Distributed Computing on Raspberry Pi

Endgame

Gawd
Joined
Jan 10, 2007
Messages
951
I figured it was time to start a Pi thread instead of hijacking someone else's thread with Pi information. Additionally, this should help people looking for information.

I'm already working on a complete set of information at the AT forums (see references at the end of this post), but I thought I would give a very quick run down here for setting up a Pi 4 to run Distributed computing.

First. why use a raspberry pi for distributed computing?
The simple answer is, a Pi is a low cost (on a per device basis), low energy use entry point into distributed computing. A single Pi 4 is about 1/3 the processing power of an old Core2 quad 9650, BUT when configured to be headless and passively cooled it will do that processing for about 3 watts of power. In other words, if you're running pre Sandy Bridge hardware somewhere with moderate or expensive priced electricity, you can probably pay for a pi in a couple of months of energy savings.

The ultra, super quick configuration guide.

1) Buy your pi (this guide assumes a 4gb pi), a 16gb flash card (I've been getting a 30GB samsung select flash card since they are cheap), a power suppy, and a heatsink case (I prefer the FLIRC pi 4 case, but many options work)
2) download and install raspbian lite on your flash card, and create an empty file called ssh in the root directory (this lets you ssh to the host without needing to first setup a monitor).
3) connect your pi to your network via ethernet (you can use wifi, but the quick shell section below turns off wifi)
4) ssh to your new raspbian lite install
5) run the following commands (loosely documented with #s)

#upgrade system
sudo apt-get update
sudo apt-get upgrade -y
sudo rpi-eeprom-update -a

#install utilities and boinc
sudo apt-get install uhubctl boinc-client boinctui zram-tools -y

#setup zram and change percentage
sudo vi /etc/default/zramswap
PERCENTAGE=100

sudo zramswap stop
sudo zramswap start

#increase swap as a last resort so we don't run out of memory
sudo dphys-swapfile swapoff
sudo vi /etc/dphys-swapfile
CONF_SWAPSIZE=1024
sudo dphys-swapfile swapon

#setup boot config and disabled unused hardware (wifi, bluetooth) to optimize power, and reduce memory assigned to video to minimum
sudo cp /boot/config.txt /boot/config.old
echo "arm_64bit=1" | sudo tee -a /boot/config.txt
echo "dtoverlay=disable-wifi" | sudo tee -a /boot/config.txt
sudo systemctl disable hciuart
echo "dtoverlay=disable-bt" | sudo tee -a /boot/config.txt
echo "gpu_mem=16" | sudo tee -a /boot/config.txt

#create power save shell script that will need to be run on every boot (sh power.sh)
echo -e '#!/bin/bash\n\n#Disable HDMI\nsudo tvservice --off\n\n#turn off power LED\necho "turning off power led"\necho 0 | sudo tee /sys/class/leds/led1/brightness\n\n#call usb script to shutdown usb\nsh ./usb.sh off' > ~/power_save.sh
echo -e '#!/bin/bash\naction=$1\nif [ $action = "on" ]; then\n val=1\n echo "setting val to $val"\nelse\n val=0\n echo "setting val to $val"\nfi\nsudo uhubctl -l 2 -a $val' > ~/usb.sh

#configure boinc - first Edit the Boinc global preferences to use more than 100% of system memory and reduce the amount of :
sudo vi /etc/boinc-client/global_prefs_override.xml

#update xml to:
<global_preferences>
<max_ncpus_pct>100.000000</max_ncpus_pct>
<vm_max_used_pct>90.000000</vm_max_used_pct>
<ram_max_used_busy_pct>150.000000</ram_max_used_busy_pct>
<ram_max_used_idle_pct>150.000000</ram_max_used_idle_pct>
<cpu_usage_limit>100.000000</cpu_usage_limit>
<work_buf_min_days>.5</work_buf_min_days>
<work_buf_additional_days>1</work_buf_additional_days>
</global_preferences>

#update cc_config to allow the arm chip.
sudo vi /etc/boinc-client/cc_config.xml

#add options after /log_flags:
<options>
<alt_platform>aarch64-unknown-linux-gnu</alt_platform>
</options>

#reboot
sudo reboot now

#after reboot use the text based boinc manager to attach to your preferred project and start running DC
boinctui

References:
My (incomplete) pi guide: https://forums.anandtech.com/threads/the-raspberry-pi-thread.2581547/
Pi 4 running Rosetta@Home: https://boinc.bakerlab.org/rosetta/forum_thread.php?id=13732
 
We like Pi(s) !

