ASUS ExpertCenter Pro ET900N G3 workstation provides enterprise server power for local neural network training with NVIDIA Grace Blackwell
ASUS begins to offer its high power hardware unit ExpertCenter Pro ET900N G3. Although it is labeled by the manufacturer as a desktop workstation, its internally calculated performance capability can rival an enterprise server architecture directly. As stated in ASUS official launch announcements, the system is designed for local training of advanced neural networks, executing massive machine learning calculations without the need of remote cloud based databases.
The CPU part of the workstation adopts an innovative processor designed by NVIDIA; it combines an NVIDIA Grace CPU with 72 cores of ARM Neoverse V2 architecture and an NVIDIA GB300 Grace Blackwell Ultra graphics accelerator. With the hardware combination the system is capable of an operational speed peak of 20 petaflops when performing demanding artificial intelligence processes which is only normally found in enterprise data centers, but now is in a desktop workstation format.
One of the most distinguishing parts of the ExpertCenter Pro ET900N G3 is the 748GB unified system memory; the RAM are split between two types of memory to give the highest throughout. The CPU utilizes 496GB of LPDDR5X system memory, while the graphics accelerator utilize 252GB of ultra fast HBM3e system memory, providing enough bandwidth to locally process and train the machine learning models with up to 10^12 parameters.
The workstation is powered and cooled efficiently with its efficient 1600W power supply and an advanced liquid cooling system to make sure that no throttling occurs when performing long computation task for days. It is also expandable with 3 PCIe 5.0 slots and 4 M.2 NVMe storage to allow for fast access of big datasets, also features high speed network interfaces for local network data transfers. The entry price for an ExpertCenter Pro ET900N G3 configuration starts at $99999.
