ASUS rolls out the ExpertCenter Pro ET900N G3, a deskside AI supercomputer that translates data‑center power into office rooms. Built on NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip, the unit delivers 20 PFLOPS of raw compute in a form factor that fits a standard desk today for.

The system ships with 748 GB of coherent unified memory, letting developers train large language models locally without cloud bottlenecks. NVIDIA’s NVLink‑C2C interconnect stitches the high‑bandwidth links, giving teams instant access to memory across the board and slashing latency that traditionally limits workstation AI for researchers and and scientists.

Stress tests with the open‑source Qwen model hit 864 tokens per second, while total input‑output throughput climbed to 1,600 tokens. These figures prove the ET900N G3 can shoulder demanding generative‑AI tasks, making it a viable alternative to cloud‑based inference for sensitive enterprise data and analysis workflows for clients worldwide.

By keeping AI workloads on premises, enterprises gain tighter data governance and lower latency for real‑time decision making. ASUS’s deskside supercomputer therefore positions itself as a cornerstone for firms that need high‑performance AI without migrating to public clouds in today’s digitized world and innovation for growth efforts.