ASUS Launches ExpertCenter Pro ET900N G3 AI Supercomputer with NVIDIA GB300

ASUS has released the ExpertCenter Pro ET900N G3 as a dedicated deskside supercomputer for artificial intelligence workloads. The system targets enterprises, researchers, and data scientists who require high-performance local computing power. This launch positions the device as a specialized tool for intensive AI training and inference tasks.

Desk-side workstation targets enterprises with unified memory and NVLink interconnects

The workstation runs on NVIDIA DGX Station GB300 architecture to handle complex computational demands. It utilizes the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip as its primary processing unit. This integrated approach consolidates all necessary hardware into one unified unit, eliminating the need for separate expansion modules.

• Processor/Superchip: NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip
• Architecture: NVIDIA DGX Station GB300 architecture
• Memory: 748 GB coherent unified memory
• AI Performance: up to 20 PFLOPS
• Interconnect: NVIDIA NVLink-C2C high-bandwidth interconnect technology

ASUS equipped the ExpertCenter Pro ET900N G3 with 748 GB of coherent unified memory for data-intensive operations. The architecture connects internal components using NVIDIA NVLink-C2C high-bandwidth interconnect technology to ensure rapid data transfer. This configuration supports up to 20 PFLOPS of raw AI performance.

Engineering stress testing demonstrated the system processing approximately 864 tokens per second on the Qwen open-source AI model. Combined input and output operations reached around 1,600 tokens per second during these evaluations. ASUS has made the ExpertCenter Pro ET900N G3 available globally to address the growing need for dedicated artificial intelligence computing resources.