GEEKOM has successfully deployed the DeepSeek V4 Flash model across a distributed cluster of four GEEKOM A9 Mega systems. Connected via USB4, each unit features the AMD Ryzen AI Max+ 395 processor, which integrates 16 Zen 5 CPU cores and Radeon 8060S graphics.
The platform utilizes Ubuntu, ROCm, and DwarfStar to distribute the model, offering an OpenAI-compatible API for AI agents. Performance benchmarks show a P95 time to first token of 0.42 seconds and 14.61 tokens per second with contexts up to 250K tokens.
This scalable hardware solution allows organizations to maintain data privacy by keeping source code and credentials on local infrastructure. It supports local retrieval-augmented generation and code analysis while scaling from one to four total systems.