What is a reliable nodes count?

[solar pier]

In the last question I asked, (which was about the energy consumption of Stockfish 15.1 and Stockfish 16), I was told that nodes are by analyzing more important than depth. So, my question is to what number of nodes (in total) is usually analyzed, if one wants to have moves, which are considered good? I usually analyzed up to depth 30+-ish with Stockfish but I don't know how that would translate to nodes.

[fallow anvil]

Probably around 10-20M nodes should be depth 30? You need exponentially more nodes for each depth.
If you want a very deep analysis of a single position and have a very good CPU and a lot of RAM then billions or hundreds of billions of nodes.
If you have a mid tier CPU and want to quickily analyze an entire game then 5-10M. For reference, Lichess uses 1.5M.
You have to use as much as you can/want. Depends on your own needs.

[radiant edge]

i can answer exactly how it can translate to nodes
sf 15.1 takes 9,416,686 nodes in ~6.3 seconds on my machine
sf 16 takes 11,373,475 nodes in ~8.6 seconds on my machine
bmi2 build for both

unsure why sf 16 takes more nodes, but then again I only started following sf's changes recently ¯_(ツ)_/¯

[fallow anvil]

how many positions did u use to get that number?

[radiant edge]

from start pos, ran twice for both
re-reading his question, it seems he didn't mean start pos

[fallow anvil]

and a single position is not really a big sample size

[radiant edge]

yeah
i could test a few hundred or thousand positions at a lower depth, but it would seem a bit pointless since we know which version is stronger

[solar pier]

I meant positions generally, not the start pos

[radiant edge]

well, what do you mean by 'good'?

[solar pier]

I'm rather trying to use it to analyze single moves individually rather than games itself

[solar pier]

What do you mean?

[radiant edge]

if one wants to have moves which are considered good

[solar pier]

That's a rather hard question to answer. I would say moves that can be calculated without investing too much resources

[radiant edge]

1 node then ;)

but honestly i would test it yourself

[solar pier]

Also an option 🙂

[fallow anvil]

what CPU do you have?

[radiant edge]

sf has a really low branching factor so even depth 18 would take minimal resources

[solar pier]

I'm not really an expert in this, what would you consider a mid CPU?

[solar pier]

Where did you test your Stockfish?

[fallow anvil]

you can just say which one do u have

[radiant edge]

make -j profile-build
./stockfish [uci options]

although this probably a terrible way to do it - i haven't looked up how to quickly test from a file of many different positions :p

[solar pier]

Is the one with the Ghz important?

[radiant edge]

what do you mean?
clock speed is important, yes, but no. of threads, cache size, etc. are also important

[fallow anvil]

I'm just asking you the name of the CPU

[solar pier]

core

intel

[fallow anvil]

Okay... that is a little bit more specific but not much

The name is something similar to this: Intel Core i5 11600K

[solar pier]

As I said, I'm not an expert with this I'm rather here for the analysis with Stockfish and its details, so excuse me if I don't know everything

[fallow anvil]

The CPU that you have is important for that. If you have a fast CPU, you can afford to analyze with more nodes.

A laptop CPU might only analyze at 1-5M nodes per second
An expensive desktop CPU can reach 50M per second

[solar pier]

I'm just gonna test it quickly

[radiant edge]

just look at the cpu info in the 'system info' (or equivalent) part of settings

[fallow anvil]

or task manager in windows

[solar pier]

its roughly 1m (its moving up and down)

[fallow anvil]

are you using all of your cores?

[radiant edge]

seems decent for a single core, but knowing the exact model would be more precise

[solar pier]

using all of them should be 100% usage (in the task manager) right?

[radiant edge]

ye

assuming it's all cpu threads, not physical cores

[solar pier]

When I used it now task manager said about 40-50% usage of the CPU

[radiant edge]

what cpu model does task manager report

[solar pier]

6200U CPU(?) If that's what you are asking for

[radiant edge]

so something like 'intel core i7-6200'?

actually i5 or i7?

[solar pier]

yes

i5

[radiant edge]

kk

huh, pretty old processor

[solar pier]

yeah not anything modern

kn is an abbreviation for nodes times 1000 (kilo-nodes ) or?

[radiant edge]

ok so back to the original question: it depends how much time you want to give the cpu. Even 1/10 of a second on any given position should give some move close to the best move

[radiant edge]

yes

kn = 1000 nodes
knps = 1000 nodes/second

[solar pier]

I'm using the engine mostly for the purpose of opening preparation with the help off chessdbcn

[radiant edge]

in that case, why do resources matter that much?

[solar pier]

Do you mean the nodes?

[radiant edge]

sorry, i meant resource consumption

unless you're concerned with battery power
actually do you use a laptop or pc?

[solar pier]

First of all I don't want it to be too warm and saving energy

[solar pier]

laptop

[radiant edge]

okay
i'm not sure why saving resources should be a big concern then
if you need to save energy, a laptop isn't a good place to start, since even the maximum power they can draw is still peanuts compared to other appliances
if you need to preserve battery power, frankly I would analyse on a low depth to save your time more than your battery
analysing any given position to, say, depth 20 would take less than a second and you wouldn't analyse any position very frequently if you're just preparing for openings
windows itself would take more of your computer's resources than stockfish would over a longer time period if you're only analysing say ~1-2 positions per minute
anyhow i have supper now so talk to you later

[solar pier]

Yeah I know that a laptop is rather weka when considered to other computers, but this what is available for me.