sleep_ms poor accuracy

[kevinjwalters]

I've noticed sleep_ms() has poor accuracy in MicroPython 2.1.2 on a micro:bit V2. A bit of REPLing shows sleep_ms vs sleep_us below. It's ranging from 11364us to 12916us in that quick bit of testing whereas sleep_us is very consistent at 12076 to 12084.

The port specific documentation describes sleep_ms as

Delay for given number of milliseconds, should be positive or 0.

The general docs have more detail:

Delay for given number of milliseconds, should be positive or 0.

This function will delay for at least the given number of milliseconds, but may take longer than that if other processing must take place, for example interrupt handlers or other threads. Passing in 0 for ms will still allow this other processing to occur. Use sleep_us() for more precise delays.

Depending on how one interprets "at least" wrt granularity at 11.364ms for a 12ms delay sleep_ms is not honouring that.

MicroPython v1.18 on 2023-10-30; micro:bit v2.1.2 with nRF52833
Type "help()" for more information.
>>> 
>>> from utime import ticks_us, ticks_diff, sleep_ms, sleep_us
>>> t1 = ticks_us() ; t2 = ticks_us()

>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12078
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12084
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12077
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12076

>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11769
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11405
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11364
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12016
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12916
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12455
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11701
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11782
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11746
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11212
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11786
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12311
>>> 

[jonnor]

Does this happen also on a recent MicroPython version, like 1.25.0 ?

Does this cause practical problems that you are looking to fix? Or is this more of a FYI for other people?

[kevinjwalters]

I was using the current version of MicroPython for micro:bit. I was seeking to raise awareness and to canvass opinions on whether behaviour was acceptable and whether the documentation should be amended.

I now have a special build of v1.23.0 for another bug investigation. The problem still exists.

MicroPython v1.23.0 on 2025-04-30; micro:bit v2.1.1 with nRF52833

Type "help()" for more information.

>>> from utime import ticks_us, ticks_diff, sleep_ms, sleep_us
>>> t1 = ticks_us() ; t2 = ticks_us()

>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12078
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12077
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12079
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12078
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12089
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12075
>>> t1 = ticks_us() ; sleep_us(12_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12077
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1076
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1078
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1074
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1075
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1076
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1075
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1079
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1075
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1088
>>> t1 = ticks_us() ; sleep_us(1_000) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1076

>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11197
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11153
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
12033
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11952
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11965
>>> t1 = ticks_us() ; sleep_ms(12) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
11170
>>> t1 = ticks_us() ; sleep_ms(1) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
883
>>> t1 = ticks_us() ; sleep_ms(1) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1142
>>> t1 = ticks_us() ; sleep_ms(1) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1771
>>> t1 = ticks_us() ; sleep_ms(1) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1484
>>> t1 = ticks_us() ; sleep_ms(1) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
1440
>>> t1 = ticks_us() ; sleep_ms(1) ; t2 = ticks_us() ; print(ticks_diff(t2, t1))
291

[davefes]

Tried your script on a STM32H723VGT6 dev board and got similar variations at 1ms. Thought I try an ESP32 and only see a +/-10us variation. I expected the ESP32 to be worse!

Do some ports use the RTC and others a system clock?

P.S. Tried a WeACT BlackPill and got 1024 +/-1us