ASUS ExpertCenter Pro ET900N G3 – NVIDIA GB300 Deskside AI Supercomputer
The ASUS ExpertCenter Pro ET900N G3 is a deskside AI supercomputer built on the NVIDIA DGX Station architecture. Powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, it delivers up to 20 petaFLOPS of AI performance and 748GB of coherent memory in a tower that sits beside your desk. Your team can train, fine-tune and run very large language models and autonomous AI agents locally, keep sensitive data on-premises, and cut recurring cloud GPU and API costs.
Buy the ASUS ET900N G3 in India from ITGadgetsOnline, with expert pre-sales guidance on configuration, networking and deployment. See our full range of ASUS Agent Computers.
ASUS ET900N G3 at a Glance
- Superchip: NVIDIA GB300 Grace Blackwell Ultra – 72-core Grace CPU + Blackwell Ultra GPU
- AI performance: up to 20 PFLOPS (FP4)
- Memory: 748GB coherent memory (496GB LPDDR5X + 252GB HBM3e)
- Networking: NVIDIA ConnectX-8 SuperNIC, dual 400G QSFP ports
- Storage: 2 x 2TB NVMe SSD pre-installed (RAID 1), 4 x M.2 slots including PCIe 6.0
- Software: Ubuntu with NVIDIA AI Developer Tools, NVIDIA NemoClaw, ASUS Zenni AI Hub
- Power: 1600W 80 Plus Titanium PSU
Extreme AI Performance with NVIDIA GB300 Grace Blackwell Ultra
At the heart of the ET900N G3 is the NVIDIA Blackwell Ultra GPU connected to a 72-core NVIDIA Grace CPU through NVIDIA NVLink-C2C. This chip-to-chip interconnect gives the CPU and GPU high-bandwidth, coherent access to a single memory pool, removing the bottleneck of traditional PCIe-connected workstations. Blackwell-generation Tensor Cores add support for 4-bit floating point (NVFP4), which lets larger models fit in memory and run faster while keeping accuracy high.
748GB Coherent Memory for Trillion-Parameter Models
Modern AI models keep growing. With 496GB of LPDDR5X CPU memory and 252GB of HBM3e GPU memory (7.1 TB/s) working as one coherent pool, the ET900N G3 can hold and run massive models that would normally need a data-centre server – right at your desk. That makes it ideal for fine-tuning large LLMs, running long-context inference and serving several models at once.
Built for Agentic AI
The ET900N G3 is designed to run AI agents that plan and carry out multi-step tasks on your behalf. It supports NVIDIA NemoClaw for secure, streamlined agent workflows and NVIDIA AI Enterprise for production deployment. Because everything runs locally, your prompts, documents and company data never leave your premises.
ASUS Zenni AI Hub – Simplify Local AI
ASUS Zenni AI Hub gives your team one platform to deploy and manage local AI:
- Deploy and manage models – deploy and route LLMs, monitor system resources, manage NVIDIA MIG partitions and benchmark performance, then connect models to your applications through APIs.
- Scale across nodes – link multiple ET900N G3 systems over the LAN into one cluster and manage GPU resources centrally.
- Multi-agent collaboration – access OpenClaw and Hermes, add your own agents and coordinate workflows from one portal.
- Knowledge+ workflows – build a private knowledge base for document Q&A, comparison and summarisation, with web search and workflow automation.
Full-Stack NVIDIA AI Software
The system ships with Ubuntu and NVIDIA AI Developer Tools, pre-configured with CUDA-X libraries, NVIDIA NIM microservices and Blueprints. It supports popular tools such as PyTorch, Jupyter and Ollama for prototyping and inference. Because the same NVIDIA stack runs from desktop to cloud, you can develop locally and deploy to the data centre without changing your code.
Multi-Instance GPU (MIG) for Teams
NVIDIA MIG can split the GPU into up to seven isolated instances, so several developers or workloads can share one ET900N G3 securely and efficiently.
NVIDIA ConnectX-8 SuperNIC Networking
Dual 400G QSFP ports powered by the NVIDIA ConnectX-8 SuperNIC (up to 800Gb/s) provide extremely fast connectivity for moving datasets and scaling out. Two DGX Station-class systems can be linked to expand model capacity and performance even further.
Add an NVIDIA RTX PRO GPU
Alongside the GB300 Superchip, the ET900N G3 can take one optional NVIDIA RTX PRO Blackwell graphics card (RTX PRO 6000 Blackwell Max-Q, RTX PRO 4000 Blackwell SFF or RTX PRO 2000 Blackwell), combining data-centre-grade AI compute with ray-traced visualisation and simulation.
Data-Centre Cooling, Office-Friendly Acoustics
A closed-loop AIO liquid cooling system keeps performance sustained under heavy AI workloads. The ET900N G3 is designed to operate from 10°C to 35°C ambient and keeps noise to around 50 dB in Auto mode at full load (without a discrete graphics card) – comparable to a quiet office. AI-based power-shifting moves power to wherever the current workload needs it.
Enterprise Security and Management
An ASPEED AST2600 BMC with AMI MegaRAC firmware supports IPMI 2.0 and the DMTF Redfish API for remote monitoring and management over a dedicated 1G LAN, and on-board TPM 2.0 protects keys and credentials.
Who Is the ASUS ET900N G3 For?
- AI research labs and universities – train and fine-tune large models without waiting for shared cluster time.
- Enterprises – run private LLMs, RAG and AI agents on confidential data on-premises.
- Healthcare and medical research – process sensitive data locally for imaging, genomics and clinical AI.
- Automotive, aerospace and smart infrastructure – develop simulation, vision and edge AI models.
- AI start-ups and developers – prototype on the same NVIDIA stack used in the data centre.
Ports and Connectivity
ASUS ExpertCenter Pro ET900N G3 Specifications
| Model | ASUS ExpertCenter Pro ET900N G3 |
| Architecture | NVIDIA DGX Station |
| Operating System | Ubuntu with NVIDIA AI Developer Tools |
| Superchip | NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip |
| CPU | NVIDIA Grace CPU, 72 Arm Neoverse V2 cores |
| GPU | NVIDIA Blackwell Ultra GPU |
| AI Performance | Up to 20 PFLOPS |
| Memory | 748GB coherent memory CPU: 496GB LPDDR5X (396 GB/s) GPU: 252GB HBM3e (7.1 TB/s) |
| Expansion Slots | 3 x PCIe x16 (Gen 5.0 x16, Gen 5.0 x8, Gen 5.0 x8) |
| Storage | 2 x M.2 2280 PCIe 5.0 x4 (OS, RAID 1) 2 x M.2 2280 PCIe 6.0 x4 (training data) 2 x 2TB NVMe M.2 SSD pre-installed |
| Networking | 2 x NVIDIA ConnectX-8 SuperNIC QSFP112 400G (transceiver required) 1 x Marvell 10G Ethernet 1 x Realtek 1G Ethernet (BMC) 1 x M.2 Key E slot |
| Optional Graphics | 1 x NVIDIA RTX PRO 6000 Blackwell Max-Q, RTX PRO 4000 Blackwell SFF or RTX PRO 2000 Blackwell |
| Front I/O | 2 x USB 10Gbps Type-C, 2 x USB 10Gbps Type-A, 1 x USB 2.0, headphone, microphone |
| Rear I/O | 4 x USB 10Gbps Type-A, 2 x 400G QSFP, 1 x 10G LAN, 1 x 1G LAN (BMC), 1 x Micro USB COM (BMC), 1 x mini DisplayPort (BMC), 3 x audio jacks |
| Management | ASPEED AST2600, AMI MegaRAC, IPMI 2.0, DMTF Redfish API |
| Security | On-board TPM 2.0 |
| Cooling | Closed-loop AIO liquid cooling |
| Power Supply | 1600W 80 Plus Titanium ATX, 115–240V AC, IEC C19 cord |
| Dimensions | 584 x 232 x 565 mm (23.0″ x 9.1″ x 22.3″) |
| Weight | 27 kg net / 32 kg gross |
| Operating Temperature | 10°C – 35°C |
| Optional Items | QSFP 400G transceivers, Wi-Fi module and antenna |
Specifications are based on ASUS information and may vary by configuration and region.
Frequently Asked Questions
What is the ASUS ExpertCenter Pro ET900N G3?
It is a deskside AI supercomputer from ASUS based on the NVIDIA DGX Station architecture. It uses the NVIDIA GB300 Grace Blackwell Ultra Superchip with 748GB of coherent memory and up to 20 petaFLOPS of AI performance.
How much memory does the ET900N G3 have?
748GB of coherent memory: 496GB LPDDR5X attached to the Grace CPU and 252GB HBM3e on the Blackwell Ultra GPU, shared through NVIDIA NVLink-C2C.
Can the ASUS ET900N G3 run large language models locally?
Yes. Its large coherent memory and FP4 support allow very large LLMs to be fine-tuned and run on-premises, and it supports NVIDIA NemoClaw, NVIDIA AI Enterprise, PyTorch and Ollama for local AI agents and inference.
What operating system does it come with?
Ubuntu with NVIDIA AI Developer Tools, pre-configured with CUDA-X libraries for AI development, data science and inference.
Is the ASUS ET900N G3 available in India?
Yes. You can order the ASUS ExpertCenter Pro ET900N G3 from ITGadgetsOnline. Contact us for configuration advice, optional RTX PRO graphics, QSFP transceivers and delivery details.
What is the difference between the ASUS ET900N G3 and the Ascent GX10?
The ASUS Ascent GX10 is a compact AI mini PC with the NVIDIA GB10 Superchip, 128GB memory and 1 petaFLOP of AI performance. The ET900N G3 is a full tower with the GB300 Superchip, 748GB memory and up to 20 petaFLOPS, built for much larger models and team workloads.
Reviews
There are no reviews yet.