UK-manufactured 2U Q-Power offers firms an on-premise alternative to premium cloud AI services
G2 Digital, a UK specialist in custom-built rack-mount computers, has launched the 2U Q-Power, a high-performance GPU server designed for complex AI workloads and rendering.
The system can accommodate up to four dual slot Nvidia RTX GPUs, offering twice the capacity of most conventional 2U servers and greater computing density within a rack. This is made possible by a custom chassis design that mounts the GPUs horizontally, parallel to the motherboard.
When configured with four Nvidia RTX Pro 6000 Blackwell Max-Q GPUs, the Q-Power delivers 384 GB of combined GPU memory, enabling organisations to run demanding AI workloads entirely on-site. This includes large language models (LLMs) with 70 billion parameters at full precision, as well as significantly larger models using lower-precision formats and quantisation.
According to G2 Digital, the system can deliver up to 18 times lower costs per million tokens than premium cloud AI APIs when operating under sustained workloads. The company also claims response latency can fall from around 1.5 seconds with cloud APIs to less than 40 milliseconds when running AI locally.
Keeping AI workloads on-premise, says G2 Digital, gives organisations greater control over their data, allowing them to train or fine-tune models using proprietary information without having to send that data to external cloud providers. This can help businesses address data sovereignty and security requirements and support compliance efforts under frameworks such as GDPR, HIPAA and SOC 2.
The 2U Q-Power can be configured with AMD Epyc or AMD Ryzen Threadripper processors, and up to two Nvidia GeForce RTX 5090 GPUs. The system is designed and manufactured in the UK.
