• Powerful Liquid Cooling System & Sustained Full Power Performance The Starlight GT15 adopts high-efficiency liquid cooling design, delivering stable continuous full-power output even under heavy AI computing workloads. The 15L compact chassis (315×225×210 mm) saves space while maintaining excellent heat dissipation for long-hour model training and rendering tasks.
• Extreme AI Computing Power, Up to 7022 TOPS Support dual-width graphics cards from NVIDIA, Moore Threads, Muxi and more, delivering peak computing power up to 7022 TOPS. It smoothly handles large language models from 27B to 235B parameters for local training, fine-tuning and inference.
• Wide Compatibility for CPU & Rich Expandable Storage Compatible with Intel Xeon, AMD EPYC and domestic Hygon series CPUs. Equipped with 4×R-DIMM DDR memory slots and 2×M.2 2280 NVMe SSD slots (max 8TB). Multiple high-speed ports including 10G SFP optical port, 2.5G Ethernet, USB3.2 Gen2, HDMI and VGA meet diverse connection demands.
• Multi-scenario AI Deployment for Industrial & Research Use Perfect for local LLM deployment, robot embodied intelligence control, edge AI vision inspection, IoT industrial gateway, private cloud & NAS. It protects data privacy by offline local computing and cuts enterprise AI development cost for universities, research labs and manufacturers.
• All-in-one Workstation for Creation, Gaming & Education Capable of professional 3D GIS/BIM design, simulation, 4K lightweight video editing, medium API services. It also works as a compact gaming PC for AAA games. Ideal for AI programming, machine learning and STEAM education training.
• 
  

Specification:

The Starlight GT15 is a 15L compact liquid-cooled AI workstation built for high-demand AI computing. Powered by high-efficiency liquid cooling, it ensures stable continuous full-power output for long-running heavy workloads. Supporting multi-brand dual-width GPUs, it delivers up to 7022 TOPS peak computing power to run, fine-tune and train large models ranging from 27B to 235B parameters.

It features broad CPU compatibility, expandable DDR memory and NVMe SSD storage, alongside abundant high-speed interfaces including 10G SFP optical port and 2.5G network. This versatile machine supports local LLM inference, robot embodied intelligence, edge industrial inspection, IoT gateway, 3D rendering, video editing and AAA gaming. It enables offline local data processing to safeguard data privacy, suitable for enterprise AI development, scientific research, education and content creation.