Skip to content

Instantly share code, notes, and snippets.

@wiedehopf
Last active November 23, 2024 11:50
Show Gist options
  • Save wiedehopf/66286a0d0ca99778ceb6a9ac80ea94a2 to your computer and use it in GitHub Desktop.
Save wiedehopf/66286a0d0ca99778ceb6a9ac80ea94a2 to your computer and use it in GitHub Desktop.
N100 PL1 PL2 power limit testing
If the aim is to run some Intel processors as efficient as possible, changing the power limit won't necessarily get it into the efficient frequency for single core loads.
For that it's necessary to set a maximum turbo frequency, this can be done like this:
for freq in /sys/devices/system/cpu/cpufreq/*/scaling_max_freq; do echo 3000000 > $freq; cat $freq; done
The default for N100 is 3.4 GHz which it will only boost to for a single core.
The following testing was done at PL1/PL2 at 15W. (i wouldn't recommend lowering PL1/PL2 below 14 W for efficiency unless you have issues with cooling)
Single core wall power draw for `stress-ng -c 1` stressing a single core:
max_freq pWallSingle geeks geekm geekl
2.0 7.8
2.2 8.2 839 2495 https://browser.geekbench.com/v6/cpu/9022108
2.4 8.9
2.6 9.5 983 2674 https://browser.geekbench.com/v6/cpu/9021962
2.8 10.2
3.0 11.0 1124 2686 https://browser.geekbench.com/v6/cpu/9021838
3.2 12.1
3.4 13.5 1224 2662 https://browser.geekbench.com/v6/cpu/9021556
I'm quite ok if it's not the fastest single thread performer, thus i'll just use 3.0 GHz max freq.
Those 400 MHz less save 1.5 W during single threaded workloads.
For multi-threaded stuff the power limit described below is the relevant figure.
15 W is a pretty reasonable number to run really and the default for most N100 mini pcs.
But it's also reasonable to run 16 or 17 W to get some more performance.
For a permanent solution you could use a crontab:
run: crontab -e
insert this line:
@reboot /usr/bin/bash -c 'for freq in /sys/devices/system/cpu/cpufreq/*/scaling_max_freq; do echo 3000000 > $freq; cat $freq; done'
PL1 and PL2 are just set to the same number for simplicity.
pWallSF is the power draw from mains.
geeks / geekm are the geekbench6 single / multicore scores
PL1/2 pWallSF StockfishScores geeks geekm geekl
10 11.5 1604841,1582075,1788214 967 1842 https://browser.geekbench.com/v6/cpu/9003729
11.625 13.2 1924550,1925746,1893007 1086 2180 https://browser.geekbench.com/v6/cpu/9005510
13.375 15.0 2021293,2090408,2105400 1175 2480 https://browser.geekbench.com/v6/cpu/9005710
15 16.5 2309462,2327378,2193946 1229 2663 https://browser.geekbench.com/v6/cpu/9007153
16.625 18.3 2757442,2408745,2382714 1252 2830 https://browser.geekbench.com/v6/cpu/9008143
18.375 20.0 2985179,2463419,2493028 1251 2947 https://browser.geekbench.com/v6/cpu/9014086
20 21.7 2582110,2876978,2788260 1251 3044 https://browser.geekbench.com/v6/cpu/9014752
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment