ASUS Packs Data-Center AI Power Into Deskside ExpertCenter Pro ET900N G3

The ExpertCenter Pro ET900N G3 centers on the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, utilizing NVLink-C2C interconnects to manage intensive workloads. Its standout feature is 748GB of coherent unified memory, allowing researchers to handle models with up to 1 trillion parameters locally. By moving these processes on-premises, enterprises gain tighter control over data privacy and reduce the operational costs associated with continuous cloud reliance.

ASUS has optimized the workstation for the modern AI stack, including support for NVIDIA NemoClaw workflows and autonomous agent development. During internal stress testing with the Qwen open-source model, the machine reached throughput speeds of roughly 864 tokens per second for output, with total processing peaking at 1,600 tokens per second. This hardware is designed to bridge the gap between traditional workstations and full-scale data centers, providing a scalable, enterprise-grade environment for deep learning, simulation, and real-time AI inference.