It's a physics/electronic thing. A CMOS sensor has one 'native' level of sensitivity, usually in the ISO 100-200 range. Electronically, it's easier to amplify a signal (gain) than to reduce it, so the latter can only be done to a point.

Bottom line, if you want long exposures in daylight, or if you want to use your f/2 or faster primes wide open in sunlight, you need ND filters.

I use B+W filters, quality is excellent. They come in single- and MRC-coated versions, the MRC is easier to clean but otherwise not really necessary, since the main purpose of a multilayer coating is to reduce transmissive light loss, and the whole point of an ND filter is to lose that light.

The next issue is how strong? For fast prime portraits and adding motion blur to waterfalls, I find a 3-stop filter usually does the job. I don't see the point in anything weaker. Also, a CPL gives ~1.75-stops of darkening, and I sometimes use that with a 3-stop ND for moving water. I also use a 10-stop ND for extreme water blur, cloud blur, and to 'eliminate' people from architectural shots.

What about sizes? Depends on the lenses you'll use them with, of course. One possibility is to get filters sized for your largest diameter lens, and get step-up rings. I tried that - got 77mm filters and adapters. Then I got the 16-35L II, and needed 82mm filters (note: only Schneider Optics, B+W's parent company, make an 82mm threaded 10-stop, AFAIK). Also, I have a 72mm 3-stop since that fits all three of my fast primes (35L, 85L II, 135L) - one issue with step up rings is they usually preclude using a hood.

One other commonly recommended option is a variable ND filter. Some swear by them, others don't (I'm among the latter). There have been some threads on here discussing them, I'd summarize as anything short of the Singh-Ray or Schneider options give you a detectable IQ hit. My issue with them is I often use NDs with wide/ultrawide lenses, and just like a CPL results in uneven polarization with ultrawide lenses, vari-NDs (which are basically a pair of stacked polarizing filters) result in a 'Maltese cross' artifact with UWA lenses.

Probably too much information, but hope that helps...

--John