Bosgame showcased a scalable private AI solution by running the DeepSeek‑V3.1 671B‑parameter model on its 7‑Node AI MAX+ 395 cluster. The cluster stitches together seven Bosgame M5 AI Mini PCs via USB4 Direct Connection, turning them into a unified platform that can handle models exceeding a single machine’s memory.

With 896 GB of Unified Memory—672 GB of that as Unified VRAM—the system delivers a substantial resource pool for memory‑intensive workloads. This modular, cost‑effective architecture lets organizations start with one node and add more as demand grows, avoiding the high deployment costs of traditional AI servers.

Beyond inference, the cluster hosts a local API powered by Llama.cpp, supporting development, debugging, Kubernetes ops, and edge AI use cases. By keeping data on local infrastructure, companies maintain tighter control over data management and deployment environments.

The demonstration underscores the growing need for distributed, locally deployed generative AI that scales with user requirements.