ASUS ExpertCenter Pro ET900N G3 GB300 AI Supercomputer

ASUS ExpertCenter Pro ET900N G3 GB300 AI Supercomputer with NVIDIA Grace Blackwell Ultra, 748GB Memory & Dual 400G Networking - Local Warranty 

Enterprise AI Supercomputer Powered by NVIDIA Grace Blackwell Ultra

The ASUS ExpertCenter Pro ET900N G3 is ASUS's flagship AI supercomputer engineered to accelerate next-generation artificial intelligence, machine learning, and high-performance computing workloads. Powered by the NVIDIA Grace Blackwell Ultra (GB300) Superchip, it delivers exceptional AI performance with up to 20 PFLOPS of AI compute, enabling organisations to train, fine-tune, and deploy large language models (LLMs), generative AI applications, AI agents, computer vision, and scientific simulations directly within their own infrastructure. It is purpose-built for enterprises, research institutions, government agencies, and AI innovators seeking uncompromising performance with complete data sovereignty.

Massive 748GB Coherent Memory for Large AI Models

Designed for today's most demanding AI workloads, the ExpertCenter Pro ET900N G3 features an incredible 748GB of coherent unified memory, allowing significantly larger AI models and datasets to be processed without the memory bottlenecks associated with conventional GPU systems. This unified memory architecture dramatically improves performance for inference, retrieval-augmented generation (RAG), AI agent frameworks, digital twins, engineering simulations, and data-intensive research applications.

Ultra-Fast Dual 400G Networking for AI Clusters

The system is equipped with dual NVIDIA ConnectX®-8 400GbE QSFP networking, enabling ultra-low latency communication between multiple AI nodes for distributed training and large-scale inference. Whether deployed as a standalone AI workstation or as part of an enterprise AI cluster, the ET900N G3 provides the networking bandwidth required for high-performance computing environments, research laboratories, universities, and private AI clouds.

Enterprise-Class Reliability and Future-Ready AI Infrastructure

Built for continuous enterprise operation, the ASUS ExpertCenter Pro ET900N G3 includes integrated Baseboard Management Controller (BMC) for secure remote monitoring and administration, support for NVIDIA Multi-Instance GPU (MIG) technology for multi-tenant AI environments, and enterprise-grade reliability for mission-critical deployments. It provides organisations with a scalable AI platform capable of supporting generative AI, foundation models, cybersecurity, healthcare, financial modelling, engineering, autonomous systems, and advanced scientific research while keeping sensitive data securely on-premises.

Technical Specifications

General

Model: ExpertCenter Pro ET900N G3
Product Name: ASUS ExpertCenter Pro ET900N G3 GB300 AI Supercomputer with NVIDIA Grace Blackwell Ultra, 748GB Memory & Dual 400G Networking
Manufacturer: ASUS
Product Series: ASUS ExpertCenter Pro
Product Type: Enterprise AI Supercomputer / AI Server
Compute Platform: NVIDIA Grace Blackwell Ultra
Superchip Platform: NVIDIA GB300
Processor Architecture: NVIDIA Grace CPU + NVIDIA Blackwell Ultra GPU
System Architecture: Unified CPU-GPU Accelerated Computing Platform
Form Factor: Enterprise Rackmount AI System
Target Deployment: Enterprise Datacentres, AI Factories, Research Laboratories, Universities, Government Agencies and Cloud Infrastructure
Primary Workloads: AI Training, AI Fine-Tuning, AI Inferencing, AI Agents, Generative AI, Machine Learning, Deep Learning, Scientific Computing and High-Performance Computing
Operating System: Enterprise Linux Distribution, Configuration Dependent
Remote Management: Integrated Baseboard Management Controller
Multi-Tenant Support: NVIDIA Multi-Instance GPU
Warranty: ASUS Enterprise Warranty, Configuration Dependent

Processor Specifications

Processor Platform: NVIDIA Grace Blackwell Ultra GB300 Superchip
CPU: NVIDIA Grace CPU
CPU Architecture: Arm
CPU Core Architecture: Arm Neoverse V2
CPU Core Count: 72 Cores
CPU-GPU Architecture: Tightly Coupled NVIDIA Grace and Blackwell Ultra Architecture
CPU-GPU Interconnect: NVIDIA NVLink-C2C
Unified CPU-GPU Address Space: Supported
Coherent CPU-GPU Memory Access: Supported
High-Speed CPU-GPU Data Transfer: Supported
Large-Scale Parallel Processing: Supported
Enterprise AI Computing: Supported
High-Performance Computing: Supported
Scientific Computing: Supported
Continuous AI Workload Operation: Supported

