- Asus ExpertCenter Pro ET900N G3 hides Nvidia’s wildest chip
- With up to 784GB of memory, it handles models that your RTX 5090 simply can’t
- No rack, no noise – just supercomputer-class AI performance on a desktop you can use
Asus has unveiled a new high-performance desktop PC which delivers petaflop-scale AI performance, not in a flashy server rack, but in a surprisingly unassuming chassis.
The Asus ExpertCenter Pro ET900N G3 looks like any standard business tower – so much so it even includes a DVD drive and a curious-looking slot that feels like a throwback to the early 2000s.
At the heart of this business PC is the Nvidia GB300 Ultra, a two-part module that combines a Grace CPU with a Blackwell GPU via NVLink-C2C, Nvidia’s high-bandwidth interconnect, making it ideal for programming and AI development.
Designed for serious AI work
The unified chip architecture enables the CPU and GPU to share a single memory pool, reducing latency and improving efficiency for large-scale AI workloads. The system can deliver up to 20 PFLOPS of performance for training large language models or running inference on high-parameter models.
It supports up to 784GB of coherent memory, more than double the combined VRAM of a workstation with four RTX 6000 Ada cards.
That scale of memory access is essential for developers and researchers working with models that exceed the capabilities of mainstream GPUs like the GeForce RTX 5090, which offers 32GB of VRAM.
The ExpertCenter Pro ET900N G3 also includes support for Nvidia ConnectX-8 SuperNIC, enabling high-throughput networking between systems. This allows it to function in clusters or within a larger enterprise AI deployment.
Despite its performance, it retains a desktop form factor, eliminating the need for rack installation, custom cooling solutions, or the infrastructure demands of a data center.
On the software side, the system runs Nvidia DGX OS, a specialized Ubuntu-based Linux distribution tailored for AI workloads. It provides native support for the full Nvidia software stack, including CUDA, TensorRT, and libraries for machine learning and data science.
It also supports remote scaling, allowing the ET900N G3 to integrate seamlessly with other DGX systems if additional compute power is needed.
