- Asus ExpertCenter Pro ET900N G3 conceals Nvidia’s powerhouse chip
- Equipped with up to 784GB of memory, it tackles models that the RTX 5090 struggles with
- Silent operation and supercomputer-like AI performance from a desktop setup
Asus has introduced a formidable desktop solution that packs petaflop-scale AI capabilities into an unsuspecting chassis—one that resembles a typical business tower rather than a flashy server.
The ExpertCenter Pro ET900N G3 sports a design that mimics standard office PCs, complete with a DVD drive and an unusual slot that might remind you of early 2000s technology.
What powers this machine is the impressive Nvidia GB300 Ultra, a dual-module setup that unites a Grace CPU with a Blackwell GPU through NVLink-C2C, Nvidia’s high-speed interconnect. This combination makes it particularly suited for programming and AI tasks.
Crafted for advanced AI applications
This unified architecture means the CPU and GPU can tap into a shared memory pool, which minimizes latency and optimizes performance for large-scale AI projects. The system boasts up to 20 PFLOPS of power, vital for training expansive language models or executing inference on high-parameter datasets.
A remarkable 784GB of coherent memory is supported, surpassing the combined VRAM of a four-card RTX 6000 Ada workstation by more than double.
This extensive memory capability is crucial for developers and researchers who work with models pushing beyond what mainstream GPUs like the GeForce RTX 5090 can manage, which maxes out at 32GB of VRAM.
Furthermore, the ExpertCenter Pro ET900N G3 is compatible with the Nvidia ConnectX-8 SuperNIC, facilitating high-bandwidth networking between multiple systems. This feature enables integration in clusters or within extensive enterprise AI frameworks.
Remarkably, despite its powerful capabilities, it remains in a desktop form factor, negating the necessity for rack mounting, custom cooling systems, or other data center prerequisites.
On the software front, the system operates on Nvidia DGX OS, a bespoke distribution of Ubuntu designed explicitly for AI workloads. It offers seamless support for the complete Nvidia software stack—encompassing CUDA, TensorRT, alongside libraries for machine learning and data science.
Add to that its ability to scale remotely, and the ET900N G3 can easily align with additional DGX systems when greater computational power is required.
