PC Engines apu2 experiences

Qinn

Thanks for the update, I will install it asap.

kevindd992002

@dugeem said in PC Engines apu2 experiences:

SMBIOS memory capability types 16 & 17

Where does one use "SMBIOS memory capability types 16 & 17"?

dugeem

@kevindd992002 For informational purposes only with dmidecode. Arguably this should have already been in the firmware but since it took time for 3mdeb to get ECC working (due to lack of AMD documentation) it probably hasn’t mattered.

cwagz

How does the APU2C4 with the new BIOS compare to a Fitlet2 J3455? Does anyone know the boost clock that the APU will get up to now? Forgive me if I missed this info earlier in the thread.

dugeem

@cwagz

With AMD CPB enabled in latest APU2 BIOS a lightly loaded (i.e. no more than single core active) APU2 CPU will boost from normal 1.0Ghz up to 1.4GHz. In the real world the gain is less - around 20% (see 3mdeb tech notes https://3mdeb.com/firmware/amd-cpu-boost/).

The quad core Celeron J3455 CPU has a base frequency of 1.5GHz and boosts to 2.3GHz - this would suggest it should offer ~50% higher performance than the AMD GX-412TC. Benchmarking would be required to confirm this.

However before selecting router hardware there are many other considerations apart from just CPU performance. EG. bandwidth, budget, form factor, interfaces, storage etc

fireodo

Hi, I made the update on my APU2C0 board to coreboot legacy ver. 4.0.25 and after 2 days everything works fine - the only thing that I can see is that the CPU temperature is slightly lower (4-5°) than before the update (same environment conditions). Maybe it is in context of AMD CPB activation.

Have a fine Weekend,
fireodo

Veldkornet

@fireodo said in PC Engines apu2 experiences:

Hi, I made the update on my APU2C0 board to coreboot legacy ver. 4.0.25 and after 2 days everything works fine - the only thing that I can see is that the CPU temperature is slightly lower (4-5°) than before the update (same environment conditions). Maybe it is in context of AMD CPB activation.

Have a fine Weekend,
fireodo

Yes, the CPU can now enter C6 idle-state as well when needed which is a deeper power saving state. More power saving = less heat.

Qinn

Is it useful to enable PowerD now?

wgentine

@qinn No. CPB is not describe in ACPI so, powerd doesn't know.

Qinn

Thanks

Veldkornet

Does anyone else use the gwled package for the other 2 lights on the front?

I noticed that it’s using massive amounts of CPU, pushing the load average well above 20. Even disabling doesn’t fix it, only removing the package helps. Below is only a snippet of "px auxw", there are still many more processes of the same.

USER      PID  %CPU %MEM    VSZ    RSS TT  STAT STARTED        TIME COMMAND
root       11 156.5  0.0      0     64  -  RNL  Wed17   16794:28.46 [idle]
root    55074   4.6  0.9  96292  35144  -  R    12:28       0:00.59 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    57431   4.5  0.9  96292  35144  -  R    12:28       0:00.60 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    58665   4.5  0.9  96292  35144  -  R    12:28       0:00.58 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    56863   4.4  0.8  96420  34636  -  R    12:28       0:00.56 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    58037   4.4  0.9  96292  35144  -  R    12:28       0:00.59 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    51369   4.3  0.9  96304  35948  -  R    12:28       0:00.65 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    52694   4.3  0.9  96292  35144  -  R    12:28       0:00.60 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    53220   4.3  0.9  96292  35144  -  R    12:28       0:00.59 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    56211   4.3  0.9  96292  35144  -  R    12:28       0:00.58 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    57826   4.3  0.9  98468  35212  -  R    12:28       0:00.56 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    58322   4.3  0.9  96292  35144  -  R    12:28       0:00.58 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    58337   4.3  0.9  98468  35212  -  R    12:28       0:00.56 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    51633   4.1  0.8  96420  34584  -  R    12:28       0:00.56 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    52119   4.1  0.9  96420  35268  -  R    12:28       0:00.57 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    55153   4.1  0.9  98468  35212  -  R    12:28       0:00.55 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    56521   4.1  0.8  98468  35036  -  R    12:28       0:00.54 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    58153   4.1  0.9  98468  35212  -  R    12:28       0:00.54 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    39184   4.0  7.5 339404 308808  -  Ss   22:56      48:49.73 /usr/local/bin/suricata -i igb1 -D -c /usr/local/etc/suricata/s
root    51525   4.0  0.9  96292  35160  -  R    12:28       0:00.58 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php
root    57568   4.0  0.9  98468  35200  -  R    12:28       0:00.54 /usr/local/bin/php-cgi -q /usr/local/bin/gwled.php

