Depth, and What It Buys You
Depth 20 is not twice as good as depth 10. What a ply is, why the returns fall away sharply, and how to tell when a position has actually been resolved.
· 3 min read
Depth is the number of plies the engine has searched. A ply is one move by one side, so depth 12 means the engine has looked six moves ahead for each player along the lines it considers most likely.
That last clause carries most of the weight. Modern engines do not search every branch to the stated depth. They search promising lines much further and cut off unpromising ones early, so "depth 20" describes the main line, not a uniform wall of twenty plies. Two engines reporting the same depth may have searched very different amounts.
The returns fall away fast
Each additional ply multiplies the work. Alpha-beta pruning and the various heuristics on top of it bring the branching factor down to somewhere around two or three in practice, but that is still exponential: going from depth 20 to depth 21 costs about as much as everything before it.
What you get for that cost shrinks at the same time. Going from depth 6 to depth 12 will routinely change the engine's mind about which move to play. Going from depth 24 to depth 30 changes the evaluation of a quiet position by a few hundredths of a pawn and almost never changes the move.
For practical use this means the interesting range is roughly 12 to 20. Below 12 you will see tactical oversights. Above 20 you are usually paying for precision you cannot use.
Some positions need much more
The average is not the point, though. Depth requirements vary enormously by position type, and knowing which kind you are looking at is more useful than knowing a number.
Quiet middlegames settle early. A closed position with no captures available is often stable by depth 12, and searching to 30 confirms what depth 12 already said.
Sharp tactical positions do not. A position with a piece hanging, several checks available and open lines can change its mind repeatedly into the twenties. Each new depth finds a resource the previous one missed.
Endgames are the worst case. Long forced sequences are exactly what a depth-limited search handles badly. A king and pawn endgame whose result hinges on a fifteen-move race needs thirty-odd plies before the engine sees the outcome, and until then it reports a number that is confidently wrong. Tablebases exist because of this.
| Position type | Usually settled by | Why |
|---|---|---|
| Closed middlegame, no captures | depth 12–14 | Few forcing lines to resolve |
| Open middlegame, pieces active | depth 16–20 | Tactics appear and disappear as the search extends |
| Sharp position, material imbalance | depth 22+ | Each ply can reverse the assessment |
| Pawn endgame | depth 30+, or a tablebase | The result hinges on a long forced race |
| Fortress or blockade | never reliably | No search depth reveals that nothing progresses |
How to tell when it is settled
Search the same position at increasing depth and watch the evaluation rather than reading it once.
If the number is stable across the last several depths, the engine has resolved the position and further searching will not change your decision. If it is still moving by half a pawn or more between consecutive depths, it has not, and the current answer is a snapshot of an unfinished search.
This is worth doing on any position you actually care about. It costs a few extra seconds and it distinguishes an answer from a guess, which is not something the number alone will ever tell you.
Depth and time are not the same
An engine given a fixed time will reach whatever depth it can. An engine given a fixed depth takes whatever time it needs. Both are legitimate, and mixing them is where confusion starts: a request for "depth 20, stop after one second" gets whatever the search reached when the second elapsed, which in a complicated position might be depth 14.
That is exactly what happens here. The analysis on this site asks for a depth and also caps the thinking time, which is why the pages say a search reached about a given depth. In a quiet position it gets there comfortably. In a sharp one, time runs out first, and honest reporting means saying so rather than printing the number that was requested.
Common questions
- What depth is good enough for chess analysis?
- For reviewing your own games, depth 15 to 18 settles almost every position that matters. Sharp tactical positions and endgames benefit from more. Beyond about depth 25 the evaluation of a normal middlegame position rarely changes enough to alter which move you would play.
- Does higher depth mean the engine is stronger?
- It means it has searched further, which usually means better. But a full ply of extra depth costs roughly three to five times the work of the one before it, so the improvement per second of thinking falls away quickly. Doubling depth does not double strength.
- Why does the evaluation change between depths?
- Because the engine found something it could not see before. A jump of more than about half a pawn between consecutive depths means the position is unresolved, and the current number should be treated as provisional rather than as an answer.
Try it on your own position
The analysis board runs the same engine in your browser, free. The app adds what a web page cannot: reading a position straight from a photograph, saved boards, and analysis with no connection.
