Raspberry Pi 4 temperature

[boznz]

I suspect anyone working headless for IOT, Industrial or robotics will care more about the system running cool than its performance. Our industrial controllers have been working flawlessly for years with the RPi2 & 3, however RPi3's are quickly being sold out and unfortunately it will need to be redesigned for the RPi 4 due to the bigger power requirement and thermal incompatibility with the current industrial enclosure. Luckily the thousand or so we have out there do not fail often if at all.

This is not actually a whinge because I know the RPi is basically a hobbyist board we have hijacked for our own use and I am pleased the new boards retain hardware GPIO compatibility with older ones. I also suspect someone will crack the under-clocking problem before our stock of RPi 3's run out

[Imperf3kt]

The Pi3 models will still be produced for many years

[jamesh]

Just a quick one to say there is currently no soft throttle on the Pi4 for technical reasons.

[graysky]

Just a quick one to say there is currently no soft throttle on the Pi4 for technical reasons.

Does that mean the RPi4 will not reduce clock speed or voltage based on temp? I read (perhaps incorrectly) that a temp of 80C causes a throttle in either CPU speed or voltage to CPU.

[jdb]

"Soft Throttle" was implemented on 3B+ because the 1.2GHz operating point had a significant power saving compared to the 1.4GHz operating point - both frequency and core voltage could be reduced. The frequency reduction is only 15% in this case.

On Pi 4, the first step down is 1GHz (33%) with no voltage reduction, so a soft-throttle wouldn't save as much power and necessarily has a bigger performance impact.

The Pi 4 will still throttle in stages at >80C - with discrete steps of 1GHz/750MHz/600/500 as temperature increases.

[B.Goode]

RPi3's are quickly being sold out

Is there any reason to think that reputable suppliers will not restock them?

Obsolescence Statement
Raspberry Pi 3 Model B will remain in production until at least January 2026

Ref: https://www.raspberrypi.org/products/ra ... 3-model-b/

Raspberry Pi 3 Model B+ will remain in production until at least January 2026

Ref: https://www.raspberrypi.org/products/ra ... el-b-plus/

Raspberry Pi 3 Model A+ will remain in production until at least January 2026

Ref: https://www.raspberrypi.org/products/ra ... el-a-plus/

[andrum99]

Raspberry Pi 2B will also remain in production until at least January 2026, although it is now using the newer BCM2837B0 SoC, rather than the 2836 it originally had.

Link: https://www.raspberrypi.org/products/ra ... 2-model-b/

[jamesh]

Raspberry Pi 2B will also remain in production until at least January 2026, although it is now using the newer BCM2837B0 SoC, rather than the 2836 it originally had.

Link: https://www.raspberrypi.org/products/ra ... 2-model-b/

Some of them, not sure all are yet made with the 2837, depends on whether Farnell or RS make them I believe.

Older devices are made in batches, so will sometimes suffer brief shortages if we have a lot of demand on, for example, the Pi4, which uses up all the production capacity.

[shadowownz]

well, not sure if it helps but here's my 2 cents: got a pi4 4gb, armor case dual fan and some decent thermal pads (Arctic 1.5mm to replace the 1mm ones, but 1mm can be used) and ran the cpubench for around 20min. Highest temp i saw was 62º C. So i gotta say, those CNC plates are definitely doing their job well! Oh and just by replacing the off brand thermal pads that came with the case, it's a 3-4º C difference on normal load (youtube 1080p, web browsing, updates and installs,...). Will try thermal grizzly minus pads 8 (8W/mK > 6W/mK from Arctic, should be a 1-2º cooler i assume) in a couple of weeks. Anyone has those to share some results?

EDIT: Oh forgot that i have over_volt=4 and cpu_freq=2000.

[ProDigit]

well, not sure if it helps but here's my 2 cents: got a pi4 4gb, armor case dual fan and some decent thermal pads (Arctic 1.5mm to replace the 1mm ones, but 1mm can be used) and ran the cpubench for around 20min. Highest temp i saw was 62º C. So i gotta say, those CNC plates are definitely doing their job well! Oh and just by replacing the off brand thermal pads that came with the case, it's a 3-4º C difference on normal load (youtube 1080p, web browsing, updates and installs,...). Will try thermal grizzly minus pads 8 (8W/mK > 6W/mK from Arctic, should be a 1-2º cooler i assume) in a couple of weeks. Anyone has those to share some results?

