Too hot to handle? Raspberry Pi 4 fans left wondering if kit should come with a heatsink Some early adopters of the Raspberry Pi 4, released on 24 June, are running into heat issues, especially with the official Pi 4 case making no provision for a heatsink or fan. The Raspberry Pi 4 has a 1.5GHz quad-core 64-bit Arm Cortex-A72 CPU, for approximately three times the performance of the previous model. That …
Something like the new FLIRC case ?
Looks very nice, but I don't think it will work with a PoE HAT.
I've got one for a Pi 3B+ and it works well but on a Pi 4 temps start at at about 70ºC and throttling starts at 80ºC so I don't think that case will do much for it.
a Pi 4 temps start at at about 70ºC and throttling starts at 80ºC so I don't think that case will do much for it.
The FLIRC case for 4B is not available yet, it's still a pre-order product. I would guess that they are still testing and honing. So it mght differ quite a lot from the 3b+ version and the pre-order photo. Maybe even have fins or mounts for further passive cooling and fans if they have a performance target to achieve.
I have just normal heat sinks on my 3B+, which is used for Pi-Hole duties and that is more than enough, but it isn't really pushed most of the time. A 4B doing the same job would probably get away with heat sinks as well - it runs headless and, depending on the number of clients on the network, doesn't get pushed hard - but if it is being used for media player or as a general PC, I'd definitely look at a proper cooling solution for it.
Mine's the flame retardant coat in the corner.
Same thoughts. Plus most of these comparisons are sensitive to the units used. People talking about "twice as hot as it was today" mean quite different things depending on whether they use Fahrenheit, Celsius, Remorqueur (however that's spelled) or Kelvin. Strictly this construction only makes sense in Kelvin. Here, the qualifier "with respect to ambient temperature" would be needed to get what is really meant here.
Yep, they are called welders :-)
I'm fairly sure that the 3D CAM versions of those is not in the hobbyist price range and an order from heat-sinks-r-us is probably a lot cheaper and easier.
Would be a fun toy to play with though - until the point you burn your garage down with it..
Indeed it's only worth while 3D printing to make a mould and then casting Aluminium. The 3D printed metal is poorer heat conductivity.
Actually carve one out of candle, pack in silver sand and pour in the aluminium or Al alloy. Then mill/grind the base and polish to mount on IC(s).
A 3D printed heatsink is a nonsense idea even if you can print metal. Maybe even cutting up an old PII or PIII heatsink is a better idea. I've kept a bunch of those to heatsink audio amps, linear regulators and RF amps. I've some 486 and GFX heatsinks too.
Icon: Dragon breath to melt metal in crucible.
Speaking as a manufacturing designer, extruded heatsinks (think car audio amplifier cases) are easier to manufacture than machined heatsinks ( think metal hedgehogs).
The only advantage to 3D printing a lost plastic part to cast aluminium is to to achieve a geometry with high surface area. Desirable, but probably not worth the faff just to achieve a bit more cooling in a certain space (or the same cooling in a smaller space). Cast parts will have a different grain structure to extruded or machined parts, but I don't know the effect of this on conduction.
Of course you can get a complicated geometry in other ways, such as bending copper pipe. And then you get the option of filling it with a suitable volatile liquid, using phase change and convection to cool to chip.
IMHO heatsinks are cheap and electronics generally like to run as cool as possible. If you're up for assembling a Pi then you can manage a few extra pennies and 30 seconds to drop in a couple of bits of ridged metal. It's currently 26C in my office, and my little Pi 2B which is idle at this time of day is reporting 47C with the lid on and 42 lid off. All nicely within spec of course, and a fan is probably overkill plus tiny fans make a noise out of all proportion to the volume of air wafted around.
What I really need is a heatsink for me...
[Tim Hartnell was] "well known for his attempts to keep the ZX80 cool by balancing a frozen carton of milk on top of it"
Back in the day I had a 386dx running at 33Mhz (I think, all a bit hazy now) and I could set it to run at 40mhz via dip switches but it got astonishingly hot, this was in the pre heatsink as standard era so it was just a flat topped chip on the board.
