Heat issue solved by separating SATA HAT and Pi

This article describes in three phases how to achieve an almost fan less SATA HAT with maintaining the lifetime of the equipment.

  • Phase I „everybody could do“
  • Phase II „read HDD temps“
  • Phase III „bigger display“

Disclaimer:
The modifications in this article are based on my experiences and if you find them useful to use you do so at your own risk.

The major issue to solve was that the Pi even in idle generate a lot of heat and this raising up heat is additionally heating the on top of it mounted HDDs which are producing heat by themself.

The solution to that is to separate HAT and Pi.

Phase I „everybody could do“

The first phase could be achieved by everybody as necessary parts could be bought and it requires no change on software, three items are needed:

  • A aluminum case, the one used here keeps the Pi 4b fan less cooled in idle at 40 degree and on 100% load at 65 degree with room temperature around 20 degree
  • A GPIO extension cable (15 cm)
  • Two USB 3.0 A-Male to A-Female extension cables

The USB 3.0 connection adapter coming with the HAT has on both ends male connectors but is wired like an „extension“ cable (1:1).
Therefore a USB 3.0 A-Male to A-Male „connector“ cable can not be used as those have a swapped wiring.
For the extension cables two things need to be considered:

  • The thickness of the female connector should not exceed 9.2 mm
  • The female connector surface should be plastic (not metal) to avoid unwanted electrical contact

Due to internal wiring of the connection adapter coming with the HAT the USB 3.0 cables need to be plugged in that way:

  • top HAT connector to lower Pi connector
  • lower HAT connector to top Pi connector

With that no software modification is required and the standard Top board can be used as indicated in the pictures.

Phase II „read HDD temps“

To ensure for all use cases a good temperature for the HDDs it is necessary to have some related measurements.

I did not find a way by available commands to measure the HDD temps when they are spun down to preserve lifetime without initiating a spin up caused by the measurement.
In that context „standby“ does not always seem to imply spun down.

The SATA HAT software already supports the reading of one w1 connected external temperature sensor. This external measurements are good indication to trigger fan if needed.

Starting from there putting things together with some not that difficult soldering work.

A board with up to 4 pluggable DS18B20 temperature sensors.

But how to connect sensors without e.g. soldering tapings on HAT GPIO?
As the CPU fan socket of the HAT is not used anymore it is a perfect candidate for w1 usage.
The socket delivers a needed voltage supply, but at 5V.
This can be reduced by a 1k Ohm / 2,2k Ohm resistor divider to the needed 3,3V and should be accompanied with the 4,7k Ohm PullUp resistor for the data wire.

As GPIO 12 is now used for w1 instead of CPU fan a few changes in software are necessary and with that already one sensor can take his duties, e. g. if only the HAT case temperature is from interest to trigger fan.
In /boot/config.txt change or add

dtoverlay=w1-gpio,gpiopin=12

In misc.py extend

f.write(content.strip() + '\ndtoverlay=w1-gpio,gpiopin=12')

In fan.py change (as CPU is passively cooled)

#    return max(t1, t2)
     return t1

and comment out

#    gpio.hardware_PWM(12, 0, 0)

and comment out

#    gpio.hardware_PWM(12, 25000, dc * 10000)

For displaying temperatures and using up to four sensors some more software changes are needed.
Below find the quick and „hard coded“ changes for two sensors.

