Raspberry Pi 4 temperature

[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.

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

True. But it also gives irrelevant data when it does throttle. ie if you insist on no throttling when running CPUBurn you almost certainly WILL need extra cooling, whereas you can quite happily run a Pi4 without any extra cooling. So in that case it possibly making you buy more kit (heatsink, fan) you don't actually need for your ACTUAL use case.

It's a bit like putting race brakes on a road car, because when someone did test on track, the road car overheated the brakes. But you only ever drive on the road....

[gsgs]

come on, it is one possible feature of the pi to do intensive calculations.
The reason why some are building clusters from pis,
I use it for chess analysis, calculating days, weeks (months,years ?) at full load
on a chess position.
Or on math-problems.
Or puzzles.
That's where the pi4b is good : massive calculation with low energy cost
(and cheap hardware)

[jamesh]

come on, it is one possible feature of the pi to do intensive calculations.
The reason why some are building clusters from pis,
I use it for chess analysis, calculating days, weeks (months,years ?) at full load
on a chess position.
Or on math-problems.
Or puzzles.
That's where the pi4b is good : massive calculation with low energy cost
(and cheap hardware)

Do you use NEON for those? Because unless you do, cpuburn is not a good representation of your workload.

Of course, if you are spending the time writing NEON optimised multiple processing multinode code for use in a cluster, then some sort of cooling system is almost certainly within budget.

[duren]

True. But it also gives irrelevant data when it does throttle. ie if you insist on no throttling when running CPUBurn you almost certainly WILL need extra cooling, whereas you can quite happily run a Pi4 without any extra cooling. So in that case it possibly making you buy more kit (heatsink, fan) you don't actually need for your ACTUAL use case.

It's a bit like putting race brakes on a road car, because when someone did test on track, the road car overheated the brakes. But you only ever drive on the road....

Happily run a pi4 without extra cooling?

I have the pi 4 with the official case running kodi 18 showing minimal CPU load and the thing gets to 70 C.

I don't think this is healthy for the pi nor any plugged in USB component.

Can we please get an update on what else is being done to lower the temperature via the firmware?

Thank you

[ProDigit]

CPUburn is excellent for testing temperatures,
Especially since gaming and emulation will come pretty close.
The CPU is packaged on the same die as the GPU, and with programs like blender and stress, only the CPU is stressed, not the GPU part.
Playing games like quake at high resolution, and high graphics, will also tax the GPU, which will result in higher heat than just CPU stress tests.

But yeah, if CPUburn keeps it at 66c, means the cooling solution is sufficient.

On my Pi3B+ with stress I have 66C on a different cooling solution. With CPU burn it hits 71-72, at which it reaches throttle temperatures.
In that case scenario, the CPU will be throttled, and one can't count on maximum performance.

I just wanted to make sure the same wasn't true here.

[rpdom]

True. But it also gives irrelevant data when it does throttle. ie if you insist on no throttling when running CPUBurn you almost certainly WILL need extra cooling, whereas you can quite happily run a Pi4 without any extra cooling. So in that case it possibly making you buy more kit (heatsink, fan) you don't actually need for your ACTUAL use case.

It's a bit like putting race brakes on a road car, because when someone did test on track, the road car overheated the brakes. But you only ever drive on the road....

Happily run a pi4 without extra cooling?

JamesH has been doing just that for quite a few months. He has been using a Pi4B as his main desktop computer since early this year (at least).

I have the pi 4 with the official case running kodi 18 showing minimal CPU load and the thing gets to 70 C.

I don't think this is healthy for the pi nor any plugged in USB component.

Why do you think that? I believe one of the other engineers said that they had been testing a PI 3B+ and Pi 4B at around 120°C and it had no problems. The reasons for capping it at 85°C are to prolong the life of the Pi and to prevent serious burns if someone touches the SoC when it is running hot.

I used to run several AMD based PCs and the default BIOS overtemp warning was at 90°C and could be set as high as 120°C. That was with large heatsinks and fans. They did sometimes get that warm as I was pushing them quite hard. That was a single core CPU.
The Pi 4B is close (if not better) in performance to those machines and runs cooler with no heatsink. That is impressive.