One day I stumbled across a ~15cm piece of machined aluminium and an idea struck me. I superglued it straight onto the top of the chip and set the dips to 40mhz.
Worked like a charm. I got my extra 5 MIPS and I didn't blow the chip up.
This week is bad, for any computer.
I'm not sure how accurate the sensors/software are on my box, but it's running around 10℃ hotter than usual, and I am not planning to transcode any video files until things are cooler. Overnight temperaturs are forecast to be plenty high enough.
Doesn't matter whether it's global warning or a freak year, when it's this hot, computers have problems. In that, the Raspberry Pi 4 isn't really all that special. But, on tests I have seen reported, I think a fan is needed. A heatsink alone doesn't make much difference.
I had to put a heatsink on my original Pi Model B back in the day.
Incidentally the lowest power Pi seems to be the Zero W, not had a single problem so far.
Runs warm but passive air cooling seems fine.
Why aren't there "micro" fans yet? You'd think a 3D printed fan would be trivial to make, using
a simple hand wound coil and beam break detector for both position feedback and stall detection.
No Hall sensors needed so loads cheaper and can work with a simple pair of LEDs run alternately
as both detectors and emitters, due to not needing to worry about spin direction.
A quiet fan is limited by physics. The amount of air you can shovel around is roughly proportional to the area (diameter squared) and speed. The smaller the fan the higher the speed and the noisier it gets. Plus you need proper bearings that need to be installed properly. I'm pretty sure that this is not an application for 3D printing...
So very large fan = low speed and quiet.
I did wonder whether a distributed maglev (tm) fan might work, with pyrolytic graphite around the outer edge to add positional feedback.
For that matter why not use a 3D printed longitudinal fan? Pressed in manufactured bearings at each end, the whole thing would have massive air flow directed onto the chips but only a small footprint.
Optimise the design so that it uses minimal material, perhaps a "double helix" structure? Should even work with a pager vibrate or drone motor if you wanted to go down the low power route.
Also see https://www.digikey.in/en/product-highlight/s/sunon-fans/me-series-maglev-fans
Large? Nah. this is large.
Joking aside, it annoys me no end that a fan is needed - I rather liked the lack of noise.
I'm a fan of fanless, so to speak.
I did wonder whether a distributed maglev (tm) fan might work, with pyrolytic graphite around the outer edge to add positional feedback.
Back in the day of early Athlons I had a couple of CPU coolers that basically had the motor's magnets in a ring on the circumference of the fan and the drive coils stationary around them. This allowed, according to the marketing blurb, to have just a small bearing at the center of the fan and the fan blades to extend inwards much further than with a hub motor, allowing airflow directly to the heatsink core sitting over the actual CPU chip. Never mind that the airflow near the fan center would be quite minimal anyway and hence add next to nothing to the overall cooling, but the fans were pretty quiet for the time.
Unfortunately the designers didn't sufficiently account for the radial forces acting on the magnets; the fans all died due to them becoming unstuck
This video shows someone's fan mount for a Pi 4. It works and stops throttling but it's a shame to have to stick a ruddy great fan on it.
I watched that the other night. It still makes a very small device.
When I get a Pi 4, it will get a fan and airguide heatsinks like he showed on YouTube - but possibly not exactly the same case
One question though. There was something he said that made me think the fan was pulling air in. I would have expected an exhaust fan there. Which is it?
I think there's plenty of debate in general as to whether fans are better sucking or blowing?
My Pi 3B+ has a case fan that sucks air out of the case, which I can understand is good in terms of actually removing heat away from the chips, however I could also configure it to blow air in. As it's positioned directly above the CPU, I would assume that a constant flow of air aimed towards the heatsink would be good in that it would cause a lot more disturbance in the airflow hitting the fins of the heatsink , and so ultimately drive more heat away from the source.
I'm genuinely curious as to the best method. Computer CPU fans still, if I recall correctly, push air into the fins of the cooler rather than draw warm air away?