1592529522753.png
 
Last edited:
As someone new to "baking Pi's" I thought I'd post a couple of things from my perspective for others who may be interested in the platform. Their main site is here https://www.raspberrypi.org/ with information about them, getting started, downloading the main OS's and some other projects that can be done with them. I've built a couple with cameras (9.00) to use as wireless security cams using the free MotioneyeOS https://github.com/ccrisan/motioneyeos/wiki .

ATM I have two running Open Pandemics and will start messing around some more with learning about them thanks to Endgame. 👍
 
As someone new to "baking Pi's" I thought I'd post a couple of things from my perspective for others who may be interested in the platform. Their main site is here https://www.raspberrypi.org/ with information about them, getting started, downloading the main OS's and some other projects that can be done with them. I've built a couple with cameras (9.00) to use as wireless security cams using the free MotioneyeOS https://github.com/ccrisan/motioneyeos/wiki .

ATM I have two running Open Pandemics and will start messing around some more with learning about them thanks to Endgame. 👍
Are you running 4 processes continuously with Open Pandemics? I would love to get some regular updates on performance / work units if you are.

*Edit* Also, have you tried installing boinc on MotioneyeOS? My guess is the security camera doesn't use much CPU, so you could probably toss 2 WCG processes on those camera Pis.

Rosetta@Home on a stock Pi 4 4GB will generate ~1100 RAC stock, +/- 100 RAC. Overclocked to 2000mhz it will do 1300 RAC +/- 125 RAC.

I may eventually dedicate a Pi 4 to WCG, but at the moment I'm testing splitting a 2GB pi between WCG and Rosetta - it gives maximum use of the resources available to the pi without totally over committing the ram. The big question is, is it more efficient to dedicate a pi to each project? In otherwords, will my 2Ghz pi 4 2GB get 650 RAC in Rosetta or some lower amount?

Also, I wanted to suggest the cheapest possible way to try out a pi - get a 2GB Pi 4, a 8GB+ micro SD card, a USB A to C cable, and plug it into a USB 3 port on your PC, and place the pi such that it sits in the airflow path of an exhaust fan. Total cost, $35 if you already have the cable and the SD card, or $45(ish) if you need to get both a cable and a SD card.
 
Are you running 4 processes continuously with Open Pandemics? I would love to get some regular updates on performance / work units if you are.

*Edit* Also, have you tried installing boinc on MotioneyeOS? My guess is the security camera doesn't use much CPU, so you could probably toss 2 WCG processes on those camera Pis.

Here's one of them that's been running the last 6 days uninterrupted.

1592572219565.png


The second one I've been running it on and off as I used it as one of the cameras before I ordered in the 2GB Pi4 to dedicate for that use. But here's its stats before that for about 7 days.
1592572591062.png


Don't "think" I can install boinc on Motioneye as it's dedicated for a camera system and boots into that.

https://github.com/ccrisan/motioneyeos/wiki/Screenshots


Edit - Both Pis are still at stock clocks.
 
Don't "think" I can install boinc on Motioneye as it's dedicated for a camera system and boots into that.

https://github.com/ccrisan/motioneyeos/wiki/Screenshots


Edit - Both Pis are still at stock clocks.
Thanks for the WCG screen shots.

I see a number of guides regarding boinc on buildroot (the OS under motion eye OS), so its likely possible. That said, I don't see a step by step guide to actually do it, so it would probably be a fair amount of research to get it running.
 
Playing around. Using Python to create a .csv file to monitor CPU temperature. I set it to check every 5 seconds while running four threads of O.P. Then suspended the work to let it cool, and then started them up again to see how it did, about two minutes worth.

1592599726580.png


It can live graph it if one wants. The code for it is on the Pi site. :)

https://projects.raspberrypi.org/en/projects/temperature-log
 
Got Rosetta running on the Pi 4 4GB last night. Cannot speak to other flavors of Pi. Due to the current Ubuntu server 20.04 build this wasn't a big deal.

OS - https://ubuntu.com/download/raspberry-pi I used Ubuntu 20.04 LTS 64bit, flashed to a 64gig SD card and ran through the initial install. Once done I downloaded a desktop for it. (easier for my use) then downloaded Boinc and selected Rosetta@Home.



