ASUS ExpertCenter Pro ET900N G3 NVIDIA DGX Station GB300 AI Supercomputer
PRODUCT DESCRIPTION
ASUS ExpertCenter Pro ET900N G3 NVIDIA DGX Station GB300 Grace Blackwell Ultra Deskside AI Supercomputer, 748GB Coherent Memory, 20 PFLOPS, Dual 400G ConnectX-8 (ET900N G3) - Local Warranty
ASUS ExpertCenter Pro ET900N G3 — NVIDIA DGX Station GB300 Deskside AI Supercomputer
The ASUS ExpertCenter Pro ET900N G3 is a deskside AI supercomputer built on NVIDIA DGX Station architecture around the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip — a 72-core NVIDIA Grace CPU (Arm Neoverse V2) paired over NVLink-C2C with an NVIDIA Blackwell Ultra GPU, sharing 748 GB of coherent memory. ASUS rates it at up to 20 petaFLOPS of AI compute, enough to develop, fine-tune and serve large language models of up to 1 trillion parameters locally.
For a Singapore AI team, research lab, university or enterprise innovation group, the ET900N G3 brings data-centre-class GB300 compute under your own roof and your own data governance — without the lead time, rack space or liquid-cooled rack infrastructure of a full DGX server deployment, and without the recurring cost of cloud GPU instances.
748 GB Coherent Memory for Trillion-Parameter Models
The Grace CPU side carries 496 GB of LPDDR5X at 396 GB/s and the Blackwell Ultra GPU side carries 252 GB of HBM3e at 7.1 TB/s — unified into a single 748 GB coherent memory pool over NVLink-C2C rather than two separate memory spaces the software has to shuffle data between. Large models that would not fit in GPU memory alone on a conventional RTX workstation load and run without manual offloading, and Multi-Instance GPU (MIG) can partition the Blackwell Ultra GPU into up to seven isolated instances so several users or workloads share one system.
Up to 20 PetaFLOPS of AI Performance, Ubuntu with NVIDIA AI Developer Tools
The system ships running Ubuntu with the NVIDIA AI Developer Tools stack preinstalled, targeting vLLM, NVIDIA NeMo, NVIDIA NIM and the wider CUDA ecosystem for model development, fine-tuning, RAG pipelines, agentic AI and inference. ASUS has stated that Windows-based AI development support is planned, but Ubuntu is the shipping configuration today.
Enterprise Storage, Dual 400G ConnectX-8 Networking and Out-of-Band Management
Two of the four M.2 2280 NVMe slots run PCIe 5.0 x4 off the Grace CPU and come pre-configured as an OS RAID 1 pair with two 2 TB NVMe SSDs installed; the remaining two slots hang off the ConnectX-8 for training-data storage, for up to 8 TB across all four slots. Three PCIe 5.0 expansion slots (one x16 and two x8) accept an optional NVIDIA RTX PRO Blackwell display GPU or additional NICs. Networking pairs two NVIDIA ConnectX-8 SuperNIC QSFP112 400G ports (transceivers sold separately) with a Marvell 10G Ethernet port and a dedicated Realtek 1G port for the BMC. An ASPEED AST2600 BMC with AMI MegaRAC firmware provides IPMI 2.0 remote management, and an onboard TPM 2.0 handles platform security.
Liquid-Cooled Deskside Tower with 1600 W Titanium Power Supply
The ET900N G3 is a deskside tower measuring 584 × 232 × 565 mm (H × W × D) and weighing 27 kg net. A closed-loop AIO liquid cooling system manages the Grace Blackwell Ultra Superchip's thermal load at approximately 50 dB in Auto Mode at maximum load, and a single 1600 W 80 PLUS Titanium ATX power supply feeds the system. Rated operating temperature is 10 to 35 °C, and the system carries CE, FCC Class A, BSMI, cTÜV SÜD US and KCC certification.
