.jpg "defaultImg.bb7db14%20(1).jpg")
The ASUS ExpertCenter Pro ET900N G3, powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, provides multi-petaflop-scale AI computing in a deskside form factor. According to ASUS, the system features 748GB of coherent unified memory and delivers up to 20 PFLOPS of AI performance, enabling enterprises to run frontier AI models with up to 1 trillion parameters locally.
How does deskside AI impact enterprise data security?
Moving AI processing from cloud-based infrastructure to local deskside units allows companies to maintain stricter governance over proprietary data. By utilizing the NVIDIA DGX Station GB300 architecture, organizations can process sensitive information on-premises, reducing the security risks inherent in transmitting data to external cloud servers. ASUS reports that this local deployment model is specifically designed for businesses requiring high-speed, secure execution of generative AI and autonomous agentic workflows.
What is the performance capacity of the GB300 architecture?
The GB300 Grace Blackwell Ultra Desktop Superchip utilizes NVLink-C2C high-bandwidth interconnect technology to bridge the CPU and GPU. This integration results in 748GB of coherent memory, a significant jump from traditional workstation capabilities. According to ASUS engineering stress tests using the vLLM framework, the system handles the Qwen open-source model at approximately 864 tokens per second. When measuring combined input and output processing, the throughput reaches roughly 1,600 tokens per second, demonstrating the system’s ability to manage large-scale inferencing tasks without the latency associated with cloud-based API calls.
Why are organizations shifting toward local agentic AI?
The rise of autonomous AI agents requires “always-on” compute capabilities that are often too costly or inefficient to maintain via standard cloud subscriptions. By deploying hardware like the ET900N G3, enterprises can integrate NVIDIA NemoClaw workflows to build autonomous assistants that operate in real-time. This trend marks a move away from the “cloud-first” mandate toward a hybrid approach where specialized, high-intensity AI tasks are handled on-site to minimize operational costs and ensure predictable performance.
Frequently Asked Questions
Can the ExpertCenter Pro ET900N G3 run 1 trillion parameter models?
Yes. ASUS states that the 748GB of coherent unified memory is specifically engineered to support the training and inference of frontier AI models reaching up to 1 trillion parameters.
Does this system require a specialized data center environment?
No. The ET900N G3 is built on the DGX Station architecture, which is specifically designed to function as an office-friendly, deskside unit without the need for external cooling or specialized server room power configurations.
What software environments does the system support?
The unit supports the NVIDIA AI software stack, including vLLM and NemoClaw, and ASUS has announced future support for Windows-based AI development environments.
Are you looking to scale your local AI research capabilities? Contact our team for a consultation on deploying enterprise-grade deskside supercomputing in your office.
