Enthusiast achieves landmark: running a 2.8 trillion-parameter model on vintage server without GPU

The escalating size of AI models, often requiring vast, expensive GPU clusters for operation, has been challenged by a recent demonstration. An enthusiast successfully ran Moonshot AI's Kimi K3, a model boasting 2.8 trillion parameters, using only an older server equipped with a 40-core processor and no dedicated graphics card. This remarkable feat suggests a significant breakthrough in model deployment, potentially through the utilization of Optane memory. The test was shared within an online community, highlighting a push against conventional hardware limitations in the field of artificial intelligence.