In misc.py add

def get_w1_temp(i):
    temp = "{:.1f}°C".format(conf['w1'][i].value)
    return temp

and add after conf[‘oled’][‘f-temp’] = cfg.getboolean(‘oled’, ‘f-temp’)

        conf['w1'][0] = mp.Value('d', -1)
        conf['w1'][1] = mp.Value('d', -2)
        conf['w1'][2] = mp.Value('d', -3)
        conf['w1'][3] = mp.Value('d', -4)

In fan.py replace the whole read_temp and adapt the w1 IDs to yours

def read_temp(cache={}):
    w1_slave = cache.get('w1_slave')
    if not w1_slave:
        try:
            w1_slave = next(Path('/sys/bus/w1/devices/').glob('28*/w1_slave'))
        except Exception:
            w1_slave = 'not exist'
        cache['w1_slave'] = w1_slave

    if w1_slave == 'not exist':
        t1 = 42
        t2 = 42
    else:
      try:
        w1_slave = '/sys/bus/w1/devices/28-3c01b5561f5f/w1_slave'
        with open(w1_slave) as f:
            t1 = int(pattern.search(f.read()).groups()[0]) / 1000.0
            misc.conf['w1'][0].value = t1
      except Exception as e:
        logging.info("show EX 1: "+str(e))
        t1 = 42

      try:
        w1_slave = '/sys/bus/w1/devices/28-3c01b5564d06/w1_slave'
        with open(w1_slave) as f:
            t2 = int(pattern.search(f.read()).groups()[0]) / 1000.0
            misc.conf['w1'][1].value = t2
      except Exception as e:
        logging.info("show EX 2: "+str(e))
        t2 = 42

    with open('/sys/class/thermal/thermal_zone0/temp') as f:
        tC = int(f.read().strip()) / 1000.0

    return max(t1, t2)

As already said it is quick code and can be made with loops and configuration more comfortable.

In oled.py e.g. replace displaying IP by temperatures

#            {'xy': (0, 21), 'text': misc.get_info('ip'), 'fill': 255, 'font': font['11']},
            {'xy': (0, 21), 'text': misc.get_w1_temp(0), 'fill': 255, 'font': font['11']},
            {'xy': (64, 21), 'text': misc.get_w1_temp(1), 'fill': 255, 'font': font['11']},

My preferred solution is a front mount of the Top board with the fan blowing.
For that a slight longer HAT to Top board cable is needed (JST PHD 2x5).

Phase III „bigger display“

For my personal needs I liked to have

  • shorter and A-male to A-male USB 3.0 cables
  • a bigger display to have information needed on one page
  • to by key toggle display on/off to preserve its lifetime

As described above the A-male to A-male need to be wired as kind of „extension“, means 1:1 without swapping wires.
I bought a standard A-male to A-male cable, cut it in two halves, identified the half with the swapped wires and considered this to „unswap“ when soldering the connector on the cut end.

I found a SH1106 controlled 128x64 OLED. To install execute

sudo -H pip3 install luma.oled

Now the software changes for the SH1106 display and the new action of the key to toggle the display on/off.
In main.py change

    #    'slider': lambda: oled.slider(lock),
    'slider': lambda: oled.slider_key(lock),

In misc.py extend

    conf = { 'disk': [], 'idx': mp.Value('d', -1), 'run': mp.Value('d', 1), 'hidden': mp.Value('d', 0) }

And finally replace oled.py with this (which is pretty close to the original oled.py)

#!/usr/bin/python3
import time
import misc
from PIL import Image
from PIL import ImageDraw
from PIL import ImageFont

from luma.core.interface.serial import i2c
from luma.oled.device import sh1106

font = {
    '10': ImageFont.truetype('fonts/DejaVuSansMono-Bold.ttf', 10),
    '11': ImageFont.truetype('fonts/DejaVuSansMono-Bold.ttf', 11),
    '12': ImageFont.truetype('fonts/DejaVuSansMono-Bold.ttf', 12),
    '14': ImageFont.truetype('fonts/DejaVuSansMono-Bold.ttf', 14),
}

device = None

def disp_init():
  global device
  serial = i2c(port=1, address=0x3C)
  device = sh1106(serial)
  return None

try:
    disp = disp_init()
except Exception:
    misc.open_w1_i2c()
    time.sleep(0.2)
    disp = disp_init()

image = Image.new('1', (128, 64))
draw = ImageDraw.Draw(image)