Enterprise Support — Confirmed at Quotation
ASUS provides onsite service after remote diagnosis for the ET900N G3, with coverage that varies by country; ASUS does not publish a flat multi-year warranty figure for this enterprise-class system. SourceIT confirms the exact ASUS enterprise support tier, term and Singapore service coverage in writing before you order, and supplies every unit with a full GST invoice and local procurement support, including ConnectX-8 transceivers, storage and any rack or power infrastructure your deployment needs.
ASUS ExpertCenter Pro ET900N G3 Datasheet — Technical Specifications
General
| Product Name | ASUS ExpertCenter Pro ET900N G3 |
| Model / Part Number | ET900N G3 (ASUS SKU family: G3-2SSD, G3-4SSD, G3-2SSD-T, G3-4SSD-T, G3-4SSDT-Pro2K) |
| Category | Deskside AI supercomputer / enterprise AI workstation (NVIDIA DGX Station GB300 platform) |
| Form Factor | Deskside tower (not rack-mount) |
| Manufacturer Support | Onsite service after remote diagnosis; coverage varies by region — SourceIT confirms current ASUS terms before order |
| Local Warranty (Singapore) | Procurement, deployment coordination and GST invoice by SourceIT; ASUS enterprise support terms confirmed at quotation |
Processor & GPU
| Compute Platform | NVIDIA DGX Station architecture — NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip |
| CPU | 1 × NVIDIA Grace CPU, 72 Arm Neoverse V2 cores |
| GPU | 1 × NVIDIA Blackwell Ultra GPU (fifth-generation Tensor Cores, second-generation Transformer Engine) |
| CPU–GPU Interconnect | NVIDIA NVLink-C2C, 900 GB/s (per NVIDIA DGX Station datasheet) |
| AI Performance | Up to 20 PFLOPS FP4 Tensor Core (with sparsity), per NVIDIA — 10 PFLOPS FP8/FP6, 5 PFLOPS FP16/BF16 |
| Model Support | AI models up to 1 trillion parameters, per NVIDIA |
| GPU Partitioning | Multi-Instance GPU (MIG) — up to 7 instances |
Memory
| Total Coherent Memory | 748 GB unified CPU + GPU memory space |
| GPU Memory | 252 GB HBM3e, 7.1 TB/s |
| CPU Memory | 496 GB LPDDR5X, 396 GB/s |
Storage
| OS Storage | 2 × M.2 2280 PCIe 5.0 x4 NVMe (from CPU), pre-configured RAID 1, 2 × 2 TB NVMe SSD installed |
| Data / Training Storage | 2 × M.2 2280 NVMe slots (from ConnectX-8) for training data, open for expansion |
| Maximum Total Storage | Up to 8 TB across four M.2 2280 slots, per ASUS |
Expansion
| PCIe Slots | 3 total — 1 × PCIe 5.0 x16, 2 × PCIe 5.0 x8 |
Networking
| High-Speed Networking | 2 × NVIDIA ConnectX-8 SuperNIC QSFP112 ports, 400 Gb/s each (up to 800 Gb/s); transceivers not included |
| General Ethernet | 1 × Marvell 10G Ethernet (RJ45) |
| Management LAN | 1 × Realtek 1G Ethernet, dedicated to the BMC |
I/O Ports
| Front I/O | 2 × USB 10Gbps Type-C, 2 × USB 10Gbps Type-A, 1 × USB 2.0, headphone and microphone jacks |
| Rear I/O | 4 × USB 10Gbps Type-A, 2 × QSFP112 400G, 1 × 10G Ethernet, 1 × 1G Ethernet (BMC), 1 × mini DisplayPort (BMC), 1 × USB Micro-B (BMC), audio jacks |
Management & Security
| BMC | ASPEED AST2600 with AMI MegaRAC-based firmware, IPMI 2.0 |
| Trusted Platform Module | Onboard TPM 2.0 |
Operating System & Software
| Operating System | Ubuntu with NVIDIA AI Developer Tools preinstalled |
| Software Stack | NVIDIA AI software stack; tested with vLLM; supports NVIDIA NeMo workflows |
Power, Physical & Environmental
| Power Supply | 1600 W 80 PLUS Titanium ATX; input 115–240 V AC, 15–8 A, 50–60 Hz |
| Cooling | Closed-loop AIO liquid cooling; approx. 50 dB in Auto Mode at maximum load (without discrete VGA) |
| Dimensions (H × W × D) | 584 × 232 × 565 mm (23.0 × 9.1 × 22.3 in) |
| Weight | 27 kg net; 32 kg gross |
| Operating Temperature | 10 – 35 °C |
| Certifications | CE, FCC Class A, BSMI, cTÜV SÜD US, KCC |
Compatibility & Software Support
NVIDIA AI Software Stack
Ships with Ubuntu Linux and the NVIDIA AI Developer Tools preinstalled — the same NVIDIA software foundation used across the DGX family. CUDA, cuDNN, TensorRT and TensorRT-LLM, NVIDIA NeMo, NVIDIA NIM microservices, Triton Inference Server, RAPIDS and NGC containers run natively on the Grace Blackwell Ultra platform, alongside PyTorch, TensorFlow, JAX, vLLM, Hugging Face Transformers, DeepSpeed and ONNX Runtime. Docker and Kubernetes deployment are supported for containerised MLOps pipelines. NVIDIA AI Enterprise is available as a separate subscription for supported production deployments.
Arm64 Architecture — Plan Your Software Accordingly
The Grace CPU is an Arm Neoverse V2 processor, so the operating system, drivers and container images are arm64 (aarch64) builds. Mainstream AI frameworks and NVIDIA NGC containers are published for arm64, but proprietary x86-only applications will not run natively — check any commercial software you rely on before ordering. NVIDIA has announced a separate DGX Station for Windows SKU (Windows plus Windows Subsystem for Linux) for a later release; the units SourceIT supplies today ship with Ubuntu.
Clustering and Networking Compatibility
The two NVIDIA ConnectX-8 SuperNIC QSFP112 ports run at 400 Gb/s each (up to 800 Gb/s aggregate) with RDMA over Converged Ethernet, GPUDirect RDMA and GPUDirect Storage support, so two or more systems can be linked for distributed training, or connected to an NVMe-over-fabric storage array. QSFP112 transceivers or direct-attach cables are ordered separately to match your switch fabric — SourceIT will quote compatible optics alongside the system. The 10GbE RJ45 port covers everyday site networking, and the dedicated management LAN keeps out-of-band BMC traffic isolated from production data.
Positioning Within the ASUS Range
The ET900N G3 sits above ASUS's conventional ExpertCenter desktops, ProArt workstations and NUC / PN-series Mini PCs — those are Windows business PCs with integrated NPUs for on-device Copilot+ features. The ET900N G3 is a different category entirely: a Linux-based, GPU-accelerated deskside supercomputer for model training, fine-tuning and inference. If you are unsure which category fits your project, email sales@sourceit.com.sg describing the workload and we will help you scope the right system.
Package Contents & Ordering Information
What Ships With the System
- ASUS ExpertCenter Pro ET900N G3 system unit with NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip
- 2 × 2 TB NVMe M.2 SSD pre-installed and configured as OS RAID 1 (base G3-2SSD configuration; 4-SSD and transceiver-bundled variants available to order)
- Power cord for the 1600 W PSU
- ASUS enterprise system documentation
QSFP112 transceivers (unless a -T variant is ordered) and additional NVMe SSDs for the training-data slots are quoted separately to match your networking and storage requirements.
NVIDIA DGX Station GB300 Systems at SourceIT Singapore
The ASUS ExpertCenter Pro ET900N G3 is scoped and quoted individually by SourceIT — we confirm SSD configuration, transceivers, power circuit requirements and the applicable ASUS enterprise support terms before finalising a quotation.