Personally, I figure that it doesn't matter until it matters. That is, it doesn't matter which way the fan is blowing. All that matters is that the cooling system is keeping the temperatures within specifications. My gut (and experience) tells me that for the vast majority of installations, a fan will achieve that regardless of which direction it blows.
I can imagine situations where air direction might actually be a critical consideration, and I'm guessing which way is best depends on the hardware. If, for instance, the heat issue is overall buildup in a case rather than a single problematic chip, then sucking is probably best. If the heat issue is a specific chip, then blowing is probably best.
"That is, it doesn't matter which way the fan is blowing."
That's probably true in 99% of cases. I've found that in a dusty environment, a fan that blows into the case is better because then you only need a filter across that one inlet. If it's sucking out, then every gap in the case draws in dusty air.
Well, I found a fan that is moderately quiet and will cope with even the heaviest loads - but you still need a heat sink to maximise the surface over which heat is transferred. It's very inefficient to use a fan straight on the CPU, there's not enough surface.
Having created conduction cooled boards in the dim and distant past they tended to be a bit thicker than that to get the copper and heat conduction layers in. Plus they were usually firmly strapped to something at the edges to dump the heat into, pcb foil layers not exactly being renowned for having massive heat capacity. So I'd assume the 'heatspreading' on this isn't adding much at all.
Hope something useful and cheap pops up to sort this out.
I got the 4GB one on launch, and after a week ordered the FanShim from Pimoroni to stop it throttling itself, because it was almost constantly bumping up at the limit (just running some containers, headlessly, albeit in a warm cupboard). But no: because it uses the whole board to (very inefficiently) dissipate heat, the whole board only seems to get slightly warm (YMMV). Even the processor itself didn't seem to get hot enough to cause a burn, unless one were to find the most sensitive patch of a baby's very sensitive skin and hold it in place for a while. I'm not sure there's much mileage in going to court with that sort of use-case.
Even at 90 degrees, the specific heat of the top of the chip isn't high enough to cause a burn without waiting a while with your finger on it. It would feel uncomfortable, but you'd be fine. The board, probably not. If you're looking for a recipe for the thing to get damaged, running the chip at a temperature above spec for a long time or putting it through loops of heating up a lot, throttling until the temp goes back down past another limit, and letting it sail back up would both be good options.
Been running a Pi kiosk in a museum which has plenty of fluorescent lights for two years. Because it's a wooden structure they kill the power each day at the main breaker. Even though the Pi gets a hard shutdown it has been booting up OK with no issues when the power is switched back on.
Maybe you should buy better SD cards.
If you buy your SD cards from the local market then power cuts will not corrupt the data - they got corrupted when you first put some data on the card. My first two Pi's were ordered in 2012. One died from a poor quality power supply. The other is still going with its original SD card and no UPS. Years ago I had problems with defective flash. Sort out a competent supplier and the problem goes away.
Also depends on use. I have several Sandisk SD cards in various Pis that have been going strong for over 4 years, with me regularly pulling the plug without safely shutting down, and one that corrupted after only one power failure while running openhabian (which seems to constantly read and write to the card - I replaced it with my own homebrew scripting and have had no such problems). But yeah, as long as you're not intensively reading/writing, I've found the SD corruption thing to not really be as big of an issue as some make out.
In my experience, the SD corruption issue was terrible with the first models, although I think that was mostly software. Every time the device was shut down improperly, there'd be problems and usually the easiest thing was rewriting the card. Then 2013 or so rolled around and that stopped happening. I wondered about that so I subjected a victim pi to a cycle of power cuts at random intervals, including during writes and boot, but it survived all of them. I'd suggest you retrieve your pis, update the OS, and try again.
These A72 cores pump the heat.
My Nanopc-t4 has a cpu overheating problem with its moderately sized heatsink (no fan).
I have to stand it on its side to improve convection air flow to cool it, will need a fan in summer.
The heat is also generated by the power supply circuitry.
Of course in the spirit of all things Pi where technical coolness is more important that cost or practicality....
