ASUS ExpertCenter Pro ET900N G3 AI Supercomputer Pairs Grace Blackwell Ultra with 748GB Coherent Memory

The ASUS ExpertCenter Pro ET900N G3 deskside AI supercomputer is now available worldwide, bringing data-center-class compute capabilities directly to local workspaces. Based on the NVIDIA DGX Station architecture, this tower workstation targets enterprises, AI developers, and researchers who run intensive local training and inference workloads. The hardware runs on the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, which provides up to 20 PFLOPS of AI performance.

This superchip combines a 72-core Grace CPU using Arm Neoverse V2 cores with a Blackwell Ultra GPU, connected via a high-bandwidth NVLink-C2C interface. This setup provides 748GB of unified memory, allowing developers to run massive AI models locally without cloud latency. It represents a significant hardware upgrade over traditional workstations, similar to how the QNAP QAI-H1290FX AI NAS brings enterprise-grade Blackwell graphics to local network storage.

Side profile view of the ASUS ExpertCenter Pro ET900N G3 workstation detailing its metal side panel and ventilation grills.
This deskside workstation includes an advanced thermal architecture for demanding workloads.
Rear view of the ASUS ExpertCenter Pro ET900N G3 deskside PC showing its back I/O ports and exhaust fan.
The rear I/O section provides high-speed networking and display outputs.

Also read: Lenovo Legion 7a Gen 11 with Ryzen AI Max Launched

Amazon Prime Day Deals

For software, the deskside supercomputer supports the NVIDIA AI Enterprise software stack and includes Ubuntu alongside CUDA-X developer tools. This setup is fully compatible with the NVIDIA NemoClaw open-source stack, enabling developers to build and execute secure, local AI agents. The NemoClaw framework includes NVIDIA OpenShell to enforce policy-based privacy guardrails, keeping sensitive data inside the local network.

ASUS validated these local agentic capabilities through stress testing with the open-source Qwen model. Running on vLLM, the workstation reached an output throughput of approximately 864 tokens per second, with combined input and output speeds hitting 1,600 tokens per second. Networking performance gets an additional boost from an NVIDIA ConnectX-8 SuperNIC, which supports transfer speeds up to 800 Gb/s and allows linking two supercomputers together.

Open side view of the ASUS ExpertCenter Pro ET900N G3 workstation interior showing liquid cooling blocks, fan frames, and internal components.
The interior has a high-performance cooling system to manage thermals under full loads.
Directly facing front view of the ASUS ExpertCenter Pro ET900N G3 deskside PC showing its carbon-patterned mesh panel and front ports.
The front panel has easy-access USB ports, audio jacks, and a power button.

The workstation includes a data center-grade thermal architecture with optimized airflow to sustain high workloads without thermal throttling. The physical tower chassis measures 565 mm deep, 584 mm tall, and 232 mm wide. For storage expansion, the motherboard has four M.2 2280 NVMe PCIe 5.0 SSD slots for a total capacity of up to 8TB.

In addition to the Grace Blackwell Ultra processor, the motherboard supports one optional NVIDIA RTX PRO Blackwell GPU to enable real-time simulation and visualization. Interested buyers can register on the ASUS website to request a pre-sales consultation and obtain localized pricing details.

Source: ASUS

Related Articles

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Latest Articles