[jamesh]

As stated so many times, running at the "high" temps that you get before throttling does not harm the Pi, they can easily run at over 100degs, but we don't let them, too dangerous for small fingers and there is a slight decrease in MTBF (from 35 to 30 years or something like that).

Also as stated, CPUBurn is a pathlogical case - very few real world scenarios comes close io what it does, and games are not one of them.

As for firmware changes that reduce temperature, the new VLI firmware drops it by a degree or two, not sure what else we have planned.

[ProDigit]

I think aside from getting burned, are there any capacitors on the board, that could suffer leakage due to high heat, or does the pi only use solid capacitors?

There's also the mental aspect, of running 50C is acceptable for hardware, running 100c will boil water. Most people prefer to run it cooler.

Overclocking also can be done higher with cooler running chips.

And also, probably the idea that cool running chips use less power (which there's some truth to it, but not entirely correct).
People love devices that barely use any power to run.

[gsgs]

when it throttles, it's not just slower but also the measurements are deranged.
It becomes unpredictable. It depends on the ambient temperature.
It's not easily reproducable.
Often you measure the running time of a program and predict
the complexity of the task from it.
When you run a chess-tournament stockfish against Lc0
-say 1000 games- the performance will vary strangely over time
because of throttling. Stockfish mainly uses the CPU
while Lc0 mainly uses the GPU.

Other than the Android smartphones and tablets the pi is especially
for education, for science with the math software packages and such.
It's been used in other devices like chess-clocks, robots, weather stations,

[Gavinmc42]

This has of to be one of the most viewed posts about the Pi4, why?
0-70C is standard, -20 to +85C is industrial temp range, -40 to 125C is automotive.
More or less, some have wider specs.
These are the ranges designers design their product for.

I think it has been said before most parts are Industrial temperature range up to 85C.
Anyone who says running above xC is risking burning and damage to Pi's?
Such FUD should not be allowed.
Yes you can get a burn on the finger at 65C if you leave it there long enough.
Are there people stupid enough to do that? Probably,.
Will they turn around and sue RPF for making dangerous products.?
"Warning contents maybe hot"?

It takes maybe a second to pop the top off the Pi4 official case and if you have a small heatsink on it then it runs mid 50'sC. Any fan will cool it below 50C.

So much noise about normal CPU operation conditions.

[Imperf3kt]

Yes you can get a burn on the finger at 65C if you leave it there long enough.
Are there people stupid enough to do that? Probably,.
Will they turn around and sue RPF for making dangerous products.?
"Warning contents maybe hot"?

Any such law suit would immediately fail as the Pi comes packages with a sheet of anti-static cardboard that has some warnings printed on it. One of them is to not touch the SoC

[stanwhite]

Greetings,

It may be that the Pi4 is happy to run at just below 80C, but it is not going to be a great fan of thermal cycling, going from maybe 15C to 80C and back again when switched off. Remember that things expand when heated, and this produces mechanical stress.

Switch it on, and leave it on forever, and you probably won't suffer reduced lifetime, but continued thermal cycling will shorten its life span, the bigger the difference between hot and cold, the worse the stress.

My Pi4 runs cool. VERY cool. As per previous post, at 2-3C above ambient. It is 24C in here right now, and the Pi is at 27C. A major cooling effort is the reason, see my post on page 17 of this thread. I didn't have a picture at the time, but I do now.

Regards,

Stan.

Pi4BV3.jpg

[stanwhite]

Sorry, attachment photo did not show? I'll try again.

Regards,

Stan.

[alphanumeric]

How does that heatsink contact the SOC?

[jamesh]

Sorry, attachment photo did not show? I'll try again.

Regards,

Stan.

Slight overkill methinks.

[LTolledo]

Slight overkill methinks.

nah... its definitely an overkill!...in huge proportions!

well if that setup can maintain 29C in a 28C ambient with the 2GHz overclocked RPi4B running some stress test for about 4 hours... then maybe its commendable...

