What is inference?
Inference is the act of using a trained model to produce output, such as answering a prompt. Training, by contrast, means creating or adjusting the model's weights, which needs far more compute. Every machine in this catalog is aimed at inference. LLM inference has two phases. In prefill, or prompt processing, the model reads your input; this phase is compute-bound and benefits from GPU power. In decode, it generates tokens one at a time; this phase is limited mainly by memory bandwidth. The Apple Mac mini (M5 Pro) illustrates both: its GPU Neural Accelerators speed prompt processing, while its 307 GB/s bandwidth drives token generation. When choosing hardware, match it to your workload. Long documents stress prefill, while long chatty answers stress decode.