Here's how you arrive at that number: 100 billion neurons (correct), each firing at 200 Hz (big overestimate, all the neurons are never firing at their max speed. A more typical number would be a few Hz), each sending signals to 1000 other neurons (underestimate, I think. The average number of synapses per neuron is about 7000). You now multiply those numbers together and say that that's the total number of calculations. That's how you get 20 million billion.

Let's do the same for a CPU. A modern Intel i7 has 1.4 billion transistors, each cycling at about 4 Ghz. Each transistor is connected to three others. So we get a total of 20 billion billion "calculations" per second.

Wow, that's a lot! But it's also nonsense. A single transistor sending a signal to another transistor isn't a useful calculation. And a single neuron firing at another neuron isn't a useful calculation either. Each neuron fires based on the total firing rate it receives, and a series of pulses is needed to stimulate it to fire itself. Secondly, lots of neurons firing together is needed to achieve even very basic things. The "20 million billion" number for the brain is probably overestimated by at least as much as the "20 billion billion" number for the CPU.

So why is the brain's output so much more impressive than a CPU's output? Probably for the same reason that a 1 MHz computer running quicksort performs better than a 1 GHz computer running bogosort. Algorithms matter.