One of the bits in the 8-bits is used to indicate if the number is negative or positive. All numbers are codified using nothing more that 1's and 0's. So if you only have 1 and 0 (bits) to create a number how do you say its negative? You use a bit as a Sign. But if you are using a bit to indicate negative/positive then you have lost the use of a bit to store the number. So an 8-bit unsigned has a greater range of numbers it can store (0-255). In a land cover raster that might be all you need to store the land categories values. Things get even more complex if you want to store numbers with decimal points and thus a float raster usually takes up a lot more disk space.
You should care because you want to be efficiently storing your data. Imagine you had a raster that only ever had the values 0,1,2,3 in them. If used stored that as a float 32 raster then a load of disk space/memory is used because it would need to accommodate the decimal point. Now if you saved that same data as a 2-bit raster that would crush your dataset size down significantly. You would also get a performance boost as the compute needs only to allocate a smaller amount of memory when processing the raster.
So bottom line, never accept the default bit-depth, think about what it is you are storing and a simple change can have real performance boosts in processing and displaying your data.
