ASUS ExpertCenter Pro ET900N G3-8T-03 8TB RTX PRO 4000 Blackwell

Code: ASUS-ECP-G3-8T-03
New
Not rated Rating details
131 147 € / pcs 108 386 € excl. VAT Measure price:
In Stock (>5 pcs)
Delivery to:
19/10/2026

ASUS ExpertCenter Pro ET900N G3-8T-03 8TB RTX PRO 4000 Blackwell – professional AI supercomputer featuring NVIDIA GB300 Grace Blackwell Ultra, a 72-core NVIDIA Grace CPU, NVIDIA Blackwell Ultra GPU, 748GB coherent memory (496GB LPDDR5X + 252GB HBM3e), dedicated NVIDIA RTX PRO 4000 Blackwell SFF graphics with 24GB GDDR7 ECC, 8TB NVMe SSD, up to 20 PFLOPS AI platform performance, NVIDIA ConnectX-8 2× 400Gb/s networking, 10GbE, 1600W Titanium power supply and Ubuntu with NVIDIA AI Developer Tools.

Detailed information

Product detailed description

The ASUS ExpertCenter Pro ET900N G3-8T-03 8TB RTX PRO 4000 Blackwell is a professional AI supercomputer based on the NVIDIA DGX Station GB300 architecture. It brings data-center-class computing power to a standalone workstation and is designed for generative AI development, large language model training and fine-tuning, scientific computing, advanced simulations and large-scale data processing. At its core is the NVIDIA GB300 Grace Blackwell Ultra Superchip, combining a 72-core NVIDIA Grace CPU with an NVIDIA Blackwell Ultra accelerator. The platform delivers up to 20 PFLOPS of performance in supported AI workloads.

The NVIDIA GB300 Grace Blackwell Ultra connects its processor and graphics accelerator through the high-speed NVIDIA NVLink-C2C interconnect. The NVIDIA Grace CPU features 72 Arm Neoverse V2 cores and 496 GB of LPDDR5X memory with bandwidth of up to 396 GB/s. The integrated NVIDIA Blackwell Ultra accelerator provides an additional 252 GB of HBM3e memory with bandwidth of up to 7.1 TB/s. The total 748 GB of coherent memory enables efficient CPU–GPU collaboration within a shared memory space, reducing the need to transfer data between separate memory pools.

The exceptional memory capacity enables local deployment and fine-tuning of large AI models, including selected models with up to 1 trillion parameters under suitable quantization and configuration conditions. The system is suitable for large language models (LLMs), multimodal AI, image and video generation, autonomous AI agents, robotics, computer vision and scientific simulations. Local data processing also helps reduce dependence on cloud infrastructure while providing greater control over sensitive information.

The configuration includes an additional professional NVIDIA RTX PRO 4000 Blackwell SFF Edition graphics card with 24 GB of GDDR7 ECC memory. The GPU features 8,960 CUDA cores, fifth-generation Tensor cores and fourth-generation RT cores. It delivers up to 24 TFLOPS of FP32 performance, up to 770 AI TOPS and up to 73 TFLOPS of RT core performance. Its 192-bit memory interface provides 432 GB/s of bandwidth. With a low 70 W power consumption and compact SFF design, it adds professional 3D rendering, CAD, visualization and other GPU-accelerated capabilities. Its dedicated 24 GB memory is separate from the GB300 Superchip's 748 GB of coherent memory.

The dedicated graphics card provides 4× Mini DisplayPort 2.1b outputs and supports up to four monitors simultaneously. Supported configurations include up to 4× 4K at 165 Hz or 2× 8K at 100 Hz. Two ninth-generation NVIDIA NVENC encoders and two sixth-generation NVIDIA NVDEC decoders provide hardware acceleration for professional video processing, including AV1. Support for NVIDIA RTX, CUDA, ray tracing and modern AI technologies expands its capabilities for design, development and content creation.

For the operating system, AI models, training datasets and projects, the system provides 8 TB of NVMe SSD storage. The platform offers four M.2 2280 slots in total. Two slots support PCI Express 5.0 x4 and are primarily intended for system drives with RAID 1 support, while two additional slots use PCI Express 6.0 x4 for data-intensive workloads. Fast SSD storage enables efficient loading of large models, handling of substantial data files and operation of professional development environments.

Professional networking includes an NVIDIA ConnectX-8 SuperNIC with two QSFP112 ports supporting up to 400 Gb/s per port, a dedicated Marvell 10GbE port and a separate Realtek 1GbE management port. High-speed connectivity enables integration with additional compute nodes, network storage and data infrastructure. Remote management is handled by an ASPEED AST2600 controller running AMI MegaRAC firmware, with support for IPMI 2.0 and DMTF Redfish. A TPM 2.0 security module is also included.

The ASUS ExpertCenter Pro ET900N G3 is ready for the professional NVIDIA AI ecosystem. It comes with Ubuntu and NVIDIA AI Developer Tools, providing an environment for developing, testing and deploying AI applications. The platform supports NVIDIA CUDA, CUDA-X and modern frameworks such as PyTorch, TensorFlow and NVIDIA TensorRT with appropriate software configuration. It is suitable for neural network training, inference, model optimization, AI agent workflows and large-scale computational tasks.