See here the difference in CPU after removing the package:

stephenw10

Hmm, odd. Actual CPU time on those processes is all pretty small though. You see it generating those everytime at boot?

Steve

cwagz

@wgentine said in PC Engines apu2 experiences:

@qinn No. CPB is not describe in ACPI so, powerd doesn't know.

Is it a best practice to disable powerd then?

stephenw10

Powerd doesn't do anything on the APU anyway, there's no driver for cpufreq to use.

Mar 17 17:48:47 	php-fpm 	92195 	/system_advanced_misc.php: The command '/usr/sbin/powerd -b 'hadp' -a 'hadp' -n 'hadp'' returned exit code '69', the output was 'powerd: no cpufreq(4) support -- aborting: No such file or directory'

Steve

Veldkornet

@stephenw10 said in PC Engines apu2 experiences:

Hmm, odd. Actual CPU time on those processes is all pretty small though. You see it generating those everytime at boot?

Steve

Hmm, I’ll have to check what it is at boot. I haven’t put much time into investigating it. Looking at the SNMP though, this is the first time the CPU has dropped down (after uninstalling the package). So, I guess it starts up pretty quick after boot since I don’t even see a dent in the SNMP.

Just watching “top”, it seems to come in bursts, like every 10 seconds for 5 seconds long for example.
A wild guess is that that’s how often it polls the gateways.

I just re-installed pfSense from scratch, not restoring any backups, and it still does it.

Since there’s not much to configure, I assume anyone could reproduce it. If not, I’m really curious as to what’s so special about my system.

You can basically choose what the last 2 lights are, which I have as:
2 - WAN_DHCP4
3 - VPN

I've made a new topic for this here

dugeem

@wgentine True - although as soon as powerd sets CPU frequency to maximum GX-412TC frequency 1000MHz via ACPI then AMD CPB is available - however whether CPB it is used depends on CPU core state & thermal parameters.

Although the question of whether powerd is still useful for APU2 given both the CPB performance gains and power savings is interesting...

Edit: some excellent CPB info here https://github.com/pcengines/apu2-documentation/blob/master/docs/debug/cpu_frequency.md

dugeem

@stephenw10 You mean the original APU (with AMD G-T40E CPU) - not the APU2 (with AMD GX-412TC CPU) right? No issues with powerd on APU2 (although whether it makes any sense now with CPB is the issue)

stephenw10

@dugeem Ah, true! That was referring to the original APU, my mistake.

Veldkornet

Is anyone using the CoDel / FQ_CoDel Traffic Shaping on the APU2?

Working well? Any problems?

cwagz

@Veldkornet said in PC Engines apu2 experiences:

Is anyone using the CoDel / FQ_CoDel Traffic Shaping on the APU2?

Working well? Any problems?

I have an APU2 box at work to provide a separate network for personal devices. It is setup with the FQ_CoDel limiter / floating rules method described towards the end of the Playing with FQ-CoDel Thread. It has been rock solid and seems to provide equal bandwidth sharing for the 30 - 50 devices connected each day and 16 - 20 GB of traffic that is passed on our 150/150 FiOS link.