7 ms·

I just tried out the Pi 4 yesterday and after a couple of hours of headless idling with no peripherals connected (just wifi) it was so hot, I almost burned my f

I just tried out the Pi 4 yesterday and after a couple of hours of headless idling with no peripherals connected (just wifi) it was so hot, I almost burned my finger. So this is definitely welcomed. But heatsinks or even fans will likely be necessary to do anything with the pi 4

penagwin 7y ago
I don't have a pi 4 to play with, but is there any chance you could try one of the mini heatsinks that just adhere onto the cpu, without any active cooling? I'm just curious to see if that helps much, they're pretty cheap and could be useful other places too (I use one on the VTX for my drone) You can check the temperature though the pi itself. From my experience depending on the material/length of contact, things start to get too hot to comfortably hold after ~55C/60C for most metals, but brief contact might barely feel warm. [0] https://www.amazon.com/Easycargo-Raspberry-Heatsink-Aluminum-conductive/dp/B07217N5LS/ https://www.amazon.com/Easycargo-Raspberry-Heatsink-Aluminum...
- nielsole 7y ago
  I am trying to hop by an electronics store tonight. Will keep you posted :)
- megous 7y ago
  Those are mostly useless (they only slow down temperature changes a little). Only thing that really helps is active cooling / moving the air. You also have to figure out the placement of the board. The air will move via convection, and if you get air movement from a hotter part of the board (PMIC) to the SoC, you'll not improve things much. I don't think perf gains in Rpi4 are of much use, if it idles at 65°C (my ambient temperature is ~32°C at home) and has to throttle down to <1GHz under persistent load.
  - close04 7y ago
    > they only slow down temperature changes a little That already helps a lot if you don't need to maintain full power for long periods. It gives just enough thermal inertia to handle spikes without throttling. For extended periods you may need to move to bigger heatsink or active cooling.
  - BubRoss 7y ago
    ARM A72 chips use out of order execution, which already gives 3x the performance.
    - penagwin 7y ago
      I'm curious what benchmarks you've seen that actually show a 3x compute improvement? Most of the benchmarks I've seen show a ~30% improvement in CPU compute, while I/O has improved tremendously, as well as the memory. But again I've only seen ~30% improvement in CPU workloads, a far cry from "3x the performance" that I keep seeing.
      - BubRoss 7y ago
        You may have seen GPU comparisons. https://www.raspberrypi.org/magpi/wp-content/uploads/2019/06/2019-06-21-15_40_11-Window.png https://www.raspberrypi.org/magpi/wp-content/uploads/2019/06... https://www.raspberrypi.org/magpi/wp-content/uploads/2019/06/2019-06-21-15_41_55-Window.png https://www.raspberrypi.org/magpi/wp-content/uploads/2019/06...
        megous 7y ago
        Speedometer (chrome) may be more dependent on mem bandwidth which is 2x of RPi 3.
        BubRoss 7y ago
        What makes you say that? Dynamic languages are usually chasing pointers which is something that out of order execution helps with tremendously.
        megous 7y ago
        This is supposedly some TODO app being run repeatedly in Chrome, so it's much more than a JS benchmark. And Chrome as a whole is memory hungry.
  - beatgammit 7y ago
    For my use case, the main attraction for me in the RPi4 is the additional RAM and better IPC/ISA, not necessarily the higher clock (though higher clock is certainly welcome).
  - Dylan16807 7y ago
    Tripling the surface area doesn't help? And those fins should help convection in most orientations.
    - megous 7y ago
      You're not trippling the surface area. (Part of) the board already serves as a heatsink. You need to count that in.
- nielsole 7y ago
  So some data is now in(Obviously with the old firmware, not the one mentioned in the article): Without fan, idle: 58°C peak With fan idle: 39°C (roughly 26°C ambient temperature) With fan 4x parallism of `yes > /dev/null`: 51°C peak The fan is a really small one, just for the CPU. And I didn't quite place in the center of the chip :/ Quite surprised of how much I overestimated idle temperature. 58°C feels really hot when you touch the SoC.
  - penagwin 7y ago
    Awesome thanks for the data! > 58°C feels really hot when you touch the SoC. This obviously depends on the heat conductivity of the surface, but from my experience with my 3D printer, ~55C/60C is about the cutoff for pain tolerance for me. Just below that temperature is holdable, after that temperature is when it's painful to touch.

