[Strilanc]

With 1 or 2 qubits you wouldn't even be able to compute 2 + 2 because storing 4 in binary requires 3 bits. 100 qubits is also not particularly useful, not because they aren't fast but because most of the interesting stuff simply doesn't fit. RSA keys are thousands of bits long, for example.

You need enough qubits to store the problem, and the working state of the algorithm you're using, and the overhead introduced by error correcting codes, and the overhead from your gates not being ideal, and so on. Austin Fowler gave a talk a few months ago where he goes through the overheads in a particular problem (finding molecular ground state energies) and ends up optimistically requiring tens of millions of qubits [1].

1: https://www.youtube.com/watch?v=XbFoXT73xVQ&feature=youtu.be...