AI performance
748 GB
total VRAM
usable 747.4 GB
748 GB
RAM
for offload and data
1×
graphics cards
NVIDIA GB300 · concurrent models
A build for flagship 70B-class models in 4-bit — quality comparable to cloud services, in your own server room.
What runs on this build (4-bit, 8k context)
| Model | Memory | Speed |
|---|---|---|
gpt-oss 20B (OpenAI) OpenAI open-weight model, fast thanks to MoE | Fits13 GB | — |
Mistral Small 3.1 24B European model, strong on documents and instructions | Fits15.7 GB | — |
Gemma 3 27B strongest Gemma, text and images | Fits17.1 GB | — |
Qwen3 30B-A3B (MoE) big-model knowledge at small-model speed | Fits19.1 GB | — |
Qwen2.5 32B demanding writing and analysis | Fits21.8 GB | — |
Llama 3.3 70B flagship class, needs 48 GB of VRAM | Fits45 GB | — |
Memory is computed from model weights and KV cache. Speed is a rough estimate from the graphics memory bandwidth; measured burn-in values are added over time.
The ASUS ExpertCenter Pro ET900N G3 is a deskside AI supercomputer built on the NVIDIA DGX Station architecture. At its heart is the NVIDIA GB300 Grace Blackwell Ultra Superchip: a 72-core Grace CPU and a Blackwell Ultra GPU share 748 GB of coherent memory, so models with up to a trillion parameters run on a single machine under your desk. Up to 20 PFLOPS (FP4) covers inference of large language models, fine-tuning of pre-trained networks and running AI agents for the whole company.
It is built for developers, researchers, data scientists and companies that want to run AI without the cloud: data stays inside your network. It ships with Ubuntu and the NVIDIA AI Developer Tools (PyTorch, TensorFlow, NVIDIA NIM microservices), ready to work as soon as it is plugged in. Two NVIDIA ConnectX-8 ports (QSFP 400G) let you stack two machines into one system with double the memory.
Available configurations: 2× 2 TB SSD (P/N 90WS00B2-M000R0) · 4× 2 TB SSD, 8 TB total (90WS00B2-M000S0) · 4× 2 TB + NVIDIA RTX PRO 2000 Blackwell GPU (90WS00B2-M001B0) · 4× 2 TB + RTX PRO 4000 SFF Blackwell (90WS00B2-M000U0) · 4× 2 TB + RTX PRO 6000 Max-Q (90WS00B2-M000V0). We confirm the configuration and price in the quote.
Equipment: 1600 W 80 PLUS Titanium power supply, 10G LAN plus a dedicated 1G LAN for remote management (BMC), front 2× USB-C 10 Gbps, 2× USB-A 10 Gbps, USB 2.0 and audio, rear 4× USB-A 10 Gbps, mini DisplayPort for BMC and 3× audio. Tower 232 × 584 × 565 mm, 27 kg, with a carrying handle.
3-year on-site warranty in the Czech Republic and Slovakia, extendable by 1 or 2 years. Price on request: the machine is built to order; we confirm the price and lead time in a quote within one business day and help you choose the configuration for the models you want to run.
Manufacturer and product safetyManufacturer: ASUSTeK Computer Inc., 1F., No. 15, Lide Rd., Beitou Dist., Taipei City 112, Tchaj-wan · zástupce v EU: ASUS Computer GmbH, Harkortstrasse 21-23, 40880 Ratingen, Německo · ProductSafety@asus.com
Product identifier: ASUS ExpertCenter Pro ET900N G3 – NVIDIA GB300 AI supercomputer (asus-expertcenter-pro-et900n-g3-nvidia-gb300)
Safety: The device is intended for dry indoor use with sufficient heat dissipation. The manufacturer's documentation and safety instructions are included with the delivery.
LOCAL AI PC
ASUS ExpertCenter Pro ET900N G3 – NVIDIA GB300 AI supercomputer
Deskside AI supercomputer with NVIDIA GB300 and 748 GB of memory: trillion-parameter models, no cloud.
Total Price
Price on requestBuilt to order · we confirm the price and lead time within 1 business day
No-obligation enquiry
Tell us what you want to run on the machine (models, number of users, network setup). We will prepare a quote with the exact configuration, price and lead time.