Peltier cooling elements are available from your favourite tat bazaar. At 5v / 1.5 A you'll need another USB power feed - but who isn't feeding their Pi from USB hub these days? At at 30 x 30 mm it's bit on the large side so a nice copper block riser (and thermal paste) might be needed to lift it clear of other components, but then you'll need to get a thermocouple in there to control it as a 65C temperature differential is likely to freeze your poor Pi. And possibly some thought given to controlling condensation to avoid embarrassing shorts.
A suitably large heatsink will of course be needed on the hot side of the peltier array.
It certainly won't get ------------------------------------------------------------------------------------------------------>
The trouble with using a Peltier is that it takes power to operate it, and so the heat sink and fan have to be much larger than for the device alone. Peltiers only make sense when it is necessary to get to below ambient temperature.
I remember a case where a CCD sensor was in a sealed box and needed cooling, so a Peltier was added between CCD and metal box. The result was that the CCD ran hotter.
Why bother with a heatsink when the fan for Pi 4 from Pimoroni works fine and is cheap. Can program it to come on at whatever temp you fancy (65C is suggested) and one can see it rapidly cooling the chip. So quiet you don't know it's cut in unless you notice the LE changing colour. Fits over the first few pins but they're stlill available.
This post has been deleted by its author
I have a Pi3 under my telly running OSMC, but the Leia builds seem to hit 100% CPU on a single core from time to time and stay there. I capped the CPU so it wouldn't bake. Runs fine for everything I need.
With the heat recently I've run it topless, but that's about it.
Why they put a spectre vulnerable CPU that was not in production (ARM Quote that no effected CPUs were in production) when the Spectre issue became piblic? Was it because they would have had the throw then in the bin otherwise?
Personally I would have preferred they stuck with a53s with the same RAM and SOC I/O throughput and didn't call it a desktop replacment just because it is atleast as vulneable as an intel desktop.
Because they are cheap. The board is /intended/ to be cheap. It is for school kids to muck about on and learn via taking risks without anyone needing to care too much if they break it. They always use chips that are out-of-date or binned aside for some other reason.
I imagine they were not even thinking about a Pi4 until Broadcom suddenly had a pallet-load of chips that were suddenly more difficult to sell. (Pure speculation there - I am not affiliated with Pi: I have played with a few, and encourage the not-IT students around here play with them too, but generally use more suitable - ake expensive - boards for my own projects).
"but generally use more suitable - ake expensive - boards for my own projects"
I mostly use Pis for toys and prototyping. I don't tend to use them in "serious" projects, primarily because they're too large (usually too tall) for the sorts of projects I tend to do. I have field-stripped all of the connectors and such off of Pis to reduce their profile, but that's a lot of work.
Instead, I tend to just use naked microcontrollers, or rarely a more serious SBC as you're talking about, depending on the project.
RPi foundation made a lot about it been capable of being a desktop replacement, so an amount of people are going to be using it for that purpose, including web browsing and online banking and without any antivirus.
Now either you believe that spectre is an issue or you do not, if the later then you might consider that the big names in CPU production do actually believe it is a problem, even intel is moving towards addressing their fail.
So when you say it is "Cheap" so it is okay then fine if you never trust it with anything important but when someone has their credentials hijacked are you willing to explain how it was all their own fault?
I actually have all but the last 2 versions, Pi 3b didn't seem worth the upgrade and the pi4 to me is just getting rid of vulnerable hardware and using a charity to do the selling. Shame on all involved.
Ceramic heatsinks pull heat away a little bit faster than aluminum ones, but that's not the reason that they are used. The benefits of ceramic heatsinks are that they weigh less and are nonconductive (so you can do cool things like print circuitry directly on the heatsink).
The primary downside is that they are more fragile.
hmm, one of my pi3's is in a nice alloy case that has a heat spreader pressed against the CPU chip.
now, my workload for it is very light weight, it runs a python script that sleeps for 2 minutes, then reads some weather data and updates a couple servers. so it runs very cool
ah, they have a pi4 version now...
https://flirc.tv/more/raspberry-pi-4-case
That's the beauty of it to you. The beauty of it to other people is that it's a teeny board with a lot of poke. Some people will want to use these as desktop replacements, so having a little fan is bearable.
