The cluster utilizes the AMD Ryzen AI Max+ 395 processor, which integrates 16 Zen 5 CPU cores and Radeon 8060S graphics. By leveraging Ubuntu, ROCm, and the DwarfStar framework, the system distributes the AI model across the hardware array. The use of USB4 interlinks removes the traditional requirement for proprietary high-speed switches or server racks, allowing for a scalable, modular architecture.
GEEKOM Deploys Local DeepSeek AI Cluster Using A9 Mega Mini PCs
By linking four A9 Mega mini PCs via USB4, GEEKOM has created a distributed platform capable of running DeepSeek V4 Flash locally. This setup allows enterprises to process sensitive data and complex AI workflows without relying on public cloud infrastructure or expensive, specialized server-grade hardware.
Decentralized Enterprise Computing
Organizations can maintain complete control over internal policies, source code, and private credentials by keeping data within local infrastructure. The system supports advanced tasks such as retrieval-augmented generation and multi-source research, handling context windows up to 250,000 tokens. Performance metrics show a single-concurrency output of approximately 14.61 tokens per second, with a P95 time to first token of 0.42 seconds. This capability allows businesses to scale from a single unit to a four-node cluster based on operational demand, providing a flexible alternative to centralized data centers for edge environments and private research labs.
Comments (0)
No comments yet. Be the first!