what is the maximum "Nodes per Second" at single core ?

[ghost]

#4166 (comment)

info depth 70 seldepth 88 multipv 1 score cp 93 wdl 169 829 2 nodes 10328639573768 nps 46749714 hashfull 999

Based on one of the issues sent by a friend , I see 46 Million NPS speed for stockfish , do this speed is really ?

what is the highest speed of NPS at single core ?

Thanks

[dav1312]

1. That is not a single core test
2. There is no hardcoded limit

[dav1312]

You should expect around 1-2Mnps

[ghost]

You should expect around 1-2Mnps

yes I agree with Intel Core i9-12900KS at single core NPS is about 2 Million

[RogerThiede]

https://ipmanchess.yolasite.com/amd---intel-chess-bench.php

[ghost]

https://ipmanchess.yolasite.com/amd---intel-chess-bench.php

thanks @RogerThiede , i see this link but test was multi thread , may i ask you what is maximum NPS at signle thread ?

[RogerThiede]

I suspect that would depend on too many variables, most notably being that the complexity of a given FEN position can cause significant slowdowns, let alone the size of the transposition table hash and whether or not it is full due to cache misses and/or extraneous memory fetches.
Using a pure neural network evaluation more similar to Leela with no strategic search algorithm of moves would likely result in the highest nodes per second possible; however that could take longer to identify the most optimal lines even though it reaches higher NPS. Stockfish spends additional resources on the move picker and other strategies.
You can rent various hardware architectures (BMI2, AVX2, AVX512, VNNI512, etc,) from cloud companies and see different results just based on different compilers (gcc, clang, etc.)