Re: RPI4 clock speeds and serial port ( temperatures idle and -j4 buildworld buildkernel )

On 2021-Mar-20, at 12:27, Mark Millard <marklmi at yahoo.com> wrote:

> On 2021-Mar-20, at 11:28, bob prohaska <fbsd at www.zefox.net> wrote:
> 
>> On Sat, Mar 20, 2021 at 10:41:48AM -0700, Mark Millard wrote:
>>> On 2021-Mar-20, at 08:56, bob prohaska <fbsd atwww.zefox.net> wrote:
>>> 
>>>> The most recent test on the 8GB Pi4 finished in 9 hours. That was after
>>>> running make cleandir twice, but having the installed system and the
>>>> sources relatively close (days) in age, on -current.
>>> 
>>> So a debug kernel was running? (Non-debug running for my builds.)
>> 
>> Yes.
>> 
>>> 
>>> Did it include a buildkernel? Or just a buildworld?
>>> 
>> 
>> Just buildworld in 9 hours. 
>> 
>>> At around 9 hours, it would take standardizing on a (equivalent
>>> of) src.conf too be able to well compare your context vs. mine
>>> now, if we ever want to do so. I'll list my file later below
>>> for reference, although I'm probably more likely to try your
>>> src.conf content in my overclocked context than you are
>>> likely to try building via my src.conf like content in your
>>> powerd context.
>>> 
>> 
>> No /etc/src.conf for this test, nor /etc/make.conf. 
> 
> I have a -j4 buildworld build kernel going on the
> FreeBSD RPi4B 8GiByte, based on using a prefix of:
> 
> env __MAKE_CONF="/dev/null" \
> SRCCONF="/dev/null" SRC_ENV_CONF="/dev/null"
> 
> I'll report the buildworld and buildkernel times
> once it is done.

World took somewhat longer than my normal build
and the kernel somewhat less, but world+kernel
somewhat longer overall:

World build completed on Sat Mar 20 20:20:42 PDT 2021
World built in 31233 seconds, ncpu: 4, make -j4
Kernel build for GENERIC completed on Sat Mar 20 20:55:21 PDT 2021
Kernel(s)  GENERIC built in 2078 seconds, ncpu: 4, make -j4

So:
World  took a little under 8 hr 45 minutes (vs. under 8 hr 10 min)
kernel took a little under      35 minues  (vs. under      40 min)
Total:      a little under 9 hr 20 minutes (vs. under 8 hr 50 min)

You are getting the vast majority of the time
benefit from powerd use compared to my build
context (based on the world build times). Avoiding
building to allow compiling non-arm could cut
your time noticeably, possibly around 30 minutes
for world.

But I do not come anywhere near what tech-lists
has reported: a little under 6 hr 45 min for
world+kernel (using -j6 and possibly ccache use,
however). My guess is that ccache avoided some
compiling/linking. (I've not used ccache.)

> The RPi4 support for powerd (cpufreq) was added
> on 2020-Jun-20 but my RPi4B activity started
> well before then. That is why I did not go down
> the powerd path in how I configured things. It
> will be interesting to compare to your time.
> 
>>> Welcome to the world of overnight (sleep-time) buildworld on a
>>> small board computer.
>>> 
>> 
>> It's a most welcome change!
> 
> You can now also buildworld and buildkernel targeting
> armv7 in the overnight time frame via either:
> 
> A) a form of cross build (using src.conf like file(s))
> or:
> B) running an armv7 world in a chroot and doing a
>   "native" style build in the chroot.
> 
> You can also set up poudriere (or just a chroot area
> that you chroot to) in order to build armv7 ports in.
> 
> This also has less of a memory limitation in that
> multiple full-sized armv7 build processes can be
> in RAM at once without swapping/paging: more total
> RAM in use than an armv7 would allow, though no
> individual process gets more than an armv7 could
> potentially allow.
> 
> You could build for the RPi3(s) or other aarch64
> machines on the RPi4(s) as well. So, likely your
> arm build activity no longer has to take up nearly
> as much time, especially the really long-wait
> builds.
> 
>>>> That's with powerd
>>>> enabled, but no other speedup tweaks. Temps were just under  60 C, with
>>>> a fan in the case but no heatsinks on the ICs. 
>>> 
>>> Ambient temperature? (It is the difference that indicates how
>>> well your cooling is working and it is the difference that
>>> is more comparable across contexts. For example, my around
>>> 15.5C ambient likely was likely around 10C lower than for
>>> many reported figures.)
>>> 
>> 
>> Ambient was around 15-17 C at the times of observation. 
>> 
>>>>> sysctl dev.cpu.0.freq will definitely show the current freq.
>>>> 
>>>> On a Pi3 with GENERIC-MMCCAM kernel  (also running buildworld) it reports
>>>> bob@www:/usr/freebsd-src % sysctl dev.cpu.0.freq
>>>> dev.cpu.0.freq: 1200
>>>> also with no other speedup efforts beyond powerd. That machine has heastinks
>>>> on the ICs but only a fan blowing in through the openings in the plastic case.
>>>> It's around 60C. 
>>>> 
>>>> So far the effects of powerd seems to be uniformly good, with 
>>>> no downsides. Next, to try a Pi2 v1.1 (armv7)
>>>> 
>> 
>> The Pi2 V1.1 is now running buildworld with powerd enabled. Heatsinks
>> on the ICs, but no fan at all, < 50 C (17C ambient) with 0% idle, 900 
>> MHz, serial console seems to work fine. 
>> 
>>> 
>>> FYI (not that you are likely to build and run such:
>>> 
>> [snipped for brevity]
>> 
>> Alas, you are right. It's already difficult to keep the 
>> experiments straight in my head......too many variations.
>> 
>> But, unless something most unexpected goes wrong, it seems
>> that powerd is useful on the Pi series computers.  
>> 
> 
> Yep.
> 


===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)

_______________________________________________
[email protected] mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-arm
To unsubscribe, send any mail to "[email protected]"