AI Performance

Maximum AI Performance: Up to 20 PFLOPS
AI Accelerator Architecture: NVIDIA Blackwell Ultra
Transformer Engine: NVIDIA Blackwell Transformer Engine
Tensor Core Acceleration: Supported
Mixed-Precision AI Computing: Supported
FP4 AI Computing: Supported
FP8 AI Computing: Supported
FP16 AI Computing: Supported
BF16 AI Computing: Supported
INT8 AI Inferencing: Supported
Dynamic Precision Management: Supported
Sparsity Acceleration: Supported
Large Language Model Acceleration: Supported
Foundation Model Training: Supported
Foundation Model Fine-Tuning: Supported
Generative AI Acceleration: Supported
AI Agent Development: Supported
Reasoning Model Processing: Supported
Multimodal AI Processing: Supported
Retrieval-Augmented Generation: Supported
Enterprise AI Inferencing: Supported
Distributed AI Training: Supported
Scientific AI Processing: Supported

AI Features

Large Language Model Training
Large Language Model Fine-Tuning
Foundation Model Development
Generative AI Applications
Agentic AI Workloads
AI Agent Development and Deployment
Retrieval-Augmented Generation
Multimodal AI Processing
Natural Language Processing
Computer Vision
Speech Recognition
Speech Generation
Recommendation Systems
Predictive Analytics
Digital Twin Simulation
Autonomous Systems Development
Robotics AI
Scientific Machine Learning
Drug Discovery
Molecular Modelling
Healthcare AI
Financial Modelling
Fraud Detection
Cybersecurity AI
Manufacturing AI
Enterprise Knowledge Assistants
Private AI Deployment
Cloud-Native AI Deployment
AI Factory Infrastructure
Production AI Inferencing
Multi-User AI Development

Memory Specifications

Total System Memory: 748GB
Memory Architecture: NVIDIA Unified Coherent Memory
CPU-GPU Shared Memory: Supported
Single Shared Memory Pool: Supported
Unified Address Space: Supported
Memory Coherency: Supported
Direct CPU and GPU Memory Access: Supported
High-Bandwidth Memory Architecture: Supported
Low-Latency Data Access: Supported
Reduced CPU-GPU Data Movement: Supported
Large AI Model Loading: Supported
Large Context Window Processing: Supported
Multi-Billion Parameter Model Support: Supported
Large Dataset Processing: Supported
AI Model Fine-Tuning Support: Supported
Enterprise Inferencing Support: Supported
Retrieval-Augmented Generation Support: Supported
Vector Database Processing: Supported
Memory Virtualisation: Supported
Multi-Tenant Memory Allocation: Supported
Continuous AI Pipeline Processing: Supported

Storage Specifications

Storage Technology: NVMe Solid-State Drive
Storage Interface: PCI Express 5.0
Drive Form Factor: M.2 2280
Storage Slot Capacity: Up to 4 × M.2 2280 NVMe PCIe 5.0 SSD Slots
Maximum Internal Storage Capacity: Up to 8TB Total Capacity When Fully Populated
Maximum Capacity per Drive: Configuration Dependent
Boot Drive Support: Supported
Enterprise NVMe SSD Support: Supported
PCIe Gen5 Storage Performance: Supported
High-Speed AI Dataset Loading: Supported
Large AI Model Repository Support: Supported
AI Model Checkpoint Storage: Supported
Vector Database Storage: Supported
Training Dataset Storage: Supported
High-Speed Data Preprocessing: Supported
Continuous Read and Write Workloads: Supported
Storage Expansion: Supported
RAID Support: Configuration Dependent
Hot-Swap Support: Configuration Dependent
Self-Encrypting Drive Support: Configuration Dependent
Storage Health Monitoring: Supported
Remote Storage Status Monitoring: Supported

