A $537 Local LLM Machine (2025) (blog.lewman.com)

A tech enthusiast has demonstrated that it's possible to run local language models (LLMs) on affordable hardware, specifically a refurbished Miniforum UM790 Pro, which boasts an AMD Ryzen 9 7940HS CPU and 64GB of RAM, all for just $537. After setting up the machine with OpenBSD and configuring Ollama for model testing, the user found that performance, while slower than high-end GPUs, is sufficient for practical use. The setup achieved between 5 to 20 tokens per second, with a typical response time of around 30 seconds per query. The significance of this development lies in its potential to democratize access to AI capabilities. Users no longer need to invest in expensive GPUs to run complex models locally, thereby reducing barriers for experimentation and application of AI/ML technologies. Additionally, the upcoming Linux kernel 6.14 promises to harness the built-in Neural Processing Unit (NPU) of the Ryzen 9, which could enhance model performance further. This breakthrough underscores that affordable hardware can adequately support local AI applications, making powerful AI tools accessible to a broader audience.