What is Depth in Chess? Different Depths for Stockfish and LCZero
When chess players start analyzing with engines, depth is usually the first number they look at. For many of them, it is the only indicator of whether they should trust the suggested moves. And although this approach is inaccurate since several other equally important factors should be considered, depth is indeed an essential part of engine analysis. So a better understanding of its role will help you use your engines more effectively and set the right expectations.
What is chess engine depth?
Depth is a value of chess engine analysis that indicates the number of half moves (a move made by one side) the engine looks ahead. A higher depth value usually means better analysis results, but several other factors also go into play.
For example, +1.3 with depth 50 means that after calculating 50 half moves (also referred to as ply/plies), the engine evaluates the given position as significantly better for white. However, depth 50 does not mean that the engine has checked 50 plies ahead for all possible variations. In order to reach such high depths, engines use a so-called pruning method, during which they cut off a part of variations at lower depths after evaluating them as ineffective. This allows the engines to focus on more promising variations and calculate them deeper.
What is a good depth for chess engines?
Basically, the higher the depth, the better. And the longer a chess engine analyzes the same position, the higher depth it reaches. However, most of the depth is reached in the first minute of engine analysis. Afterward, it starts increasing very slowly: usually by a single point every few minutes.
Some players believe that depth is the most important value during engine analysis, so they readily leave their engines running for several hours just to get higher depths. But although depth is important, it greatly depends on several factors, such as the NPS speed and the computer on which the engine is running.
Depth 25 on 1 CPU is not the same as depth 25 reached on 8 CPUs. More CPU cores allow chess engines to look into more variations, which helps them increase the credibility of their analysis but slows down their ability to reach higher depths. So the answer to what is considered a good depth can be given only for specific cases.
The main factors that impact the depth of an engine and the time it takes to reach that depth (known as time-to-depth speedup) include:
- The strength of the computer on which the engine is running
- The engine itself (Stockfish, LCZero, Houdini, Komodo, etc.) - see the top chess engines for 2023
- The type of chess position – in openings, for example, where many good moves are available, the depth increases slower
- The number of requested lines, etc.
Stockfish depth vs LCZero depth
Stockfish and LCZero are the most popular engines among chess players, but the two engines work very differently. They even require different processors: while Stockfish runs on CPUs, LCZero works best on GPU cores.
If you leave the two engines to work for the same amount of time, you may get the same move suggestions but completely different depth and NPS values. Let’s take a look at this screenshot of 300,000 kN/s speed Stockfish and 100 kN/s LCZero (lastest-T60 network) after 1 minute of analysis.
Although the suggested lines and evaluation are similar, depth and speed are disparate. Stockfish analyzes around 1000X more positions per second than LCZero, but the latter has a better evaluation method. The difference is in their algorithms.
Stockfish takes a wide tree of possibilities and gives them a very basic evaluation. It then prunes (cuts off) those moves that get a bad score, and continues to look deeper into the remaining moves. The process continues until Stockfish manages to reduce down to a few best options.
In contrast, Leela Chess Zero looks into fewer possibilities but uses a much more advanced evaluation method than Stockfish. Due to the complexity of its evaluation, Lc0 works slower and requires more time to reach the same depth as Stockfish. However, lower depth or speed does not affect LCZero's efficiency of analysis. In fact, if you compare 1 depth of Lc0 to that of Stockfish, then Lc0 will always win.
So what is an acceptable depth range for each of them? For somewhat accurate analysis and given that both engines run on adequate hardware, you would not want anything below 20 with Stockfish, while LCZero can offer quite credible lines and evaluation even at depth 10.
Helpful links for further discussion
If you want to dive into the technicalities of chess engine analysis deeper, then these are some links that will be helpful to you.