GEEKOM has demonstrated a four-node cluster of its A9 Mega mini PCs, linked via USB4, running the DeepSeek V4 Flash AI model locally instead of in the cloud. Each A9 Mega uses an AMD Ryzen AI Max+ 395 chip, letting organizations build a private, scalable AI platform for document analysis, code review, and agent workflows while keeping data in-house.
GEEKOM has shown that its A9 Mega mini PCs can do more than sit on a desk running everyday workloads — the company deployed DeepSeek V4 Flash across a cluster of four A9 Mega units connected over USB4, according to TechPowerUp. Rather than relying on a data-center server or a proprietary high-speed switch, the four compact machines act as a single distributed AI platform, each contributing its AMD Ryzen AI Max+ 395 chip, 16 Zen 5 CPU cores, Radeon 8060S graphics, and unified memory to the effort.
The setup runs on Ubuntu with ROCm and DwarfStar handling the distribution of an optimized DeepSeek V4 Flash model across the four nodes, with an OpenAI-compatible API connecting applications and AI agents to the cluster. This matters for organizations that want to keep prompts, documents, source code, credentials, and intermediate results on local infrastructure rather than sending them to a public cloud — useful for private knowledge assistants, document review, source-code analysis, local retrieval-augmented generation, and controlled workflow automation for agent systems like Hermes Agent.
In testing, the cluster handled contexts up to 250K tokens, positioning it for long documents, large codebases, and multi-step jobs rather than just quick chat responses. At single concurrency, GEEKOM reported roughly 14.61 tokens per second with a P95 time to first token of about 0.42 seconds in 32- and 128-token tests — figures the company frames as evidence of stronger acceleration for long prompts rather than raw generation speed alone.
What stands out is the scalability: businesses can start with one or two A9 Mega units and grow to four as needs increase, with each machine able to operate independently or as part of the distributed cluster. For labs, classrooms, and edge environments looking to run capable AI locally without provisioning traditional server hardware, this approach turns a set of mini PCs into a flexible, self-contained AI platform.
What We Know
| Spec | Detail |
|---|---|
| Processor | AMD Ryzen AI Max+ 395 |
| CPU Cores | 16 Zen 5 cores |
| Integrated Graphics | Radeon 8060S |
| Memory | Unified memory architecture |
| Cluster Interconnect | USB4 (up to four A9 Mega units) |
| AI Software Stack | Ubuntu, ROCm, DwarfStar, OpenAI-compatible API |
| Context Window | Up to 250K tokens |
| Single-Concurrency Performance | ~14.61 tokens/sec, P95 TTFT ~0.42s (32/128-token tests) |
Frequently Asked Questions
How many GEEKOM A9 Mega units make up the AI cluster?
GEEKOM demonstrated the cluster using four A9 Mega mini PCs connected together via USB4, according to TechPowerUp.
What processor powers each A9 Mega in the cluster?
Each A9 Mega is powered by the AMD Ryzen AI Max+ 395, which combines 16 Zen 5 CPU cores with Radeon 8060S graphics and unified memory.
How do the mini PCs communicate in the cluster?
The four A9 Mega systems connect through USB4 as a distributed platform, eliminating the need for a proprietary high-speed switch or server rack.
Can businesses start small and scale up the cluster?
Yes, organizations can begin with one or two A9 Mega systems and scale up to four as their AI workload needs grow, with each unit able to run independently or as part of the cluster.
