Artemis III Rack

OVERVIEW

Rack solutions for next-generation AI factories.

 

Artemis III, Inventec’s rack solution featuring NVIDIA Vera Rubin NVL72 and NVIDIA Spectrum-X Ethernet networking, is designed to address demanding AI training, inference, and agentic reasoning workloads for next-generation AI factories.

Alongside Artemis III, Inventec will present an Inventec-defined rack solution, demonstrating its ability to design and integrate rack-level infrastructure based on customer requirements. By drawing from Inventec’s broad server portfolio, customers can select the computing platform that best fits their workload and deployment needs and scale that platform into a complete rack solution, giving organizations greater choice in how they build and expand their computing infrastructure.

NVIDIA Vera Rubin NVL72NVIDIA Spectrum-X EthernetInventec-defined rack100% liquid cooling solution

228KW

TDP for NVIDIA Vera Rubin NVL72

5000A+

Power Busbar Current Rating

2x 110 kW

3RU 110 kW Power-Shelf, 6x 18.3kW PSU each

48RU

NVIDIA® MGX™ rack

KEY FEATURES

Built for next-generation AI training

Four design pillars give Artemis III the power density, cooling and serviceability needed for NVIDIA Vera Rubin NVL72 deployments.

01

1U

NVIDIA® MGX™ 1U architecture

NVIDIA® MGX™ Design

  • •Comply with NVIDIA® MGX™ 1U architecture

02

228KW

TDP for NVIDIA Vera Rubin NVL72 building blocks

High-Power Density

  • •228KW TDP for NVIDIA Vera Rubin NVL72 building blocks
  • •Power Busbar – 5000A+ Current Rating
  • •3RU 19" 110 kW Power-Shelf with 6x 18.3kW PSU and additional bulk capacitor

03

100%

Liquid cooling Compute and Switch tray solution

Liquid Cooling

  • •Support different liquid cooling solution: sidecar/row based
  • •Leakage detection solution
  • •100% liquid cooling Compute and Switch tray solution

04

48RU

NVIDIA® MGX™ rack

MGX Rack Integration

  • •48RU NVIDIA® MGX™ rack
  • •1GbE OOB MGMT switch
  • •Great serviceability with blind mate design in NVIDIA NVLink™ cable cartridge and manifold

RACK CONFIGURATION

Artemis III Rack Overview

Reference Configuration: 18x 1U NVIDIA Compute Tray, 9x 1U Non-scalable NVIDIA NVLink™ Switch tray.

FRONT VIEW

 
 
 
 
2x1GbE MGMT Switch
 
 
 
1x3RU 110KW Power Shelf
 
 
 
10x1U NVIDIA Compute Tray
 
 
 
9x1U Non-scalable NVIDIA NVLink™ Switch tray
 
 
 
8x1U NVIDIA Compute Tray
 
 
 
1x3RU 110KW Power Shelf

BACK VIEW

 
 
 
 
Full rack blind mate manifold
 
 
 
100 A single AC input is required per power shelf
 
 
 
NVIDIA NVLink™ Cable Cartridge

NETWORKING & STORAGE

High-bandwidth fabric for every tray

NVIDIA ConnectX®-9 SuperNIC™ and NVIDIA BlueField®-4 DPU for back-end and front-end networking, with E1.s NVMe local storage.

BACK-END NETWORK

1.6T NVIDIA InfiniBand

NVIDIA ConnectX®-9 SuperNIC™ on each carrier board for back-end network.

FRONT-END NETWORK

800Gb/s

NVIDIA BlueField®-4 DPU 800Gb/s for front-end.

LOCAL STORAGE

4x E1.s NVMe

Support 4 pcs E1.s NVMe as local storage.

DIMENSIONS

Standard footprint, rack-scale power

Rolling height is 2296 mm. The depth includes extended frame.

HEIGHT

2236mm

WIDTH

600mm

DEPTH

1200mm

Ready to deploy NVIDIA Vera Rubin NVL72 at scale?

Our rack-scale engineering team will help you plan power, cooling and integration.

Contact Us
NEW

Artemis III Rack

A magnificently flexible architecture adapts to the extensibility from the all-in-one to the cloud scale to empower the wide spectrum of AI training applications
• Comply with NVIDIA® MGX™ 1U architecture
• 228KW TDP for NVIDIA Vera Rubin NVL72 building blocks
• Power Busbar – 5000A+Current Rating
• 3RU 19" 110 kW Power-Shelf with 6x 18.3kW PSU and additional bulk capacitor
• Support different liquid cooling solution: sidecar/row based
• Leakage detection solution
• 100% liquid cooling Compute and Switch tray solution
• 48RU NVIDIA® MGX™ rack
• 1GbE OOB MGMT switch
• Great serviceability with blind mate design in NVIDIA NVLink™ cable cartridge and manifold