penagwin 7y ago

I don't have a pi 4 to play with, but is there any chance you could try one of the mini heatsinks that just adhere onto the cpu, without any active cooling? I'm just curious to see if that helps much, they're pretty cheap and could be useful other places too (I use one on the VTX for my drone) You can check the temperature though the pi itself. From my experience depending on the material/length of contact, things start to get too hot to comfortably hold after ~55C/60C for most metals, but brief contact might barely feel warm. [0] https://www.amazon.com/Easycargo-Raspberry-Heatsink-Aluminum-conductive/dp/B07217N5LS/ https://www.amazon.com/Easycargo-Raspberry-Heatsink-Aluminum...

nielsole 7y ago
I am trying to hop by an electronics store tonight. Will keep you posted :)
megous 7y ago
Those are mostly useless (they only slow down temperature changes a little). Only thing that really helps is active cooling / moving the air. You also have to figure out the placement of the board. The air will move via convection, and if you get air movement from a hotter part of the board (PMIC) to the SoC, you'll not improve things much. I don't think perf gains in Rpi4 are of much use, if it idles at 65°C (my ambient temperature is ~32°C at home) and has to throttle down to <1GHz under persistent load.
- close04 7y ago
  > they only slow down temperature changes a little That already helps a lot if you don't need to maintain full power for long periods. It gives just enough thermal inertia to handle spikes without throttling. For extended periods you may need to move to bigger heatsink or active cooling.
- BubRoss 7y ago
  ARM A72 chips use out of order execution, which already gives 3x the performance.
  - penagwin 7y ago
    I'm curious what benchmarks you've seen that actually show a 3x compute improvement? Most of the benchmarks I've seen show a ~30% improvement in CPU compute, while I/O has improved tremendously, as well as the memory. But again I've only seen ~30% improvement in CPU workloads, a far cry from "3x the performance" that I keep seeing.
    - BubRoss 7y ago
      You may have seen GPU comparisons. https://www.raspberrypi.org/magpi/wp-content/uploads/2019/06/2019-06-21-15_40_11-Window.png https://www.raspberrypi.org/magpi/wp-content/uploads/2019/06... https://www.raspberrypi.org/magpi/wp-content/uploads/2019/06/2019-06-21-15_41_55-Window.png https://www.raspberrypi.org/magpi/wp-content/uploads/2019/06...
      - megous 7y ago
        Speedometer (chrome) may be more dependent on mem bandwidth which is 2x of RPi 3.
        BubRoss 7y ago
        What makes you say that? Dynamic languages are usually chasing pointers which is something that out of order execution helps with tremendously.
        megous 7y ago
        This is supposedly some TODO app being run repeatedly in Chrome, so it's much more than a JS benchmark. And Chrome as a whole is memory hungry.
- beatgammit 7y ago
  For my use case, the main attraction for me in the RPi4 is the additional RAM and better IPC/ISA, not necessarily the higher clock (though higher clock is certainly welcome).
- Dylan16807 7y ago
  Tripling the surface area doesn't help? And those fins should help convection in most orientations.
  - megous 7y ago
    You're not trippling the surface area. (Part of) the board already serves as a heatsink. You need to count that in.
nielsole 7y ago
So some data is now in(Obviously with the old firmware, not the one mentioned in the article): Without fan, idle: 58°C peak With fan idle: 39°C (roughly 26°C ambient temperature) With fan 4x parallism of `yes > /dev/null`: 51°C peak The fan is a really small one, just for the CPU. And I didn't quite place in the center of the chip :/ Quite surprised of how much I overestimated idle temperature. 58°C feels really hot when you touch the SoC.
- penagwin 7y ago
  Awesome thanks for the data! > 58°C feels really hot when you touch the SoC. This obviously depends on the heat conductivity of the surface, but from my experience with my 3D printer, ~55C/60C is about the cutoff for pain tolerance for me. Just below that temperature is holdable, after that temperature is when it's painful to touch.