Memory and storage

With 32-bit addresses a processor can name 4,294,967,296 bytes, exactly 4 GiB. Switch the power off and RAM forgets all of it.

"My laptop has 16 GB of memory, so it can hold 16 GB of photos." That sentence mixes up two different parts, and the laptop advert itself shows why. It lists two sizes, for example 16 GB of memory and 512 GB of storage. Memory, or RAM, holds the programs and data the CPU is using right now. Storage, a solid-state drive (SSD) or an older hard disk, keeps your files when the power is off. The photos live on the 512 GB side. The 16 GB is the working space where a photo goes while you are editing it.

Why RAM forgets when the power goes

Most RAM stores each bit as a tiny charge that leaks away unless it is refreshed many times a second, so everything in it is lost when the power goes. That is what "volatile" means. Storage keeps data without power. An SSD traps charge in flash memory cells that hold it for years, and a hard disk records bits as magnetised spots on spinning platters. Storage is much bigger and cheaper per byte, but much slower to reach, so programs are copied from storage into RAM before they run. That copying is what is happening while an app says it is loading.

How the CPU finds one byte among billions

Memory is a long row of cells, each holding one byte, and each cell has a number called its address, starting from 0. To read or write, the CPU puts an address on a set of wires and gets back the byte stored there. The number of address bits sets how much memory can be reached. Four address bits, as in the toy computer, give 2 to the power 4 = 16 cells. Sixteen bits give 65,536. Thirty-two bits give 4,294,967,296 bytes, which is exactly 4 GiB, and that is why older 32-bit systems could not use more than 4 GiB of RAM. Today's 64-bit processors can name far more.

A cell has no idea what its bits mean. In the toy computer the 8 bits 00000100 in a data cell are the number 4, but read as an instruction the same bits mean HALT 4. Only the way the CPU reaches a cell decides which, as the next instruction or as data named by an instruction. That follows directly from von Neumann's idea of keeping both in one memory. It is also why a bug that writes to the wrong address can wreck a whole program.

Each layer down is bigger and slower

Fast memory is expensive, so computers use layers. Closest to the circuits are the registers, a few dozen values the CPU works on directly, like the accumulator in the toy. Next come caches, small blocks of fast memory on the processor chip itself, usually named L1, L2 and L3, which keep copies of recently used parts of RAM. Then RAM, then the SSD or hard disk, and beyond that the network. Each layer down is bigger and slower. A program runs fast when most of what it needs is already in the top layers. That is why reading data in order is usually quicker than jumping around.