ASUS has taken an ambitious leap by integrating NVIDIA’s GB300 Grace Blackwell Ultra—originally designed for server infrastructure—into a desktop form factor. This custom system, unveiled as the ExpertCenter Pro ET900N G3, harnesses the hybrid design of NVIDIA’s latest CPU-GPU superchip, offering an extraordinary 20 PetaFLOPS of FP4 precision compute. It features a massive 784 GB of unified cache-coherent memory, split between 288 GB of HBM3E situated on the GPU and 496 GB of LPDDR5X dedicated to the Grace CPU. This marks a new paradigm where server-grade computational density moves into workstation desktops specifically for advanced computing workloads.
Distinct from standard desktops, ASUS’s implementation follows NVIDIA’s DGX blueprint, encouraging OEMs to craft purpose-built systems. Connectivity keeps pace with compute performance, thanks to the inclusion of an 800 Gb/s ConnectX-8 SuperNIC, designed to funnel data at hyperscale rates. The system runs NVIDIA’s DGX OS, a customized Ubuntu derivative, featuring kernel tweaks and optimizations to squeeze every drop of performance from the Blackwell architecture. These software-level enhancements underscore the platform’s intent for serious, purpose-driven research rather than general office work.
Designed for professionals and Artificial Intelligence researchers, the motherboard inside the ET900N G3 is engineered for expansion and extreme workloads. It offers three full-length PCIe x16 slots for GPU stacking or deploying specialty accelerators, alongside three M.2 slots for ultra-fast SSDs. Power infrastructure is equally robust, supporting up to 1,800 Watts via dedicated GPU power connectors beyond standard ATX and EPS12V plugs—highlighting readiness for multi-GPU configurations in compute-intensive environments. While ASUS has yet to announce pricing or availability, the advanced hardware and enterprise focus suggest a premium exceeding typical workstation costs, catering directly to those integrating desktop-scale supercomputing into Artificial Intelligence research and development pipelines.