The robust chassis measures 232 × 584 × 565 mm and features a 1600 W Titanium-class power supply. Its construction is designed for sustained computational workloads and efficient cooling of high-performance components. Front connectivity includes two 10 Gb/s USB-C ports, two 10 Gb/s USB-A ports, one USB 2.0 port and separate headphone and microphone jacks. The rear panel provides four additional 10 Gb/s USB-A ports, network interfaces, remote management connections, a Mini DisplayPort for BMC and three analog audio jacks. The system weighs approximately 27 kg.

Key features

  • Model: ASUS ExpertCenter Pro ET900N G3-8T-03
  • Device type: professional AI supercomputer / workstation
  • Platform: NVIDIA DGX Station GB300
  • Superchip: NVIDIA GB300 Grace Blackwell Ultra
  • Processor: NVIDIA Grace, 72 Arm Neoverse V2 cores
  • Primary AI accelerator: NVIDIA Blackwell Ultra
  • CPU–GPU interconnect: NVIDIA NVLink-C2C
  • Total coherent memory: 748 GB
  • CPU memory: 496 GB LPDDR5X
  • CPU memory bandwidth: up to 396 GB/s
  • Primary GPU memory: 252 GB HBM3e
  • Primary GPU memory bandwidth: up to 7.1 TB/s
  • AI platform performance: up to 20 PFLOPS in supported AI workloads
  • Dedicated graphics card: NVIDIA RTX PRO 4000 Blackwell SFF Edition
  • Dedicated GPU architecture: NVIDIA Blackwell
  • Dedicated GPU CUDA cores: 8,960
  • Dedicated GPU Tensor cores: 5th generation
  • Dedicated GPU RT cores: 4th generation
  • Dedicated GPU FP32 performance: up to 24 TFLOPS
  • Dedicated GPU AI performance: up to 770 AI TOPS
  • Dedicated GPU RT performance: up to 73 TFLOPS
  • Dedicated GPU memory: 24 GB GDDR7 ECC
  • Dedicated GPU memory interface: 192-bit
  • Dedicated GPU memory bandwidth: 432 GB/s
  • Dedicated GPU interface: PCI Express 5.0 x8
  • Dedicated GPU outputs: 4× Mini DisplayPort 2.1b
  • Maximum displays: 4
  • Supported display configurations: 4× 4K at 165 Hz or 2× 8K at 100 Hz
  • Dedicated GPU video encoders: 2× NVIDIA NVENC, 9th generation
  • Dedicated GPU video decoders: 2× NVIDIA NVDEC, 6th generation
  • Dedicated GPU maximum power: 70 W
  • Dedicated GPU cooling: active
  • Dedicated GPU form factor: SFF, low-profile, dual-slot
  • Storage: 8 TB NVMe SSD
  • M.2 slots: 4× M.2 2280
  • System storage: 2× M.2 PCIe 5.0 x4, RAID 1 support
  • Data storage: 2× M.2 PCIe 6.0 x4
  • Expansion slots: 1× PCIe 5.0 x16, 2× PCIe 5.0 x8
  • High-speed networking: NVIDIA ConnectX-8 SuperNIC, 2× QSFP112, up to 400 Gb/s per port
  • Ethernet: 1× Marvell 10GbE
  • Management LAN: 1× Realtek 1GbE
  • Remote management controller: ASPEED AST2600
  • Management: AMI MegaRAC, IPMI 2.0, DMTF Redfish
  • Security: TPM 2.0
  • Front USB: 2× USB-C 10 Gb/s, 2× USB-A 10 Gb/s, 1× USB 2.0
  • Front audio: headphone and microphone jacks
  • Rear USB: 4× USB-A 10 Gb/s, 1× Micro USB for BMC
  • Rear system video output: 1× Mini DisplayPort for BMC
  • Rear audio: 3× analog audio jacks
  • Operating system: Ubuntu
  • Development tools: NVIDIA AI Developer Tools
  • Power supply: 1600 W Titanium
  • Operating temperature: 10–35 °C
  • Dimensions: 232 × 584 × 565 mm
  • Weight: approximately 27 kg
  • Applications: generative AI, LLMs, inference, fine-tuning, deep learning, AI agents, scientific simulation, CAD, 3D rendering, professional visualization

Additional parameters

Category: AI Computers (AI Systems)
Warranty: 36 months (Company 36)
EAN: 4711636684392

Product rating

Be the first who will post an article to this item!

Do not fill out this field

How would you rate this product? Choose 1 to 5 stars, where 1 is the worst and 5 is the best rating.

By submitting a review, you agree to the Privacy Policy.

Discussion

Be the first who will post an article to this item!

Do not fill out this field

Security check

By submitting a comment, you agree to the Privacy Policy.

ASUS represents a combination of high technology, precision design and reliable performance. For more than three decades it has been among the most important manufacturers of computer components, notebooks, monitors and gaming equipment. Its products are known for quality, innovation and attention to detail.

In the game segment ASUS became famous mainly by series ROG (Republic of Gamers) and TUF Gaming, setting performance and resistance standards. The brand has gained the confidence of professional players and demanding users worldwide thanks to continuous development and technical innovation.

ASUS is synonymous with reliability, high quality processing and modern design. Each product is designed to provide maximum performance and bring users an exceptional experience - whether it is work, acting or everyday use.

Manufacturing company : ASUSTeK Computer Inc.
Address : No. 15, Li-Te Rd., Beitou Dist., Taipei City 112, Taiwan
E-mail : info@asus.com
EU representative of the manufacturer : ASUS Computer GmbH
Address of the EU representative : Harkortstr. 21-23, 40880 Ratingen, Germany
Email of the EU representative : support@asus.com