Checked on it this morning and looks like it completed 4 units overnight and started on the others. :)

rosetta.png



rosetta1.png



rosetta2.png



All of this is some initial data. But passing it on in case someone is considering the Pi 4 4GB and wants to run Rosetta. This was all pretty simple to set up and get going, no real magic involved. 👍

The "bad" side is I shut down one of my Pis that was running Open Pandemics to play with this and get it working with Rosetta and in order to let this keep running for several days/weeks to test it with completion times and pass this info on in the D.C. forum. So I ordered another Pi 4 / 4GB last night and will stick that MicroSD card in it and start that O.P. back running on it once it arrives.
 
Also guys, when you are posting info about the Rosetta work units, you should include how long you have your target CPU run time set for. Default is 8 hours but you can adjust this anywhere from 2 hours to 1 day 12 hours. This will give people a better idea if their runtimes are on par with yours.

RosettaTargetRuntime.JPG
 
Also guys, when you are posting info about the Rosetta work units, you should include how long you have your target CPU run time set for. Default is 8 hours but you can adjust this anywhere from 2 hours to 1 day 12 hours. This will give people a better idea if their runtimes are on par with yours.

View attachment 255350
I always use default, and have found no concrete reason to change from the default
 
I change from the default to reduce the data usage as I have data limits on my cable internet. By setting them for longer work units, it is less data transferred and less strain on project servers.
 
I change from the default to reduce the data usage as I have data limits on my cable internet. By setting them for longer work units, it is less data transferred and less strain on project servers.
That would be a good reason then!
 
Also guys, when you are posting info about the Rosetta work units, you should include how long you have your target CPU run time set for. Default is 8 hours but you can adjust this anywhere from 2 hours to 1 day 12 hours. This will give people a better idea if their runtimes are on par with yours.

1592789016081.png



:)
 
Last edited:
Just fired up another Pi 4 2GB (CoconutCream) and split it between Rosetta and WCG. I've realized that naming the hosts after pies may have not been the best idea... I'm already starting to run out of pies that I can think of.

Edit:

In regards to temperature tracking, I just run this little 1 line shell script to keep an eye on things:

while true; do date; vcgencmd measure_temp; vcgencmd measure_volts core; vcgencmd measure_clock arm; temp=`/opt/vc/bin/vcgencmd measure_temp | awk -F "=" '{print $2}' | awk -F "'" '{print int ( $1 )}'`; if [[ $temp -gt $maxtemp ]]; then maxtemp=$temp; fi; echo "Max temp: $maxtemp"; echo; sleep 30; done

output is:

Mon 22 Jun 09:34:29 EDT 2020
temp=45.0'C
volt=0.9563V
frequency(48)=2000478464
Max temp: 46
 
Added another 4GB pi, though Rosetta is currently out of tasks (they are adding more, but there is more available compute capacity than jobs at the moment).

among some of my other work, I’ve moved the project directory for all my boinc installs On my pis to my freenas as phase 1 of no longer using local SD cards. I picked up a 512 970 pro to install on my freenas, which i will split between an ARC cache and remote swap space for the pis. I’ve also spun up a centOS vm on my freenas so that I can setup dns, etc, to PXE boot the pis.

I’ve also now filled my reserve 8 port switch, so I’ll need to pickup a new switch eventually. I’m currently wondering if I should just go with a switch with a 10gbit uplink so i can throw extra bandwidth at the freenas. Decisions decisions.
 
Added another 4GB pi, though Rosetta is currently out of tasks (they are adding more, but there is more available compute capacity than jobs at the moment).

Ahh that explains it. Checked it this morning and it was doing nothing. Thanks for the info. 👍
 
A few more power saving options to apply to /boot/config.txt I've found digging around the official pi forum.

#Disable Ethernet LEDs:
dtparam=eth_led0=4
dtparam=eth_led1=4

# Disable DRM VC4 V3D driver on top of the dispmanx display stack (comment existing line)
#dtoverlay=vc4-fkms-v3d

#disable hdmi frambuffer (does not appear to be covered by tvservice --off)
max_framebuffers=0

#disable audio (minimal power saving, but frees up memory)
dtparam=audio=off
 
Per a suggestion from the raspberry pi forums, I tried forcing the ethernet card to 100mbit. (I did it via ethtool with: sudo ethtool -s eth0 speed 100 duplex full). Surprisingly, it actually saves about .25 watts per pi. When dealing with the fact that a pi is pulling 4.5 watts (ish), that's actually a 5% power saving, though the question is, is .25 wats worth 10x slower network speed. Any opinions out there?
 