Graphics Specifications

GPU Platform: NVIDIA Blackwell Ultra
GPU Architecture: NVIDIA Blackwell Ultra
GPU Type: Enterprise AI and HPC Accelerator
Tensor Cores: NVIDIA Tensor Cores
Transformer Engine: Supported
CUDA Parallel Computing: Supported
CUDA Toolkit Compatibility: Supported
NVIDIA cuDNN Compatibility: Supported
NVIDIA TensorRT Compatibility: Supported
NVIDIA RAPIDS Compatibility: Supported
NVIDIA NeMo Compatibility: Supported
NVIDIA NGC Container Compatibility: Supported
NVIDIA AI Enterprise Compatibility: Supported
GPUDirect RDMA: Supported
GPUDirect Storage: Supported
Multi-Instance GPU: Supported
GPU Resource Partitioning: Supported
Secure GPU Workload Isolation: Supported
Multi-Tenant GPU Allocation: Supported
Enterprise GPU Virtualisation: Supported
Mixed-Precision Computing: Supported
Large Language Model Acceleration: Supported
Scientific Simulation Acceleration: Supported
High-Performance Data Analytics: Supported

Networking Specifications

Integrated Network Adapter: NVIDIA ConnectX-8
Network Interface Type: High-Speed Enterprise AI Networking
Network Ports: 2 × 400GbE QSFP Ports
Connector Type: QSFP
Maximum Per-Port Bandwidth: Up to 400Gbps
Maximum Aggregate Network Bandwidth: Up to 800Gbps
Ethernet Support: Supported
InfiniBand Support: Platform and Configuration Dependent
RDMA Support: Supported
RoCE Support: Supported
RoCE v2 Support: Supported
GPUDirect RDMA: Supported
GPUDirect Storage: Supported
Ultra-Low-Latency Networking: Supported
Distributed AI Training: Supported
AI Cluster Interconnect: Supported
High-Speed Storage Fabric: Supported
Scale-Out AI Infrastructure: Supported
Enterprise AI Fabric Integration: Supported
Multi-Node Model Training: Supported
High-Performance Computing Cluster Support: Supported

Ports & Expansion

2 × NVIDIA ConnectX-8 400GbE QSFP Network Ports
Up to 4 × M.2 2280 NVMe PCIe 5.0 SSD Slots
PCI Express 5.0 Expansion Support
Enterprise Network Expansion Support
High-Speed Storage Expansion Support
AI Accelerator Expansion: Configuration Dependent
Management Network Port: Configuration Dependent
USB Ports: Configuration Dependent
Video Output: Configuration Dependent
Serial Management Port: Configuration Dependent
Dedicated BMC Management Interface: Supported
Rack Infrastructure Integration: Supported
Future Enterprise Expansion: Supported

Multi-Tenant & Virtualisation Features

NVIDIA Multi-Instance GPU Support
GPU Resource Partitioning
Secure Workload Isolation
Multiple Independent GPU Instances
Concurrent AI Workload Processing
Multi-User AI Development
Department-Level Resource Allocation
Containerised AI Workloads
Virtual Machine Support
Enterprise GPU Virtualisation
Cloud-Native AI Deployment
Kubernetes Resource Scheduling
Docker Container Support
Red Hat OpenShift Support
Workload-Orchestrated GPU Allocation
Shared AI Infrastructure Deployment
Improved GPU Utilisation
Multi-Tenant Enterprise AI Services

Security Features

Secure Boot
Firmware Protection
Hardware Root of Trust
Trusted Platform Module Support: Configuration Dependent
Role-Based Access Control
Secure Remote Management
Encrypted Management Sessions
BMC User Authentication
Management Account Separation
Secure Firmware Updating
System Event Logging
Hardware Health Monitoring
Audit Logging
Secure Workload Isolation
Multi-Tenant GPU Isolation
Enterprise Authentication Integration
Self-Encrypting SSD Support: Configuration Dependent
Datacentre Security Integration
Remote Access Control
Administrative Privilege Management

Management Features

Integrated Baseboard Management Controller
Out-of-Band Management
Dedicated Remote Management
Remote KVM Access
Remote Console Access
Remote Power On
Remote Power Off
Remote Power Cycling
Remote System Reset
Remote BIOS Configuration
Remote Firmware Updates
Remote Hardware Diagnostics
System Health Monitoring
Processor Status Monitoring
GPU Status Monitoring
Memory Status Monitoring
Storage Health Monitoring
Network Status Monitoring
Temperature Monitoring
Fan Speed Monitoring
Voltage Monitoring
Power Consumption Monitoring
System Event Logging
Hardware Inventory Management
Asset Management
Enterprise Fleet Management
Alert and Notification Support
24×7 Remote Administration
Datacentre Management Integration