def disp_show():
    im = image.rotate(180) if misc.conf['oled']['rotate'] else image
    try: 
      device.display(im)
    except Exception as e:
      pass
    draw.rectangle((0, 0, 128, 64), outline=0, fill=0)

def welcome():
    draw.text((0, 0), 'ROCK Pi SATA HAT', font=font['14'], fill=255)
    draw.text((32, 16), 'loading...', font=font['12'], fill=255)
    disp_show()

def goodbye():
    draw.text((32, 8), 'Good Bye ~', font=font['14'], fill=255)
    disp_show()
    time.sleep(2)
    disp_show()  # clear

def gen_pages():
    pages = {
        0: [
            {'xy': (0, -3), 'text': time.strftime ('%d.%m.   %H:%M'), 'fill': 255, 'font': font['11']},
            {'xy': (0, 9), 'text': misc.get_cpu_temp(), 'fill': 255, 'font': font['11']},
            {'xy': (0, 20), 'text': misc.get_info('cpu'), 'fill': 255, 'font': font['11']},
            {'xy': (0, 31), 'text': misc.get_w1_temp(0), 'fill': 255, 'font': font['11']},
            {'xy': (0, 42), 'text': misc.get_w1_temp(1), 'fill': 255, 'font': font['11']},
            {'xy': (0, 53), 'text': misc.get_info('ip'), 'fill': 255, 'font': font['11']},
        ]
    }
    return pages

def slider_key(lock):
    misc.conf['hidden'].value = not(misc.conf['hidden'].value)
    slider(lock)
    global device
    if not(misc.conf['hidden'].value): 
      device.show()
    else:
      device.hide()

def slider(lock):
    with lock:
      if not(misc.conf['hidden'].value): 
        for item in misc.slider_next(gen_pages()):
           draw.text(**item)
        disp_show()

def auto_slider(lock):
    while misc.conf['slider']['auto']:
        slider(lock)
        misc.slider_sleep()
    else:
        slider(lock)

`

Hope you enjoyed reading and find it inspiring – have fun :slight_smile:
Michael

Everything you do - you do so at your own risk

7 Likes

Awesome post @michael! I dont have the Pi4 or SATA hat, yet found it a very enjoyable read. Well done and thanks for sharing!

Thank you very much! I really love that tiny device, it was a lot of fun to combine soldering, acryl glas and software together.

1 Like

Some observations after a one and a half month of service.
The main usage is as Plex Mediaserver (and PiHole and mini DLNA and sharedrive for family).

I started with two HDDs and added a third one. To keep HDD spin ups as low as possible I added as fourth disk a SSD for usage as “incoming” for new and most recent viewed videos.
But with that step I had some strange cases.

First case - 50 minutes cycle
At the time as HAT was HDD only, in the night (= with close to no load) the fan never spun up - as intended.
But after adding the SSD the logs indicated nightly fan activity.
It seemed to occur with a periodity of 50 minutes, which does not point to a cron job (= typically hourly).
Another possible explanation was that the idle SDD produces heat to “stable” trigger fan threshold of 35 degree every 50 minutes.
To prove I put all data in excel and generated a chart and indeed:


The upper blue line shows the CPU temperature, the green line the fan duty cycles.
From far it looked like a flat line but zooming in in the 32-35 degree area of the here two temperature sensors (blue and gray) it became visible.
A stable control circle - possible as room temperature is stable.
With that indication I found in a datasheet that the old Crucial MX100 256GB has a high idle power consumption.
Ok, as I like to keep fan spin as low as possible I bought a recent Crucial BX500 240GB and entered the

Second case - never wake up
With the new SSD no nightly fan activity occured - great.
But next evening the video did not start.
After some testing it became evident that in this setup if the BX falls asleep it never wakes up - except by rebooting the system.
This is different to all other HDDs and SSDs I put over the time into the system.
I found an article in the internet describing same behavoiur in the context of a music player system.
So I returned it and got a WD Green 240GB.