For BOINC, there is not really a concern with slower speed as the transfers are usually less than 1 meg at most projects anyways. If this was a streaming device or a gaming device, maybe but for DC...not really and issue.
 
For BOINC, there is not really a concern with slower speed as the transfers are usually less than 1 meg at most projects anyways. If this was a streaming device or a gaming device, maybe but for DC...not really and issue.
That was my original thought, but I’m already running my boinc project directory on nfs. My final plan is to move to network booting the pis and running the entire file system either from nfs or iscsi. Swap over nfs hasn’t been working terribly well (the pis tend to hang if they allocate more than 800mb), so iSCSI is probably the correct solution. With a large dependency on network, gigabit speeds may be preferred.
 
I'm curious what's the crunching efficiency of pi when compare to 3950x when running WCG OP tasks. Normally the crunching efficiency is quoted in PPD/watt (PPD = points per day. I'm using BOINC credit system in the calculation below. For WCG credits, multiply by 7)

My 3950x currently gives about 40 points per hour per thread (pphr/thread). There is the WCG OP forum thread that discusses the performance of other 3950x here. 40 pphr/thread is reasonable for under-clock system. So the PPD = 40 x 32 threads x 24 hours = 30720 PPD. It's lower than stock performance. Most 3950x rigs produce more than 30K PPD in OP if they are running at stock. Currently my 3950x is running at 3.52 GHz at 0.96V drawing 127W from the wall. So PPD/W = 30720PPD/(127W x 24 hrs) = 10.08 PPD/watt.



1593571908888.png



Using the stat from The_Heretic, I calculated that total BOINC points earned during that 7 days is 5195.86 points (convert from WCG to BOINC pts) for a total run time of 557.31 hrs (~23.22 days). This yields about 5195.86/557.31 = 9.32 pphr per thread versus 40 points per hour per thread for 3950x. Assuming pi total power consumption is 4W (?) from the wall, then the PPD/W = (9.32 x 4 threads x 24 hrs)/(4W x 24hrs) = 9.32 PPD/watt. Hopefully my logic and math are correct.

Well 3950x is heck a lot more expensive (very high upfront cost) than pi but just trying to see what makes sense in terms of PPD/watt. Please correct my assumptions if any.

BTW, the pi with 4 threads produces about 895 PPD, so to match the 3950x PPD, would need about 30720/895 =34.3 Pi. Well I built my 3950x with used parts and the total cost is less than $1000. I already have PSU, SSD, case and a GPU. One can use PCPartPicker to estimate brand new system for a fair comparison. Without doing the cost analysis in detail and perhaps very situational, I'm guessing 3950x would give Pi a run for the money.
1593572676915.png
 
I'm curious what's the crunching efficiency of pi when compare to 3950x when running WCG OP tasks. Normally the crunching efficiency is quoted in PPD/watt (PPD = points per day. I'm using BOINC credit system in the calculation below. For WCG credits, multiply by 7)

My 3950x currently gives about 40 points per hour per thread (pphr/thread). There is the WCG OP forum thread that discusses the performance of other 3950x here. 40 pphr/thread is reasonable for under-clock system. So the PPD = 40 x 32 threads x 24 hours = 30720 PPD. It's lower than stock performance. Most 3950x rigs produce more than 30K PPD in OP if they are running at stock. Currently my 3950x is running at 3.52 GHz at 0.96V drawing 127W from the wall. So PPD/W = 30720PPD/(127W x 24 hrs) = 10.08 PPD/watt.

Using the stat from The_Heretic, I calculated that total BOINC points earned during that 7 days is 5195.86 points (convert from WCG to BOINC pts) for a total run time of 557.31 hrs (~23.22 days). This yields about 5195.86/557.31 = 9.32 pphr per thread versus 40 points per hour per thread for 3950x. Assuming pi total power consumption is 4W (?) from the wall, then the PPD/W = (9.32 x 4 threads x 24 hrs)/(4W x 24hrs) = 9.32 PPD/watt. Hopefully my logic and math are correct.

