RAM cooling, thermal pads, paste, and height difference

[foresto]

The wiki says we can use either the thermal pad or thermal paste between the heat sink and the rk3399. However, the RAM chips and SoC have different heights, so using the usual thin layer of paste would leave the RAM unprotected. (Worse than that, actually, since the heat sink would block airflow across the RAM surface.)

This leads me to a few questions:

1. What exactly is the height difference between the RAM and SoC? Is a mechanical diagram available? (This would tell us the thickness required for copper shims to compensate for the difference.)

2. If we were to cut the thermal pad (included with 30mm heat sink) into smaller pieces for use on the RAM chips, and then apply a very thin layer of paste to the SoC (with no pad), would the pads on the RAM prevent the heat sink from sitting flush against the SoC + paste?

3. Thomas Kaiser (tkaiser of Armbian) mentions in this post the risk of a short when using the heat sink with thermal paste alone, but the picture he refers to is no longer available. What contacts might short in this scenario?

4. Can the DRAM ever get hot enough to require a heat sink? (Maybe it doesn't need cooling?) If so, what sort of load causes it to get that hot?

[tllim]

The wiki says we can use either the thermal pad or thermal paste between the heat sink and the rk3399. However, the RAM chips and SoC have different heights, so using the usual thin layer of paste would leave the RAM unprotected. (Worse than that, actually, since the heat sink would block airflow across the RAM surface.)

This leads me to a few questions:

1. What exactly is the height difference between the RAM and SoC? Is a mechanical diagram available? (This would tell us the thickness required for copper shims to compensate for the difference.)

2. If we were to cut the thermal pad (included with 30mm heat sink) into smaller pieces for use on the RAM chips, and then apply a very thin layer of paste to the SoC (with no pad), would the pads on the RAM prevent the heat sink from sitting flush against the SoC + paste?

3. Thomas Kaiser (tkaiser of Armbian) mentions in this post the risk of a short when using the heat sink with thermal paste alone, but the picture he refers to is no longer available. What contacts might short in this scenario?

4. Can the DRAM ever get hot enough to require a heat sink? (Maybe it doesn't need cooling?) If so, what sort of load causes it to get that hot?

Heatsink just needed on top of CPU, not necessary on DRAM.

[Gienek]

Hello!
I have considered the same question, and finally measured temperature of the RAMs. They are significantly cooler than the CPU. Even during intensive RAM-load test. So connecting them to the heat sink of CPU would actually cause heating the RAMs.
I have mounted CPU heat sink (the original 30mm tall one from Pine64) only with thermal paste. As far as I could see, it is not touching any element except the CPU. The paste, I used is not electrically conductive (the new MX-4 from ARCTIC). So there is also no danger if I applied too much.
Since then, CPU temperatures are much lower!
Best regards,
Gienek.

[foresto]

I have considered the same question, and finally measured temperature of the RAMs. They are significantly cooler than the CPU. Even during intensive RAM-load test.

I'm interested; can you share the details of your test?

What load did you apply to the RAM? Was the load constant? For how long had the peak load temperature remained constant before you took your final measurement? What were the temperature readings before and during the load test? What was the ambient temperature? How did you measure temperatures?

[Gienek]

Hello,
  Sorry for late answer. I was trying to retrieve my notes from the test. Unfortunately, unsuccessful. So I repeat from my memory.
I used two TMP36 temperature sensors. One attached to RAM, one – to the CPU. Both were read by two analog channels of Arduino PRO Mini. In addition, I checked temperature of CPU delivered by /sys/class/thermal/thermal_zone0/temp (and ... thermal_zone1... ). I had to attach external sensor to a RAM since I do not know about any internal sensor reading RAM temperature.
Ambient temperature was between 18 and 19 Celsius, but not really controlled. Just room temperature in winter here in Germany. The board was simply supported on corners, on a table – no case.
Temperature data from all sensors was collected every 8 seconds, and stored in a plain file, together with a time stamp. Unfortunately, I cannot find these files anymore. Aim of the test was to decide whether I need cooling of the RAMs, so after the decision I have not cared of the data.
I executed 3 tests:
- CPU load with stress -c 6 -t <several minutes>
- CPU load with stress -c 6 -t <several minutes>, plus playing high quality video
- Memory load with stress -m 6 -t <several minutes>
Since I could not attach sensor to RAM with the heat sink mounted to CPU. The test was done without it. The second sensor was attached to CPU, but it cannot dissipate heat significantly. Despite of that, CPU has not reached 80 Celsius during the test.
I remember that temperature shown by the sensor attached to CPU was few degrees lower than the internal sensors value. The difference between temperature read from RAM and from CPU was varying, but never less than 20 degrees of Celsius. Also they were heating and cooling with different speed.
Now, when you asked this question again, I realized that the CPU temperature with attached heat sink is much lower than during my measurement. So it would be very interesting to check the RAMs temperature with the heat sink on CPU. I will try to attach a temperature sensor to RAM somehow. TMP36 seems too big for this purpose. I need to find some way... And it is summer here. Air in that room where ROCKPro is located is terribly hot – around 27 Celsius.
Best regards,
Gienek.