Now eveverything is fine
SSD wakes up when video need to be streamed and there is no nightly fan activity.


You could see the expected fan duty cycles during video activity: viewing and upload.
And the opening of the window in the morning (07:21:51) to refresh room, dropping room temperature :slight_smile:

Hope you enjoyed reading – have fun :slight_smile:
Michael

1 Like

One year of service


fan duty close to zero, display usage duty close to zero

On updates ensure to re-set in /lib/systemd/system/pigpiod.service
ExecStart=/usr/bin/pigpiod -l -m -n 127.0.0.1 -p 8888
otherwise fans will not run.

I whish you all a Happy New Year :slight_smile:
Michael

I bought this USB-powered fan: https://www.amazon.com/gp/product/B07S51M3R5/ref=ppx_yo_dt_b_asin_title_o01_s00?ie=UTF8&th=1

and I placed it under the quadsata and plugged it into one of the USB 2.0 ports.

My temp dropped to 34C on idle! You can see the temp dropping on the graph. I’m happy now. :grinning:

Now 3 years of service

2 TB SSD
I replaced the 240 GB WD Green SSD with a 2 TB Crucial MX500 SSD.
This - besides hosting the ftp and the family shared drive with all pictures - acts as the 240 GB before as an incoming drive for new videos.
With that 90% of the workload goes to that SSD and significantly lower the load on the three HDDs which are acting more as an all time ready archive.

Control HDDs fan on real HDD temps based on lsof
As written above in first post it would be great to know the real HDD temps, especially when they are active. Just recap:
   Using smartctl when discs are sleeping wake them up
   Ongoing smartctl prevent them falling asleep
So the task is just to use smartctl only when they are active.
But even hdparm to know if they are active wakes them up as well.
So I wrote an shell script which each minute lsof the mounts of the four drives to detect activity on a drive and if yes, then to smartctl its real temperature and to write the highest value of the active ones in a file.
The reading of that file was easly integarted as another temparture source in the fan.py. For sure that lsof functionality can be fully integrated in fan.py.
But why the effort? It runs so well as is :slight_smile:

I wish you all a Happy New Year
Michael

i wanna try this, i see that you conected the fan to the jst phd conector of the hat, what pins did you use? can you provide a basic circuit diagram for the temperature board, where do you conected that?

The fan is connected aligned with top board documentation, which enables me for test purposes to still connect the original hat
For the fan I used pin 4, 8, 9

 
Please find below the wiring diagram based on the soldering dots:
left side, blue marked pins: JST to Hat
right side, green marked pins: connector to temperature sensors board


As indicated in the above section "Phase II read HDD temps“ the JST connects to the CPU fan socket of the HAT. Doing so, as then GPIO 12 becomes input and no longer PWM, the software needs to be adapted as described.

If you are heading for the new Penta Hat with PI5 making advantage of PCIe:
I started evaluating a setup with no external temp sensors and no self soldered USB3 cables which I plan to post this winter.

2 Likes

First of all thanks for the reply and the explanation, I’m working in the new Penta SATA Hat, in the reverse of the board I can find a fan socket but with two pins, this is blowing my mind… I purchased the sata hat during 2024

ya, that socket is 2 pin only, so no PWM and 100% speed on that header
on the 5x2 pin ones PWM is working great

I will check again the Whacking diagram because always the fan is full Throttle

I’m going to the origin, please really sorry for my basic speech but I need to know if I missed something.

This is my fan. It has4 pins, one for ground (black) another one 5V (red) and pwn control (yellow) the blue pin is for show the rpm that fan is working (output info from fan)

The black goes to pin 9, red to pin4 and yellow to pin8.

Has any config I need to set in my raspberry to control that? I installed the rockpi package to set the threshold with levels.

Where is the mistake in this post…

please see my answer in your starting thread

That one? Yeah I aligned with you that 2 pin is not for pwn, but i’m lost with the mistake