[PiModules]

Hi,
If you run PiCoolFAN 4 at 25%-50% of speed (ultra low noise) the on such environment it is fine. If during the time your application requirements increases temperature, the FAN speed will increased automatically, and when needs drop down, fan reduce their speed. You can use it also with a small heat-sink below.

See below, I tested here and working very good, so offering as an alternative solution, to big heat-sink. Here, with ambient temperature of 30 Celsius. I think that a big heat-sinks are not the solution, it is better to have combination of a small (low profile) heat-sink and low speed (therefore very low noise) FAN.

https://pimodules.com/download/picoolfan4-user-manual

https://pimodules.com/product/picoolfan4

Kind Regards
Ioannis

[stanwhite]

A solid block of aluminium is screwed to the bottom of the heatsink, and it gently presses down on the CPU. This block is 30mm square, (not 40mm as I erroneously posted earlier|), and has arctic silver 5 thermal paste between the CPU and the block, and between the block and the heatsink.

Yes, it is overkill, but thermal cycling is reduced to a minimum, which is a great benefit. I gave it a stress test, (results on P17 of this thread).

It IS big, its NOT pretty (well to me it is )

The fans are 12V, powered from a small boost inverter run from the Pi's 5V. There is also a 30mm fan keeping the inverter happy

Regards

Stan

[ProDigit]

I think the majority of people will find temperatures of 40-65c acceptable.
Aside from a metal case, or active cooling,
You can further cool the pi, by squeezing in thermal pads
or thermal tape, (depending on how much space you have) between the bottom of the board and preferably a metal case. As well as cooling most large chips on the board, save for the broadcom one.

[bensimmo]

if it failed due to thermal stress, then it would be under warranty. why worry.

my only two (and a half) problems are the same from the start of this thread.
with children, schools.

the board is too hot when not cased and in use.

when in a case (e.g RPi case) is over heats and throttles a lot reducing it useful processing power. (not always a problem)

When you want to use the gpio in schools, it needs to be out of the case, you cannot use a fan as that uses the gpio pins and cover the area for Hats and plain gets in yhe way.

it would be really nice if there was a Pi4 Lite.
sort of cool and usable with HATs and out of the case.
the processing power can be lower, as long as it can still run the usual stuff. just without the heat.

If the A72s are not capable of running cool under use, no matter what, then that's a shame.

No doubt that's a small use case as actual use within Education or CodeClubs etc is no where, i am assuming, near the general user/enthusiast and maker community size (ignoring industry).

Which begs the question i don't and haven't seen the answer to.
How slow can the Pi4 go before it becomes unusable?
Webbrowsing and email, office.
Scratch3 with a SenseHAT on.
Media

but not to worry, mines working away fine, two monitors now and is useful as a desktop with a fan gently blowing air over.
and i have my pi3 series for accesible gpio work.

Can't wait to see what they come up with next

[pfletch101]

it would be really nice if there was a Pi4 Lite.
sort of cool and usable with HATs and out of the case.
the processing power can be lower, as long as it can still run the usual stuff. just without the heat.

Isn't that called a 3B(+)?

[bensimmo]

it would be really nice if there was a Pi4 Lite.
sort of cool and usable with HATs and out of the case.
the processing power can be lower, as long as it can still run the usual stuff. just without the heat.

Isn't that called a 3B(+)?

at the moment-ish,
but you do loose the newer Pi4 capabilities.
like dual monitor, multiple uart, usb3, full Gbit Ethernet and that very useful 2GB of RAM.

[ProDigit]

Since you have a quad core, you'll want the speed of each core to be ~1,5Ghz for snappy response, and ~1,2Ghz for acceptable response.
Anything below 1-1,2Ghz gets a sluggish GUI experience.
It at least has the benefit of faster RAM, but the Pi Zero W only operates acceptably at 1.1Ghz.

[jamesh]

Recent (today) change to bootloader has a tweak to SD RAM timings which will give a slight power saving at idle, which will help very slightly with temperatures.

[martinrowan]

Recent (today) change to bootloader has a tweak to SD RAM timings which will give a slight power saving at idle, which will help very slightly with temperatures.

James is that released, or just internal change being tested?
Thanks
Martin