This article introduces a method using Intel Optimum-Intel to optimize its performance. The author utilized this method on a LattePanda Sigma, an x86 single board computer/server, leveraging the integrated GPU to accelerate LLaMA 3.1's inference speed, with impressive results.