- MSI XpertStation WS300 — NVIDIA DGX Station GB300 platform, same 748 GB / 20 PFLOPS core compute
- HP ZGX Fury AI Station G1n — NVIDIA DGX Station GB300 platform, same 748 GB / 20 PFLOPS core compute
- GIGABYTE W775-V10-L01 — NVIDIA DGX Station GB300 platform, same 748 GB / 20 PFLOPS core compute
- NVIDIA DGX Spark — desktop-scale GB10 system (128 GB unified memory, 1 PFLOP FP4) for individual developers
Browse the full NVIDIA DGX Spark & DGX Station collection. For volume, project, research-grant or education pricing, GST invoicing and current lead times, email sales@sourceit.com.sg.
Datasheet & Downloads
Official ASUS documentation for the ExpertCenter Pro ET900N G3:
- ASUS ExpertCenter Pro ET900N G3 datasheet (PDF)
- ASUS ExpertCenter Pro ET900N G3 — product page (Singapore)
- ASUS ExpertCenter Pro ET900N G3 — full technical specifications
- NVIDIA DGX Station (GB300) — official platform page (Singapore)
- NVIDIA DGX Station datasheet (PDF)
For deployment guides, current ASUS enterprise support terms or Singapore availability, email sales@sourceit.com.sg quoting ET900N G3.
Frequently Asked Questions — ASUS ExpertCenter Pro ET900N G3
What is the ASUS ExpertCenter Pro ET900N G3?
A deskside AI supercomputer built on NVIDIA DGX Station architecture around the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, with 748 GB of coherent CPU–GPU memory and up to 20 PFLOPS of AI performance for developing, fine-tuning and serving large language models of up to 1 trillion parameters locally. It is not a general office PC or a conventional engineering workstation.
Is this the same as the NVIDIA DGX Station GB300?
Yes — the ET900N G3 is ASUS's build of the NVIDIA DGX Station GB300 reference platform. The GB300 Superchip, 748 GB coherent memory, 20 PFLOPS FP4 compute, ConnectX-8 networking and NVIDIA AI software stack are the same across every OEM DGX Station; the OEMs differ in chassis, cooling, storage configuration, management firmware and support terms.
How does it compare to the NVIDIA DGX Spark?
Different class entirely. The DGX Spark is a GB10 desktop system with 128 GB unified memory and 1 PFLOP FP4 for models up to about 200 billion parameters; the ET900N G3 delivers 748 GB and up to 20 PFLOPS for models up to 1 trillion parameters. Choose the Spark for an individual developer, the ET900N G3 for a team or production-scale AI.
What operating system does it run?
Ubuntu with the NVIDIA AI Developer Tools stack preinstalled. ASUS has indicated that Windows-based AI development support is planned, but Ubuntu is the shipping configuration today.
Can several people share one system?
Yes. Multi-Instance GPU (MIG) partitions the Blackwell Ultra GPU into up to seven isolated instances, so multiple users or workloads run concurrently with guaranteed resources.
Does it need a server room or a liquid-cooled rack?
No rack is required — cooling is a closed-loop AIO liquid loop built into the deskside tower. Plan an air-conditioned space within the 10–35 °C operating range and a dedicated power circuit for the 1600 W Titanium PSU under sustained AI load.
What networking does it need, and are transceivers included?
Two NVIDIA ConnectX-8 QSFP112 400G ports handle clustering and high-speed storage, with 10G and 1G Ethernet for general and management traffic. Transceivers are not included on the base SKU (ASUS offers -T variants that add two) and must match your switch fabric — SourceIT quotes compatible optics alongside the unit.
What warranty or support does ASUS provide in Singapore?
ASUS states onsite service after remote diagnosis, with coverage varying by country, and publishes no flat multi-year figure for this system. SourceIT confirms the current ASUS enterprise support term and Singapore coverage in writing as part of your quotation.
How do I buy the ASUS ET900N G3 in Singapore?
SourceIT is an authorised IT reseller in Singapore and supplies the ET900N G3 with a full GST invoice, local procurement support and deployment scoping — email sales@sourceit.com.sg with your workload and site details for a quotation and current lead time.
