User Runs Large AI Model on CPU Using Optane PMem

A Reddit user successfully ran Moonshot AI's Kimi K3, a 2.8-trillion-parameter AI model, solely on a CPU. This was achieved by utilizing Optane PMem 200 with 1-bit quantization, demonstrating an inference speed of 0.85 tokens per second. This achievement highlights the potential for deploying large AI models efficiently on non-GPU hardware, potentially broadening access to advanced AI capabilities.