Software & AI Framework Support

NVIDIA AI Enterprise
NVIDIA CUDA Toolkit
NVIDIA cuDNN
NVIDIA TensorRT
NVIDIA Triton Inference Server
NVIDIA NeMo
NVIDIA RAPIDS
NVIDIA NGC Containers
PyTorch
TensorFlow
JAX
ONNX Runtime
Hugging Face Transformers
DeepSpeed
Megatron-LM
Kubernetes
Docker
Red Hat OpenShift
Slurm Workload Manager
NVIDIA Base Command Manager
MLflow
Ray
Apache Spark
Vector Database Platforms
Enterprise MLOps Platforms
AI Model Serving Frameworks
Distributed Training Frameworks

Operating System Compatibility

Ubuntu Server
Red Hat Enterprise Linux
Rocky Linux
SUSE Linux Enterprise Server
NVIDIA-Certified Linux Environments
Enterprise Linux Distributions
Container-Optimised Operating Environments
Operating System Support: Configuration and Certification Dependent

Cooling & Reliability

Enterprise Air-Cooled Architecture
Optimised Datacentre Airflow
High-Efficiency Thermal Management
Continuous AI Workload Cooling
CPU Thermal Monitoring
GPU Thermal Monitoring
Memory Thermal Monitoring
Storage Thermal Monitoring
Variable-Speed Fan Control
System Fan Redundancy: Configuration Dependent
Thermal Protection
Automated Fan Management
Hardware Health Alerts
High-Availability Design
Enterprise-Grade Components
Continuous 24×7 Operation
Long-Duration AI Training Support
Datacentre Rack Deployment Ready

Power Specifications

Power Supply Type: Enterprise High-Efficiency Power Supply
Power Supply Configuration: Configuration Dependent
Redundant Power Supply Support: Configuration Dependent
Hot-Plug Power Supply Support: Configuration Dependent
Power Input: Datacentre AC Power, Configuration Dependent
Power Consumption: Workload and Configuration Dependent
Remote Power Monitoring: Supported
Remote Power Control: Supported
Power Usage Reporting: Supported
Power Fault Monitoring: Supported
Continuous High-Performance Operation: Supported
Datacentre Power Optimisation: Supported
Energy-Efficient Workload Processing: Supported

Compatibility

Compatible AI Workloads:

Large Language Models
Foundation Models
Generative AI
AI Agents
Agentic AI
Retrieval-Augmented Generation
Reasoning Models
Multimodal AI
Computer Vision
Natural Language Processing
Speech AI
Recommendation Engines
Scientific Machine Learning
Predictive Analytics
Digital Twins
Autonomous Systems
Enterprise AI Inferencing

Compatible Development Frameworks:

PyTorch
TensorFlow
JAX
ONNX Runtime
Hugging Face Transformers
DeepSpeed
NVIDIA NeMo
NVIDIA TensorRT
NVIDIA Triton Inference Server
NVIDIA RAPIDS

Compatible Deployment Platforms:

Docker
Kubernetes
Red Hat OpenShift
Slurm
NVIDIA Base Command Manager
Enterprise MLOps Platforms
Private Cloud Infrastructure
Hybrid Cloud Infrastructure
AI Factory Infrastructure

Compatible Network Environments:

400GbE Ethernet Fabrics
NVIDIA ConnectX Networking
RDMA Networks
RoCE Networks
High-Performance Storage Networks
Distributed AI Clusters
High-Performance Computing Clusters

Design & Construction

Enterprise Rackmount Chassis
Datacentre-Optimised Mechanical Design
Professional AI Server Construction
High-Density Compute Architecture
Optimised Front-to-Rear Airflow
Rack Installation Support
Enterprise Cable Management
Serviceable Internal Components
Continuous 24×7 Operational Design
Remote Serviceability
High-Availability Architecture
Datacentre Integration Ready

Environmental Specifications

Operating Environment: Enterprise Datacentre
Operating Temperature: Configuration Dependent
Storage Temperature: Configuration Dependent
Operating Humidity: Configuration Dependent
Storage Humidity: Configuration Dependent
Maximum Operating Altitude: Configuration Dependent
Thermal Monitoring: Supported
Environmental Monitoring: Supported
Airflow Requirement: Datacentre Front-to-Rear Airflow
Continuous Operation: Supported
Compliance and Certifications: Region and Configuration Dependent

Physical Specifications

Product Type: Enterprise AI Supercomputer
Form Factor: Rackmount AI Server
Rack Unit Height: Configuration Dependent
Chassis Colour: Black
Dimensions: Configuration Dependent
System Weight: Configuration Dependent
Installed Weight: Configuration Dependent
Rack Mounting: Supported
Rail Kit: Configuration Dependent
Cable Management Arm: Configuration Dependent
Service Access: Enterprise Rack Service Design

