It takes a long time to compute the size of 20 files when a division by 1000 takes 300 odd cycles on a 10kHz machine. It doesn't take such a long time when a right shift 10 takes 1 cycle.
This must be the most clueless post about the 1000/1024 divide so far. It never had anything to do with the computer's performance, it's that when you build a digital computer a lot of things will be sizes of two because what you can address with n bytes will be 2^n. Physical memory, memory pages, caches, buffers, floppy and hard drive sectors all the "microunits" in the computer are powers of two. Hint: No actual hard drive gives you 1MB = 1000000 bytes because it's not divisible with 512, in reality they give you 1954*512 = 1000448 so they don't underdeliver. Actually make that divisible by 4096 for modern HDD drives with 4K (no, not 1000) sectors.
There is a single reason why computer scientists usurped the prefix kilo and that is because they needed to describe "one thousand and twenty four bytes" - or multiples of that - very, very often. They needed a shorter name, they never needed the unit "1000 bytes" and so "one kilobyte" became their shorthand for 1024 bytes. And unless you're really good at doing math in your head, tell me how much is seven kilobytes exactly? (And if you answer 7000 I'll slap you). We still say 512GB of RAM. Nobody wants to say 549.755813888 GB of RAM, because multiply that with a billion and you have how many bytes that is. It's not some nice, round number.
Either way you're going to run into some f*cked up conversions if you mix GiB and GB, which I'll use now for clarity. If you have 512GiB of RAM (hey, servers do) and load 512GB from disk, how much of your RAM have you used up? Now while you're calculating that, this other person who uses a GiB system says so that was like ~477 GiB so like ~35 GiB free? Or you have to say you have 549.8 (rounded) GB RAM and use exactly 512 GB. Of course in reality file sizes are probably a rather random size so you'll have two long floating point numbers. At least with base 2 you just have one, because you have exactly 512 GiB RAM.
And when you do have base 2 numbers then multiplication/division gives other nice base 2 numbers like 10 MiB / 2 KiB = 5 KiB. 10.485760 MB / 2.048 KB = how much? It's a lot uglier if you numbers are 2^n values, which again they will be a lot of the time. At least far more often than base 10 as long as you're working with the computer itself and not business data or whatever. If you for example want to make something fit in L3 cache to optimize and algorithm, the numbers will be in base 2. You can't "bugfix" your way out of that.