True Parallelism (The Kitchen Analogy)

CPUs are fast, but they are sequential devices. Here's another way to think about it. A CPU is sort of like a person working in a kitchen. A multi-core CPU is sort of like multiple people working in the same kitchen. When working together you can create some amazing dishes/meals, with each of you focusing on some specialty and moving around the kitchen in such a way as to share resources. If you both need the same knife or pot, you need to wait for someone to finish. If you are sharing the preparation of a dish, you need to hand it off at the right moment. You may have many tasks running in parallel (oven, mixer, blender, etc), but you can only focus on one at a time. And the resources are typically shared between the various people in the kitchen.

An FPGA, on the other hand, is more like a food factory. Imagine you need to crank out thousands of identical (or very similar) dishes, with a high degree of predictability (determinism). The most efficient way to do that is to set up a large parallel assembly-line like operation, in which each station does only one task, and throughput is achieved by increasing the number of stations doing that task. That means that each station has a dedicated set of resources (e.g, a cutting board, knife, onions), but that you may have many of those stations operating in parallel to increase the number of onions chopped per hour.

The second part of that food factory analogy is the interconnect. If you have ever seen the inside of a food factory, it has equipment meant to connect the different stations together. Conveyor belts, usually, but it could be vehicles, robots, etc. These move partial products from one part of the factory to another.

Now, the reason I like the factory/kitchen analogy is that it fits on a bunch of different levels. A factory is good for producing at scale, identical (and often simplified) recipies. A kitchen excels at flexibility, improvisation, and adaptation. Did you plan on having a block of Tofu, only to find someone else already used it? Find a substitute! Or stop and go get some. Or make it. Or bother a neighbor. You get the idea.

Factories, on the other hand, do not deal well with shortages. How do you adapt an onion station when there are no onions?

This tradeoff between flexibility, and adaptability and the ability to scale is key to why FPGAs are still relevant. They have fewer applications these days than they used to, but there are still plenty of applications where an FPGA can elegantly and efficiently solve a problem that is very hard to do on a microprocessor. Incidentally, you can, of course, build microprocessors on FPGAs, and some FPGAs come with CPUs built in. I won't focus on either of those at this point. A microprocessor on an FPGA is an excellent way to bridge the gap between the two technologies, but requires some fairly advanced techniques that we won't start with.