I will give an example on a 64-bit double in a fraction of 52 bits, if we use BCD in these 52 bits, we will get much more accurate numbers (13 significant digits, in a regular double 15-17, but there will be cumulative errors …).
The image of double

FPU calculates IEEE 754, but then you can also create a BCDU for calculating such BCD with sign and exponent to calcuate such numbers with speed of float. But why is this not done?

Or DPD(it was mentioned by @chtz)?

I know that in decimal we can only divide in 2 and 5 but it will be more useful for people, because we use decimal, and even 1/3, will be more accurate in decimal than binary, because in binary the smallest difference between numbers is 0.015625, in decimal it is 0.000001.

But still binary numbers in fraction have some advantages than BCD or DPD in fraction, that make it the most popular, what are they?