RAM speed and latency, and when they matter

A memory kit is sold on two numbers — a speed and a latency — and neither tells you much alone. The speed is how much data the memory can move; the latency is how long you wait for the first of it. Together they set how quickly memory answers, and reading one without the other is how a kit that looks faster turns out not to be.

What the speed number measures

The headline figure, in MT/s — megatransfers per second — is the rate the memory moves data once it is answering: its bandwidth. Higher means more data per second across the bus, and it is the number that rises each generation and between grades within one. DDR4 and DDR5 sit in different bands of it, DDR5 starting roughly where DDR4's fast kits leave off and climbing well past them.

Bandwidth is what you have plenty of for most work and short of only in particular places. Moving large blocks — and feeding a graphics processor that borrows system memory for its own — leans on it. A single program working through modest amounts of data at a time rarely reaches the ceiling, which is why the speed number moves everyday responsiveness less than its size suggests.

What CAS latency measures, and why it is only half

CAS latency — the CL on a kit's label — is how many memory cycles pass between asking for data and the first of it arriving. Lower looks better, and within one speed it is. The catch is that a cycle is not a fixed length: it gets shorter as the speed rises, so a higher CL at a higher speed can mean the same wait in real time as a lower CL at a lower speed.

The wait that matters is in nanoseconds, and it is the two numbers together: CL divided by the speed. Run that on two kits before deciding one is quicker, because a fast kit with loose timings and a slower kit with tight ones often land within a hair of each other on the figure that actually counts. The label shows CL because it is one number; the real latency is the arithmetic behind it.

When speed and latency actually matter

For most machines, moving from a mid kit to a fast one changes little you would feel — the work is not waiting on memory. Three cases are the exceptions. Integrated graphics, which have no memory of their own and run on the system's, gain directly from more bandwidth. AMD's Ryzen chips tie part of their internal fabric to memory speed, so faster memory lifts more than the memory alone. And a handful of games that are held up by the processor rather than the graphics card respond to lower real latency. Outside those, memory speed is a small lever, and the money it costs often does more elsewhere in the build — including in capacity, which moves far more for far more people than speed does.

DDR4 or DDR5

This one is not a tuning choice but a compatibility one. The two generations are physically keyed apart and a motherboard is built for a single one, so the board decides which you buy — there is no fitting DDR5 into a DDR4 board or the reverse. DDR5 brings more bandwidth and adds error correction on the module itself, and early in its life it ran at a higher real latency than mature DDR4 before catching up. When you are choosing a platform the question is which generation the board takes; once the board is chosen, the generation is settled and only the grade within it is left to pick.

What we compare on

We compare memory on three figures: capacity, speed in MT/s, and CAS latency. Higher is better on the first two and lower on the last, and that is the whole of what the comparison knows. The full timing tuple beyond CL, the voltage, and whether a kit carries error correction are fields a kit may state, but most listings give only the headline pair, so they are not something the comparison leans on. That absence is the market's habit, not a judgement that the rest does not matter.

Every RAM kit in our catalog is on one page, with specs from the datasheet and prices tracked over time.