Me, I'm in the middle. Some of mine are used for Kodi and others for MiniSatIP. MiniSatIP doesn't take much grunt and is in the attic, so no-moving-parts is nice there. Kodi takes a bit of power, but I like it to be silent, so to me there's a limit on how many Watts it can draw before it runs too hot. It may be that a Pi4, underclocked hard, could run all the 4k H.265 stuff from my UHD Blu-Rays without cooking, and that would be adequate for me.
Other people will have different use cases which are just as valid as yours or mine.
https://www.banggood.com/Aluminum-Alloy-Armor-Enclosure-Case-Metal-Shell-Compatible-with-Raspberry-Pi-3-Model-Bplus3B-p-1416646.html
or:
https://www.banggood.com/Aluminum-Alloy-CNC-Enclosure-Case-Metal-Shell-Compatible-with-Raspberry-Pi-3-Model-Bplus3B-p-1425732.html
or with fans:
https://www.banggood.com/CNC-V3-Aluminum-Alloy-Armor-Protective-Case-Dual-Cooling-Fan-Kit-For-Raspberry-Pi-3-Model-B-3BPlus-p-1456092.html
*As per the title: these are 2-3B(+) ones, but I'd put money on 4-compatible versions already rolling off production lines.*
Yep! Good spot, Olivier2553!
Just showed up on BangGood too. (Nothing particularly special about BG, but I can only justify time to tracking new arrivals on one such site and they have, so far, provided an e-shopping experience that hasn't outright driven me away!).
https://www.banggood.com/Armor-Aluminum-Alloy-Case-Protective-Shell-Metal-Enclosure-for-Raspberry-Pi-4-Model-B-Only-p-1539228.html
https://www.banggood.com/Black-Silver-Aluminum-Case-Enclosure-Shell-With-Cooling-Fan-For-Raspberry-Pi-4-Model-B-p-1539229.html
https://www.banggood.com/9-Layer-Acrylic-Case-Cooling-Fan-Micro-HDMI-Adapter-Kit-for-Raspberry-Pi-4-Model-B-p-1539218.html
and unrelated to Pi, but likely of interest to El Reg's primary demographic:
https://www.banggood.com/10-LED-Festival-String-Lantern-Hanging-Lamp-Party-Patio-Home-Decorative-Night-Light-p-1538527.html
:-)
I can highly recommend the Pimoroni 'Fan Shim'. My Pi 4 4Gb is in a PiBow Coupe case with the Fan Shim fitted and is now idling at 42°C (not using 4K). Before, I think it was approaching 65-70°C. They also provide a bit of code which can make the fan kick in and run when a certain temperature threshold is reached which works well. However, I decided to go without the code and just have it run it all the time as it is pretty quiet anyway!
Didn't the Raspberry group make pre-production samples? First they screwed up the power circuitry and now it's getting too hot?
Even when I design a board for personal use I always have a few samples made, if only to make sure the board stuffer can do it's job, and the stuffed PCB can fit in the enclosure.
The Raspberry group should remember a lot of it's users are D-I-Y people with very tight financial resources.
I'm not entirely clear if you're complaining because it's too cheap or too overpowered, but I'm certain that the Pi foundation will be glad for your advice to try out a few prototypes first. I certainly hadn't thought of that, and I doubt anyone else has either.
The fan shim seems to be better at cooling obviously compared to the heatsink, with or without FLIRC. Adding both a heatsink and fan shim makes it worse... https://blog.pimoroni.com/raspberry-pi-4-thermals-and-fan-shim/ - FLIRC doesn't eem to make any difference (https://www.phoronix.com/scan.php?page=article&item=raspberrypi-4-cooling&num=1), depends on your aesthetic needs. I use just a Pibow Coupé 4 with the Fan shim myself, it's totally silent
Also useful to note that there are small fans available.
I have 3 broken 4Bs on order, hopefully can do some experimentation and see what works.
Interestingly others have reported that if a bad unit has its memory card cloned, it can sometimes
work again. What gives?
Thinking card wear here because it generally happens with a working-but-slow card or one very close to
the minimum speed.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