Package Contents

ASUS ExpertCenter Pro ET900N G3 AI Supercomputer
NVIDIA Grace Blackwell Ultra GB300 Platform
Enterprise Power Cables
Rack Mounting Rail Kit: Configuration Dependent
Cable Management Accessories: Configuration Dependent
Quick Installation Guide
Safety Documentation
Warranty Documentation
Enterprise Support Information

Key Features

NVIDIA Grace Blackwell Ultra GB300 Superchip
72-Core NVIDIA Grace CPU
Arm Neoverse V2 CPU Architecture
NVIDIA Blackwell Ultra GPU
Up to 20 PFLOPS AI Performance
748GB Unified Coherent Memory
Shared CPU-GPU Memory Architecture
Up to 4 × M.2 2280 NVMe PCIe 5.0 SSD Slots
Up to 8TB Total Internal NVMe Storage
Dual NVIDIA ConnectX-8 400GbE QSFP Networking
Up to 800Gbps Aggregate Network Bandwidth
NVIDIA Multi-Instance GPU Support
Secure Multi-Tenant AI Workloads
Integrated Baseboard Management Controller
Out-of-Band Remote Management
NVIDIA AI Enterprise Ready
CUDA and TensorRT Optimised
Distributed AI Training Ready
Enterprise AI Factory Deployment
Large Language Model Training and Inferencing
High-Performance Computing Support
Enterprise Rackmount Architecture

Performance & Features

The ASUS ExpertCenter Pro ET900N G3 is an enterprise AI supercomputer powered by the NVIDIA Grace Blackwell Ultra GB300 platform. It combines a 72-core NVIDIA Grace CPU based on the Arm Neoverse V2 architecture with an NVIDIA Blackwell Ultra GPU to deliver up to 20 PFLOPS of AI performance for demanding model training, fine-tuning and production inferencing workloads.

The system features 748GB of coherent unified memory, allowing the Grace CPU and Blackwell Ultra GPU to access a shared memory pool through a unified address space. This architecture reduces unnecessary data transfers between separate CPU and GPU memory pools, improves processing efficiency and enables the system to handle larger language models, longer context windows and memory-intensive scientific applications.

Internal storage supports up to four M.2 2280 NVMe PCIe 5.0 SSD slots, with a maximum total capacity of up to 8TB when fully populated. The PCIe Gen5 storage architecture provides high-speed access to training datasets, model checkpoints, vector databases, AI application containers and production inference assets.

Dual NVIDIA ConnectX-8 400GbE QSFP network interfaces provide up to 800Gbps of aggregate network bandwidth. Support for RDMA, RoCE, GPUDirect RDMA and GPUDirect Storage enables low-latency communication between compute nodes, storage systems and GPU resources in distributed AI and high-performance computing environments.

NVIDIA Multi-Instance GPU technology allows the GPU to be partitioned into isolated processing instances for different users, departments or AI services. This enables organisations to run multiple workloads securely on the same platform while improving overall hardware utilisation and maintaining predictable resource allocation.

The integrated Baseboard Management Controller provides out-of-band access for remote administration, system monitoring, firmware management, diagnostics and power control. Administrators can monitor processor, GPU, memory, storage, networking, thermal and power conditions without requiring direct operating system access.

Support for NVIDIA AI Enterprise, CUDA, TensorRT, Triton Inference Server, NeMo, RAPIDS, PyTorch, TensorFlow, JAX, Kubernetes and Docker provides a comprehensive environment for developing, training, optimising and deploying enterprise AI applications.

Typical Use Cases

Large Language Model Training
Large Language Model Fine-Tuning
Foundation Model Development
Generative AI Applications
AI Agent Development
Agentic AI Deployment
Retrieval-Augmented Generation
Reasoning Model Processing
Multimodal AI
Production AI Inferencing
Machine Learning
Deep Learning
Computer Vision
Natural Language Processing
Speech AI
Recommendation Systems
Enterprise Knowledge Assistants
Cybersecurity AI
Financial Services AI
Healthcare AI Research
Drug Discovery
Molecular Simulation
Scientific Computing
High-Performance Computing
Digital Twin Simulation
Manufacturing AI
Autonomous Systems Development
Robotics Research
University Research Clusters
Government AI Infrastructure
Private AI Datacentres
Hybrid Cloud AI Infrastructure
Cloud Service Provider Infrastructure
Multi-Tenant AI Platforms
Enterprise AI Factories