The cluster runs DeepSeek V4 Flash locally, bringing private and scalable AI to businesses.
TAIPEI, Aug. 17, 2026 /PRNewswire/ -- GEEKOM, a leading innovator in high-performance Mini PCs, has deployed DeepSeek V4 Flash across four GEEKOM A9 Mega systems. The achievement shows how GEEKOM is extending its compact-computing expertise into practical enterprise AI.
Rather than relying on a data-center server, the cluster connects four A9 Mega mini PCs through USB4 as one distributed platform. Each A9 Mega is powered by the AMD Ryzen AI Max+ 395, combining 16 Zen 5 CPU cores, Radeon 8060S graphics, and unified memory in a compact chassis.
Ubuntu, ROCm, and DwarfStar distribute an optimized DeepSeek V4 Flash model across the machines. An OpenAI-compatible API connects applications and AI agents, while USB4 eliminates the need for a proprietary high-speed switch or server rack.
Bringing Enterprise AI Closer to Business
The cluster enables organizations to use advanced AI while keeping sensitive information under their control. Prompts, documents, source code, credentials, and intermediate results can remain within local infrastructure instead of being sent to a public cloud.
Businesses can create private knowledge assistants that search and analyze policies, manuals, contracts, reports, and other internal documents. The platform also supports document review, source-code analysis, local retrieval-augmented generation, multi-source research, and controlled workflow automation.
For agent systems such as Hermes Agent, the cluster can process tools, policies, memory, logs, code, and retrieved information before taking action. Its technical path has operated with contexts up to 250K tokens, making it well suited to long documents, large codebases, and complex multi-step jobs.
Proven Performance with Flexible Expansion
At single concurrency, the system delivered approximately 14.61 tokens per second and a P95 time to first token of about 0.42 seconds in the 32- and 128-token tests. More importantly, it provides greater capacity and stronger acceleration for long prompts — not merely faster text generation.
Organizations can start with one or two systems and scale to four as needs grow. Every A9 Mega can operate independently or contribute to distributed inference, maximizing flexibility and long-term value. （Related： Taiwan Weeks 2026 set for Nov. 3-13 with expanded international participation ｜ Latest ）