EDIT: Oh forgot that i have over_volt=4 and cpu_freq=2000.

Would be nice to see if you could use the stock thermal pads in combination with thermal paste, to see if that improves the heat transfer?

[Ernst]

Would be nice to see if you could use the stock thermal pads in combination with thermal paste, to see if that improves the heat transfer?

Do you understand the purpose of thermal paste ?

[ProDigit]

Would be nice to see if you could use the stock thermal pads in combination with thermal paste, to see if that improves the heat transfer?

Do you understand the purpose of thermal paste ?

The cases need pads, because they have a gap between the CPU and cooling solution. One can't just use thermal paste with them.

[pfletch101]

Would be nice to see if you could use the stock thermal pads in combination with thermal paste, to see if that improves the heat transfer?

Do you understand the purpose of thermal paste ?

The primary purpose of thermal paste is to fill in small irregularities in two imperfectly mating surfaces so that there are no air gaps to impede heat transfer between them. The primary purpose of a thermal pad is/should be to fill in what would otherwise be a larger gap between a source of heat and a heat sink or other 'device' which is intended to cool it and/or to act as part of an adhesive connection between the two. While one doesn't generally want/need to use both, they are not doing exactly the same thing, so one can conceive of circumstances under which one might (e.g.one or both of the surfaces being connected is unusually rough or uneven).

[PeterO]

If pads are sticky they can be used to secure a heatsink to a device.

I doubt a pad would stick if one (or both) surfaces had paste applied but I've not got any pads here to waste trying it out.

PeterO

[pfletch101]

Pads are sticky and can be used to secure a heatsink to a device.

I doubt a pad would stick if one (or both) surfaces had paste applied but I've not got any pads here to waste trying it out.

PeterO

Don't generalize from limited experience! Pads may be sticky and act as an adhesive connection between two components. They may, on the other hand (and the pads supplied with Flirc and similar cases are an excellent example of this) simply fill in a gap between two components that are positioned and secured by other means. Non-sticky sheets of thermal material are an ordinary item of commerce.

[PeterO]

Pads are sticky and can be used to secure a heatsink to a device.

I doubt a pad would stick if one (or both) surfaces had paste applied but I've not got any pads here to waste trying it out.

PeterO

Don't generalize from limited experience!

Fixed it !
PeterO

[martinrowan]

For those that might be interested in looking at temperatures of the Pi 4 in various cases under varying CPU loads and looking at thermal throttling, I've just posted an update to my blog with my latest test results:
https://www.martinrowan.co.uk/2019/09/r ... nder-load/

I've also recently looked at the impact on WiFi Performance of various case choices, something again which may be of interest to some: https://www.martinrowan.co.uk/2019/08/w ... i-4-cases/

Not convinced that burying this information in an 18-page topic is the best idea. But at least it's out there so that it might help someone that stumbles over it.

[shadowownz]

well, not sure if it helps but here's my 2 cents: got a pi4 4gb, armor case dual fan and some decent thermal pads (Arctic 1.5mm to replace the 1mm ones, but 1mm can be used) and ran the cpubench for around 20min. Highest temp i saw was 62º C. So i gotta say, those CNC plates are definitely doing their job well! Oh and just by replacing the off brand thermal pads that came with the case, it's a 3-4º C difference on normal load (youtube 1080p, web browsing, updates and installs,...). Will try thermal grizzly minus pads 8 (8W/mK > 6W/mK from Arctic, should be a 1-2º cooler i assume) in a couple of weeks. Anyone has those to share some results?

EDIT: Oh forgot that i have over_volt=4 and cpu_freq=2000.

Would be nice to see if you could use the stock thermal pads in combination with thermal paste, to see if that improves the heat transfer?

