You need a fixed data size for good performance. If you use fixed precision, you get absolute nonsense when doing very common calculations like `tan(x)`. IEEE 754 was masterfully engineered to produce the fastest, most correct results for the most common operations.

We use it because it's fantastic.