to room air — no facility water loop required.

L2L · Liquid-to-Liquid

In-Row CDU

A row-based CDU connects to facility water and serves multiple racks for maximum heat rejection.

4U · Optional

Rack-Mounted CDU

Inventec self-developed 4U CDU integrated inside the rack, configurable to customer requirements.

DIMENSIONS

Standard footprint, rack-scale power

Rolling height 2230 mm. Fits standard 600 mm data center rows.

HEIGHT

2236mm

WIDTH

600mm

DEPTH

1068mm

Ready to deploy NVIDIA GB300 NVL72 at scale?

Contact Us
NEW

Artemis II Rack

The intelligent foundation for limitless AI scalability

• Comply with NVIDIA® MGX™ 1U  architecture 
• High-power density design for NVIDIA® 
• Max 136kW for NVIDIA GB300 NVL72 building blocks
• Support different liquid cooling solution: sidecar/row based
• Support Leakage detection solution
• Great serviceability with blind mate design in NVIDIA® NVLink™ cable cartridge and manifold