You could, but you won't have any gains, on the contrary. As someone said, the purpose of pads vs paste is different but the goal is the same: transfer heat from the SoC / chips to the heatsink in order to be dissipated. Thermal paste USUALLY has higher conductivity (around 12W/mk for a good one) than pads (check fujipoly pads for that matter) but may not apply to every case. There's a kit for Pi 4 with a heatsink & fan, just like a normal computer, that uses screws to attach it in place, so my educated guess is that probably it uses thermal paste because the gap between heatsink and SoC is very small. But for the majoraty of the pi 4 cases, there is a small gap between the SoC and the heatsink of the case and thermal paste in that case wont be any good because it doesnt have the thickness required to bridge both the SoC and the case heatsinks. But IMO, there wont be much difference between a thermal pad of 8W/mK and a paste of 12W/mK, maybe 1 or 2 degrees Celsius tops. Oh and there's a thermal conductivity property stating that the use of 2 different conductivity elements IS ACTUALLY WORSE than using one of them alone (thermals engineer / physicist needed here!). In short the more stuff you put between your heat source and heatsink = less and less heat transfer effectively.

[LTolledo]

I dont use those "pads" over the SoC, as I find those inefficient in heat transfer.

in place of the pads, I use copper sheets/blocks, the thickness of which is dependent on the measured gap, less 0.1mm. Thermal adhesive/paste is used to secure the copper sheet/block to the heatsink. thermal grease between the copper block and SoC.
something like this

RPi4B-4G_Armorcase-open.jpg (200.56 KiB) Viewed 2475 times

[ProDigit]

For those that might be interested in looking at temperatures of the Pi 4 in various cases under varying CPU loads and looking at thermal throttling, I've just posted an update to my blog with my latest test results:
https://www.martinrowan.co.uk/2019/09/r ... nder-load/

I've also recently looked at the impact on WiFi Performance of various case choices, something again which may be of interest to some: https://www.martinrowan.co.uk/2019/08/w ... i-4-cases/

Not convinced that burying this information in an 18-page topic is the best idea. But at least it's out there so that it might help someone that stumbles over it.

Seems like the passive radiator (the one which I also use on my Pi 3B+) is a very good (passive) solution that prevents throttling altogether, and allows the Pi 4B to run at it's full potential!

Can you test CPUBurn on that aluminum radiator case (without the fan)?
CPUBurn will heat your system much more than stress does.

[martinrowan]

Can you test CPUBurn on that aluminum radiator case (without the fan)?
CPUBurn will heat your system much more than stress does.

I'm currently testing another 3D printed case at the moment. But will look into CPUBurn. Though at the same time I think so long as the test is the same across all scenarios be it idle, stress, CPUBurn, overclocking etc. They will all be relative and provide the same indicative outcomes. Though I'll admit until I do such test I can't be 100% sure. So maybe I'll just do a subset of tests to confirm the correlation, as testing takes time and like most here I'm doing this for fun and gaining knowledge for my own uses cases and sharing that information freely.

I personally think my testing of idle (older post) and the 1 core stress are likely to be more indicative of most of the scenarios, rather than the plethora of tests under full load and overlocked scenarios. But maybe I'm wrong?

Martin

[martinrowan]

Can you test CPUBurn on that aluminum radiator case (without the fan)?
CPUBurn will heat your system much more than stress does.

To see what difference CPUburn made compared to stress, I left CPUburn running overnight and the maximum temperature was 66°C, compared to 63°C when using stress. So there is a small difference, at least with this radiator case design.


The decrease in temperature over time is just ambient temperature reducing. from 23 to 21°C

[jamesh]

I've said it before, but things like cpuburn are a pathological/extreme/very worst case. It's very rare to find that sort of workload on a device. It's loading up all four cores, with NEON running full tilt. Normal desktop use won't do that, or anything near it. The only use case I can think of that might get close,and still be relatively common, is a highly NEON optimised, multithreaded codec.

[gkreidl]

I've said it before, but things like cpuburn are a pathological/extreme/very worst case. It's very rare to find that sort of workload on a device. It's loading up all four cores, with NEON running full tilt. Normal desktop use won't do that, or anything near it. The only use case I can think of that might get close,and still be relatively common, is a highly NEON optimised, multithreaded codec.

My real world test case has always been converting a video with HandBrake.

[Imperf3kt]

I've said it before, but things like cpuburn are a pathological/extreme/very worst case. It's very rare to find that sort of workload on a device. It's loading up all four cores, with NEON running full tilt. Normal desktop use won't do that, or anything near it. The only use case I can think of that might get close,and still be relatively common, is a highly NEON optimised, multithreaded codec.

Which is precisely why it is useful - if it won't throttle while doing that, it'll never throttle under sane workloads.