Just realized that the word “fast” is very serial. When you can do one thing at a time, the latency (how fast you are) determines the throughput. But with parallelism, throughput is decoupled from latency. For example, a business that sells 12-year-old whiskey has high latency but high throughput that you can make infinite batches per year. And if you make a batch every year, then after 12 years you get to sell a lot every year. And this is called pipelining.