Published 15 August 2026 · Last updated 15 August 2026
The two terms get used interchangeably and they should not be. Ghosting describes a correctness problem in the keyboard's switch matrix — certain specific combinations produce an ambiguous electrical state. Rollover describes a capacity limit — how many keys the keyboard can hold and report simultaneously, regardless of which ones they are.
A board can have excellent rollover on paper and still block a particular combination. A board can be immune to ghosting and still be capped at six keys by the connection. Different questions, different tests: the ghosting test answers the first, the n-key rollover test answers the second.
Side by side
| Ghosting | Key rollover | |
|---|---|---|
| The question it answers | Does this particular combination register correctly? | How many keys can register at once, whichever keys they are? |
| Root cause | Shared row and column wiring in the switch matrix creating an ambiguous state | How many key slots the report format and the firmware can carry |
| Hardware fix | A diode in series with each switch | Diodes plus a report format that can express more than six keys |
| Typical symptom | One combination fails while a similar one works fine | Every combination works up to a fixed count, then the next key is ignored |
| Position-dependent? | Yes — it depends where the keys sit in the matrix | No — the limit applies to any keys |
| Measured with | Ghosting test | N-key rollover test |
The quick diagnostic: if four particular keys fail but a different four work, that is a ghosting or blocking problem. If any seventh key you add is ignored no matter which one it is, that is a rollover limit.
Ghosting, briefly
Keys are wired as a grid. When three held keys form three corners of a rectangle in that grid, current can sneak around through the pressed switches and arrive exactly where the fourth corner would be, so the controller cannot distinguish "three keys down" from "four keys down". Diodes make the current one-way and remove the ambiguity.
Almost every keyboard sold today resolves this by suppressing input rather than inventing it: it drops the extra key instead of reporting a phantom. That is blocking, and it is what people usually mean when they complain about ghosting. The full ghosting explainer goes through the matrix and the diode fix in detail.
Rollover: 2KRO, 6KRO and NKRO
Rollover comes from typewriter terminology — the ability to press the next key before releasing the last one and still have both come out in order. On a modern keyboard it means how many keys the controller can hold as "currently down" and report to the computer.
| Level | What it means | Typically found on |
|---|---|---|
| 2KRO | Two ordinary keys reliably, plus modifiers. A third arbitrary key may or may not work depending on where it sits in the matrix. | Budget membrane keyboards, some laptop keyboards |
| 6KRO | Six ordinary keys plus modifiers, in any combination. This matches the standard keyboard report layout used by the USB HID boot protocol. | The great majority of keyboards, including most mechanical boards by default |
| NKRO | Every key independently tracked, with no fixed cap on how many can be held at once. | Keyboards with per-key diodes and a report format that can express more than six keys |
Some marketing describes a board as "6KRO with full anti-ghosting", which is a coherent claim: no combination will produce a phantom key, and up to six ordinary keys will be reported at once. Some describes it as "26-key rollover" or similar, which usually means the board has diodes across a defined region and is still limited by the report format elsewhere. Treat these numbers as a description of the region covered, not a hard guarantee.
Why modifiers are counted separately
Hold Ctrl, Shift and Alt together with six letters on a plain 6KRO keyboard and all nine usually register. This surprises people, but it follows directly from the report format.
The standard keyboard report used by the USB HID boot protocol carries the eight modifier keys — left and right Ctrl, Shift, Alt and Meta (Windows or Command) — as individual bits in a dedicated byte. Each modifier has its own bit, permanently reserved for it. The remaining space in the report holds a short list of ordinary key codes, and it is that list which is six entries long.
So modifiers never compete with letters for space. Six is the limit on non-modifier keys, which is why the practical ceiling on a 6KRO board is comfortably higher than six for real-world shortcuts and game bindings.
This is the reason a browser test should be read carefully: if you hold four modifiers and four letters and the page reports eight keys, that does not demonstrate NKRO. Add a fifth, sixth and seventh letter to test the real limit.
Which limit is actually stopping you?
- Count what you were holding. Include modifiers separately from ordinary keys.
- Six or fewer ordinary keys, one combination failing? Ghosting or blocking. Try shifting one key to a different part of the keyboard — if the combination starts working, the matrix layout was the problem.
- Seven or more ordinary keys? You have probably hit a rollover ceiling, most likely the six-key report limit rather than a matrix fault.
- Works on one computer but not another? Suspect the connection or the operating system rather than the keyboard: the same board can behave differently over a hub, over Bluetooth, or in a boot environment.
What browser tests cannot tell you
Both tests report what the browser received. They cannot see inside the keyboard's matrix, count its diodes, or distinguish a limit imposed by the keyboard from one imposed by the USB stack, the operating system or the browser. Treat the result as an accurate description of behaviour on this machine, not a hardware specification. The methodology page sets out exactly what is and is not observable from a web page.
Which one should you care about?
For most people, neither — until a specific combination fails. Ordinary typing never holds enough keys to hit either limit, and standard application shortcuts are built from one or two modifiers plus a single letter.
Ghosting behaviour is the one worth checking if you game with movement plus modifiers, because that pattern lands on exactly the combinations partial-diode boards struggle with. Rollover is worth checking if you use software that expects many simultaneous keys — music and rhythm software, stenography layouts, or a game with a genuinely dense keybind set. The NKRO guide covers how it is implemented and who genuinely needs it.
Neither has anything to do with speed. Rollover and ghosting describe key coverage; if what you actually want to know is how quickly input reaches the screen, run the browser keyboard latency test and read what keyboard latency actually is.
Frequently asked questions
Is anti-ghosting the same as n-key rollover?
No. Anti-ghosting usually means the keyboard will not report a key you did not press, and on many boards it is only implemented for a subset of keys. N-key rollover is a stronger claim: every key can be tracked and reported simultaneously. Full NKRO implies anti-ghosting, but a board can be advertised as anti-ghosting without offering NKRO.
Why do Ctrl, Shift and Alt not count toward the six-key limit?
The standard keyboard report used by the USB HID boot protocol carries the modifier keys in a separate bitfield, with one bit per modifier, alongside a small number of slots for regular keys. Because the modifiers have their own bits, holding Ctrl and Shift does not consume any of those regular key slots.
Do I need NKRO for gaming?
Usually not. Most games ask for three to five keys at once including modifiers, which six-key rollover handles comfortably. NKRO matters when a specific combination you actually use is blocked, or for rhythm games, music software and stenography where many keys are held together.
Can a browser test tell me my keyboard's rollover?
It tells you how many simultaneous keys your browser registered, which is the practical answer for that machine and that connection. It is not a direct reading of the keyboard hardware, because the operating system, the connection mode and the browser can each impose their own limits before the page sees anything.