clear shore
I want to run a benchmark on stockfish to test my engine against it, but when i run bench 1 1 5 current pertf as I saw in a youtube video, it shows Nodes searched : 0 (in the vid it showed the numbers correctly) could you please help me? I have 12th Gen Intel(R) Core(TM) i5-12500H if it's relevant.
stark summit
Perft count is correctly reported just above
319....
But this does show a weird output artifact that should be solved
clear shore
sorry i forgot to mention, I want to see the nodes per second
stark summit
You have the total nodes and the total seconds
clear shore
oh :))))))
stark summit
You can probably extract the ratio if you think hard enough about it
(you just need a division)
clear shore
thx i didnt think about that
stark summit
Np
Since you are presumably working on move generation , if you ever need to debug perft I suggest using webperft over SF
For any general engine dev question feel free to ask in #engines-dev
clear shore
no, I already debugged my move generation, right now I just want to see how fast it is
stark summit
Well if it's just purely to measure speed then sf is a good target
Are you also doing bulk counting?
Because sf does
clear shore
For any general engine dev question feel free to ask in <#882956631514689597>
thx, I might ask around if I stumble upon anything
stark summit
So that inflates the final result by an order of magnitude
clear shore
Are you also doing bulk counting?
no, I want per position
stark summit
That's a wrong answer
clear shore
I saw if I use the default option it bulk counts them
clear shore
yes
gleaming stream
I think this is a regression
stark summit
Bulk counting is a technique you can do if you have fully legal movegen
It lets you skip the last layer of make moves, it increases perft speed (but not actual search speed) by a factor of 10 or more
If you aren't doing it SF perft performance will appear very inflated compared to what you can get
clear shore
ill search around, thanks for the insight
stark summit
Yw
hot comet
I think this is a regression
yeah, sf16 works fine but 16.1 doesn't