Well 3950x is heck a lot more expensive (very high upfront cost) than pi but just trying to see what makes sense in terms of PPD/watt. Please correct my assumptions if any.

BTW, the pi with 4 threads produces about 895 PPD, so to match the 3950x PPD, would need about 30720/895 =34.3 Pi. Well I built my 3950x with used parts and the total cost is less than $1000. I already have PSU, SSD, case and a GPU. One can use PCPartPicker to estimate brand new system for a fair comparison. Without doing the cost analysis in detail and perhaps very situational, I'm guessing 3950x would give Pi a run for the money.
I'm trimming some of the post to cut down on reply length, but excellent post. I've been trying to narrow down the PPD/watt on the pis as well, as I'm trying to determine how "all in" I want to go with Pis. The Rosetta WU shortage screwed with all my RAC numbers, AND it screwed with my WCG numbers, because the pis I have split between the projects ended up switching to 100% WCG OP while Rosetta was out of work. When I did this exercise with Rosetta, I was finding that a pi 4 power efficiency is approximately the same as a Threadripper 3990X, which I think is pretty similar to a 3950, so if they came out similar I wouldn't be surprised.

My Pi 4 current cluster power use at the wall, with all optimizations in this thread: 34 watts (4.25 W average per pi)

Caveats to my measurements.
1) I do not have a power meter outside of the one integrated into my APC UPs, and it only displays power in whole watts, meaning for a single pi its extremely difficult to get an accurate reading.
2) To resolve 1, I'm averaging the power draw of 8 Pis. However, they are of mixed ram quantity (1x 8GB, 5x 4GB, 2x 2GB), and I've not been able to find any concrete power draw numbers for the extra ram. That means that my average will be high for 2GB pi 4s, low for 8GB pi 4s, and maybe(?) close to correct for the 4GB models.
3) I have not yet finished power optimizations, so this can change as I read the documentation and get more clues on what to tweak / turn off. Ex, I'm expecting to drop some power usage when I switch to network boot and remove SD cards.

My WCG OP points from a Pi 4 running 4 threads while Rosetta was out of work, then some rebalancing when Rosetta WU returned (however Boinc decided to do it, it was running 4 rosetta threads for a while) then 2 threads :
1593610505438.png


I think I'll convert this one to be 100% WCG OP to get a better reading, as its kind of all over the board.

Edit, as for price, my current breakdown per pi, assuming a plan to run a mix of projects to not exceed 2GB ram, and net booting and running without local sd card:

2GB Pi 4: 35
USB A -> C Power Cable: 2.66
1/6 of 6 port Power Brick: 3.99
1/16 of Network Switch: 2.75
Ramsink: ?. I've found 5x 8 packs of ramsinks in my drawer of stuff, and a pi 4 stood on end, in an area with airflow, with ramsink on the arm chip, works at 100% cpu and doesn't throttle.
Total: $44.40 + ramsink
 
Last edited:
If you can somehow measure the current and voltage to the Pi, you should be able to figure out the power for one Pi system. I have also read EPYC rome build that is also very power efficient (probably better than the threadripper) but can't find any normalized PPD/watt number for OP project.
 
If you can somehow measure the current and voltage to the Pi, you should be able to figure out the power for one Pi system. I have also read EPYC rome build that is also very power efficient (probably better than the threadripper) but can't find any normalized PPD/watt number for OP project.
Yeah, I've been considering picking up the tools to measure power draw exactly, but I'm not sure I care that much. Pre-pandemic with the kids in daycare with a regular sleep schedule I would have already done it. Now I would very much prefer someone from the raspberry pi foundation just provide official power usage numbers for every component (good luck, right?).
 
Many libraries let you check out a Kill-A-Watt device now. You may want to check your local establishment...
 
