ASUS desktop AI supercomputer features 748GB memory and Blackwell Ultra GPU
ASUS has launched the ExpertCenter Pro ET900N G3, a deskside AI system based on NVIDIA DGX Station GB300 architecture. The system uses the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip and is aimed at local AI training, inference and agentic AI workloads.
Unlike most AI systems based on data-center hardware, the ET900N G3 looks like a large desktop tower. ASUS calls it a deskside AI supercomputer, which means it is designed for office or lab deployment rather than rack installation.
The main specification is memory capacity. ASUS lists 748GB of coherent memory, split between 496GB LPDDR5X memory for the Grace CPU and 252GB HBM3e memory for the Blackwell Ultra GPU. The CPU side offers 396 GB/s bandwidth, while the GPU memory is rated at 7.1 TB/s.
The system uses a 72-core Arm Neoverse V2 Grace CPU and a single NVIDIA Blackwell Ultra GPU. ASUS lists up to 20 PFLOPS of AI performance and NVIDIA ConnectX-8 SuperNIC networking. The platform also supports NVIDIA NVLink-C2C, which connects the CPU and GPU through a coherent chip-to-chip interface.
Windows support coming
ASUS currently lists Ubuntu with NVIDIA AI Developer Tools as the operating system. However, the company also confirms that future Windows support is planned for Windows-based AI development and agentic environments. This follows NVIDIA’s earlier confirmation that DGX Station for Windows is coming to systems from ASUS, Dell Technologies, GIGABYTE, HP, MSI and Supermicro.
ASUS ExpertCenter Pro ET900N G3
Based on NVIDIA DGX™ Station Architecture
The ASUS ExpertCenter Pro ET900N G3 is part of a new class of deskside AI supercomputers. Designed from the ground up to build and run AI, this revolutionary system unleashes the NVIDIA ® GB300 Grace Blackwell Ultra Desktop Superchip and a massive 748GB of large coherent memory, with future support planned for Windows-based AI development and agentic environments. Fully compatible with NVIDIA NemoClaw and NVIDIA AI Enterprise, it enables the local deployment of trillion-parameter models and autonomous AI agents at the deskside . Capable of developing and running large local AI training and inferencing workloads, this desktop system uses the NVIDIA ® AI Software Stack to put an unprecedented amount of compute performance at the fingertips of AI development teams.
The ET900N G3 supports up to four M.2 2280 NVMe SSDs. ASUS lists two PCIe 5.0 x4 slots for OS RAID 1 and two PCIe 6.0 x4 slots for training data. The system also includes three PCIe expansion slots, with one Gen5 x16 slot and two Gen5 x8 slots.
This is not the first time we have seen the ET900N G3 name. Last year we reported on ASUS GB300 desktop system with a similar design, but the earlier listing used up to 784GB coherent memory. The updated ASUS specification now lists 748GB.
According to UK retailer, the system costs an eye-watering £117.599,99.
Source: Scan UK
Source: ASUS
