Hi there,
I thought I'd share the results of my passive cooling testing and my efforts to avoid over-baking my Pi 4.
Aim:
Run the Pi 4 at high load without the use of active cooling. Fan noise is a pet hate of mine. Generic IC heatsinks were the preferred option due to their low cost and good availability.
Test setup:
![Image]()
* Raspberry Pi 4 Model B Rev 1.5
* Official Pi 4 USB-C Power Supply
* Pi 4 Official Case (Lid removed as per photo)
* 14.5x14x13mm SoC heatsink
* 14x10x6mm RAM heatsink
* 9x9x5mm USB hub heatsink
* Bookworm (64-bit, Wayfire)
* arm_boost disabled in /boot/config.txt - 1.5 GHz max ARM clock
* 4k60 mode enabled in /boot/config.txt or raspi-config - 550 MHz GPU clock
Tests were run on the Wayfire desktop, connected to a 1440p 120 Hz monitor.
One USB keyboard/mouse receiver was connected to one of the USB 2.0 ports.
Wifi was enabled and connected to a network.
The ambient room temperature was 19C (because Scotland).
Tests:
CPU stressor:GPU stressor:
These commands provoke 99-100% CPU and GPU usage in Task manager (The `grim` screenshot tool temporarily drops the GPU usage while capturing as can be seen in the screenshot).
The tests were run until the temperature plateaued for 10 minutes.
Results:
Right-click on the below screenshot and select 'Open image in new tab' to view the full resolution image (and be able read the text).
![Image]()
As can be seen, the reported temperatures are coming in just below the throttling threshold.
Analysis:
While I am pleased with this result, I think that during a warm summer day that we would see thermal throttling due to increased ambient room temperature. However I am still satisfied with the cooling performance considering that this test is well beyond the hardware utilisation that normal desktop usage demands.
The SoC heatsink is a step above the basic offerings, with a good number of tall fins in a dense array. This provides increased surface area for thermal transfer, while still being compatible with the official case. Touching the the top of the heatsink during the stress tests cannot be done for any longer than 2-3 seconds, which can be used as a simple method to gauge thermal transfer performance between the heat source and heatsink.
The memory heatsink gets very hot to the touch during these stress tests. Not quite as hot as the SoC heatsink, but still hot to the point that one wouldn't want to keep a digit pressed down on it for more than 5-10 seconds. Which is why I was surprised not to see a thermal pad for the Pi 5's memory chip on the official Active Cooler.
The USB hub heatsink doesn't get very warm. I can only hypothesise that perhaps the USB interface would need to be hammered with transfers to get this chip to heat up. For these tests, this heatsink is basically cosmetic and not required.
It could be argued that disabling the 1.8 GHz ARM boost clock is excessive - indeed the default boost behaviour is most likely more optimal for most users. The increased thermal headroom (of around 5C or so at max load, from my observations) is a personal preference on my part.
I still think the official case could have done with a redesign to a the 3-piece format like that of the Pi 3 and 5's though, as with the lid clipped on the Pi 4 is only usable for light tasks before throttling occurs. I leave the lid off during operation for this reason, then clip the protective lid back on once the Pi is switched off. It isn't unworkable, but definitely not as practical a design as the 3 and 5 cases which are excellent.
In terms of sustained performance, I have left the stress tests running for the entire time I have been typing up this post (~1hr), with the SoC temperature staying stable at ~75C.
Closing Comments:
I hope this information is useful to someone - the early days of the Pi 4 being a toaster are long gone and it is a very well-behaved SBC now, in my opinion. Now the Pi 5 can take that prestigious title - as it is, in fact, a very speedy toaster.
Thanks for reading.
I thought I'd share the results of my passive cooling testing and my efforts to avoid over-baking my Pi 4.
Aim:
Run the Pi 4 at high load without the use of active cooling. Fan noise is a pet hate of mine. Generic IC heatsinks were the preferred option due to their low cost and good availability.
Test setup:
* Raspberry Pi 4 Model B Rev 1.5
* Official Pi 4 USB-C Power Supply
* Pi 4 Official Case (Lid removed as per photo)
* 14.5x14x13mm SoC heatsink
* 14x10x6mm RAM heatsink
* 9x9x5mm USB hub heatsink
* Bookworm (64-bit, Wayfire)
* arm_boost disabled in /boot/config.txt - 1.5 GHz max ARM clock
* 4k60 mode enabled in /boot/config.txt or raspi-config - 550 MHz GPU clock
Tests were run on the Wayfire desktop, connected to a 1440p 120 Hz monitor.
One USB keyboard/mouse receiver was connected to one of the USB 2.0 ports.
Wifi was enabled and connected to a network.
The ambient room temperature was 19C (because Scotland).
Tests:
CPU stressor:
Code:
stress-ng --matrix 0 -t 0Code:
glmark2-es2 -b shadow:duration=60 -s 1440x1200 --run-foreverThese commands provoke 99-100% CPU and GPU usage in Task manager (The `grim` screenshot tool temporarily drops the GPU usage while capturing as can be seen in the screenshot).
The tests were run until the temperature plateaued for 10 minutes.
Results:
Right-click on the below screenshot and select 'Open image in new tab' to view the full resolution image (and be able read the text).
As can be seen, the reported temperatures are coming in just below the throttling threshold.
Analysis:
While I am pleased with this result, I think that during a warm summer day that we would see thermal throttling due to increased ambient room temperature. However I am still satisfied with the cooling performance considering that this test is well beyond the hardware utilisation that normal desktop usage demands.
The SoC heatsink is a step above the basic offerings, with a good number of tall fins in a dense array. This provides increased surface area for thermal transfer, while still being compatible with the official case. Touching the the top of the heatsink during the stress tests cannot be done for any longer than 2-3 seconds, which can be used as a simple method to gauge thermal transfer performance between the heat source and heatsink.
The memory heatsink gets very hot to the touch during these stress tests. Not quite as hot as the SoC heatsink, but still hot to the point that one wouldn't want to keep a digit pressed down on it for more than 5-10 seconds. Which is why I was surprised not to see a thermal pad for the Pi 5's memory chip on the official Active Cooler.
The USB hub heatsink doesn't get very warm. I can only hypothesise that perhaps the USB interface would need to be hammered with transfers to get this chip to heat up. For these tests, this heatsink is basically cosmetic and not required.
It could be argued that disabling the 1.8 GHz ARM boost clock is excessive - indeed the default boost behaviour is most likely more optimal for most users. The increased thermal headroom (of around 5C or so at max load, from my observations) is a personal preference on my part.
I still think the official case could have done with a redesign to a the 3-piece format like that of the Pi 3 and 5's though, as with the lid clipped on the Pi 4 is only usable for light tasks before throttling occurs. I leave the lid off during operation for this reason, then clip the protective lid back on once the Pi is switched off. It isn't unworkable, but definitely not as practical a design as the 3 and 5 cases which are excellent.
In terms of sustained performance, I have left the stress tests running for the entire time I have been typing up this post (~1hr), with the SoC temperature staying stable at ~75C.
Closing Comments:
I hope this information is useful to someone - the early days of the Pi 4 being a toaster are long gone and it is a very well-behaved SBC now, in my opinion. Now the Pi 5 can take that prestigious title - as it is, in fact, a very speedy toaster.
Thanks for reading.
Statistics: Posted by JimJamJamie — Sat Jan 20, 2024 2:25 am — Replies 5 — Views 381