I've three kill-a-watt and all are used for DCing. Since most of my rigs are headless and I'm running a single project most of the time on each PC, the kill-a-watt can tell me if something is wrong with the PC (e.g. PC unresponsive) or the project (e.g. out of work). Just fyi. Personally, it is worth the investment for me since my local electricity rate (plus taxes) is high. I'm always looking for ways to run the rigs more efficiently. The 3950x is my first AMD build and I like it. Dumping xeons unless they are really cheap to acquire for DC. I'm really interested in pi in regards to power efficiency, so please keep up the good work. ;)
 
I've three kill-a-watt and all are used for DCing. Since most of my rigs are headless and I'm running a single project most of the time on each PC, the kill-a-watt can tell me if something is wrong with the PC (e.g. PC unresponsive) or the project (e.g. out of work). Just fyi. Personally, it is worth the investment for me since my local electricity rate (plus taxes) is high. I'm always looking for ways to run the rigs more efficiently. The 3950x is my first AMD build and I like it. Dumping xeons unless they are really cheap to acquire for DC. I'm really interested in pi in regards to power efficiency, so please keep up the good work. ;)
How accurate is the kill-a-watt at low wattage? Does it read milliamperes?
 
I still need to e-mail APC to find out the accuracy on their 1500 VA UPS. I'm guessing the kill-a-watt is more accurate.

Does it display fractional watts, or is it in whole watts only?

Edit - I agree on wanting to track this for efficiency sake. For me, my wife is onboard with using the extra power generated by our Solar panels, but she isn't interested in actually spending money on power for my computers. I really want to stretch the power available to me, so as I add more pis I'm turning off older hardware. I've already decommissioned an old i3 CPU M 370 laptop I was using, and next to be powered down will be an A10-5800K and a Core2Quad 9650.

Edit 2 - The Broadcom chips in the Pi 4s are 14nm chips, and they are competing favorably against AMD at 7nm, so I guess I shouldn't be too hard on the Pis. That said, managing 12 pis is already a bit of work, and I can tell it would be substantially worse for someone that doesn't have IT work as a day job.
 
Last edited:
Does it display fractional watts, or is it in whole watts only?
For less than 100W, it will show one decimal place, eg 14.3W. Above 100W, it is in whole watts. I can't recall if it can show two decimal places for the power when reading below 10W.
 
Managing a gazillion devices is and was the worst part of having a phone farm or phARM. I'm not going out of my way to get more as it was a PITA to keep track of them and make sure they were happy. I still add to my collection as I find super great deals or freebies, but I'm certainly not investing in them. At least not for DC uses.
 
Managing a gazillion devices is and was the worst part of having a phone farm or phARM. I'm not going out of my way to get more as it was a PITA to keep track of them and make sure they were happy. I still add to my collection as I find super great deals or freebies, but I'm certainly not investing in them. At least not for DC uses.
I’m thinking of running boinc in containers and building a kubernetes cluster to do all the Container management with Prometheus to monitor the whole thing. That has a performance penalty though, so if I can’t get the pis more efficient than AMD as a baseline I’m not very enthusiastic about making them less efficient.
 
One of the pi's for 21 days on Open Pandemics.

rasp_pi_21days.jpg


The numbers come out to an average of 8.19 results per day and 4994.9 points (ehh 5K) over a 3 week period undisturbed. This is with stock pi at 1.5Ghz, nothing on the board disabled, using the wireless.
 
A few notes on the 8gb pi.

1)running Rosetta at 1525 MHz, an 8gb pi 4 scores about the same 1350 RAC as an overclocked 4GB model. It’s around a 25% boost over a stock 4gb model’s 1100 - I guess zram has a bigger impact than I thought

2) per the raspberry pi forums, the power delivery components on the 8gb pi had to be increased to support 8gb of ram. This is confirmation of my suspicion that extra ram increases power draw a measurable amount. I really need that killawatt to get some exact numbers.
 
Seems ok but never own one. The manual says accuracy standard class 1.0 meaning error can go up to 2% which is comparable to kill-a-watt.
 
I did it old school here with a multimeter to measure current draw and using Ohm's Law. (being that I'm an old tech)


First the pi sitting booted up to the desktop in the Pi OS and at idle. Again with mine the CPU is a stock clocks and no hardware is disabled on the pi.

rasppiopenpand1.jpg


Next I loaded Open Pandemics with 4 threads running and the pi under load.

rasppiopenpand2.jpg


Using Ohm's Law (P=E x I) so 5 vdc x 890 milliamps = 4.45W


I've included a video below showing about 12 seconds of it running under load to show how that current draw is varying some, in this as high as 4.8W for a second. Also I did see it touch 1 amp (5W) for a second after it first started running O.P. So in my particular case it's running under load from around 4.4 to peaks of 5W.
 
Back
Top