A New Standard in Workstation Power
ASUS has officially launched its ExpertCenter Pro ET900N G3, a workstation engineered for elite computing tasks. Powered by Nvidia’s GB300 Grace Blackwell Ultra Superchip, this system is designed for users who require data-center-level performance within a desktop form factor.
The unit is substantial, weighing 27kg and measuring 584 x 232 x 565 mm. It is clearly intended for fixed installations rather than mobile use. The chassis features an industrial aesthetic, characterized by a brushed silver finish, a dark mesh front, and heavy-duty top-mounted metal handles, which allow for safer handling of its considerable mass.
Advanced Cooling and Architecture
To manage the significant heat generated by its 72 Arm cores and high-capacity memory, ASUS has implemented a sophisticated enterprise-grade, sealed liquid cooling loop. This system utilizes copper cold plates to dissipate heat from the Superchip, memory modules, and network interface. Additionally, a dedicated internal fan ensures airflow reaches the optical transceivers on the network card, mitigating the risk of thermal throttling.
Energy requirements are met by a 1,600-watt 80 PLUS Titanium-certified power supply, ensuring maximum efficiency during heavy computational loads. For storage reliability, the system comes with a mirrored array of two 2TB NVMe drives. Furthermore, the internal configuration supports future expansion via three PCIe slots and two additional bays for supplementary storage or GPUs.
Performance Capabilities
The architecture of the ET900N G3 excels in AI-driven workloads. By utilizing a coherent memory fabric totaling 748GB, the system minimizes latency, allowing it to process massive models efficiently. In benchmark tests, such as MiniMax M2.7, the workstation achieved an output of 4,793 tokens per second across 128 simultaneous streams.
Industry experts have praised the device's capabilities, with Brian Beeler from StorageReview noting: «The GB300 DGX Station remains the most capable thing we've ever put on a desk.»
The workstation’s connectivity is equally impressive, featuring two 400-gigabit networking ports. These allow for a combined bandwidth of 800Gb/s, enabling two stations to be linked to create a unified pool of 1,496GB of coherent memory. This configuration significantly enhances the ability to process even larger AI models.
Real-World Utility
During testing, tech creator Alex Ziskind demonstrated the machine's ability to run dozens of coding, research, and writing agents concurrently without performance degradation. Reflecting on the experience, Ziskind stated: «Everything you're about to see, dozens of AIs all thinking at once, is running on this one machine, not a data center.»
While ASUS has not provided official pricing, industry estimates suggest that the cost of this powerhouse workstation will range between $100,000 and $120,000.