I get great thermals. Currently I put it in the official case (with drilled holes) and get about 25 over ambient with USB turned off and a conservative CPU governor.
Interestingly we (also a Raspberry Pi Approved Reseller) tested these heatsink cases with fans attached and found that the fans didn't improve thermals due to the manner in which they are mounted.
In the end we decided the extra cost, power consumption, noise, and moving parts simply weren't good value for the customer!
Adding fans could (and should) make a huge difference but this specific design was very poor at taking advantage of them.
Why does the image for the case featured at [1] have so many editing artifacts? and why is the image for the case linked from shopify [2] when the product is being sold at piaustralia.com.au? Seems kind of sketchy.
I took a look on amazon and it seems like there are quite [3] a [4] few [5] resellers offering what appears to be the same product or design:
With so many clones in the market it becomes difficult to trust anyone, which is frustrating to say the least. I would love some unbiased advice on what product to purchase for a raspberry pi 4 which will probably be playing a lot of games and running at high cpu often.
Here's a fun thing I just did with my RPI4. I've got "watch -n 1 vcgencmd measure_temp" running.
Equilibrium temp is arount 73C running a realtime website. Take a £2 coin and place it on the CPU. Temp drops 10C within seconds. I can get it under 60C by swapping a second coin. They stay pretty hot after you take them off. Also the equilibrium temp when you just leave the coin there seems to be a good few degrees less than with no coin. Possible the larger surface area helps remove the heat, just not as good as an actually designed heatsink.
When you lie the heat sink you have a rough surface of air bubbles. Super glue would fill those bubbles with a solid material which always has a better heat conduction than air
That's pretty interesting. I don't know why they didn't just include a small heatsink, it can't cost more than 25 or 50 cents and it improves the performance significantly by cutting down the thermal throttling.
Both times I’ve bought raspberry pi’s from Canakit.com they included adhesive heat sinks. But these were added by the vendor. It certainly seems like they should come with the pi from the factory.
But then anyone who wants to use an aftermarket cooler will need to peel off the stock cooler first. I don't think that's consistent with the Pi philosophy, not to mention you might damage the SoC if you apply too much force while removing an adhesive heat sink.
I have tried many options (including powerful fans), but the Flirc case [1] (whole-case passive heat sink) is my favority so far and cools well enough for a 1.75 GHz overclock under torture testing (2+ GHz would crash under torture).
[1] https://flirc.tv/more/raspberry-pi-4-case (I'm not affiliated, just a content customer).
Might as well go all the way to AstroPi[1] at that rate. Though I wonder if Flir couldn't make a copper-top version for less money than either, really...
In which way is this "all" more than the flirc case? The Flirc case makes physical contact with the CPU and radiates heat all over the case. Don't let the black (removable) plastic top fool you - it's one solid piece the whole way.
Having recently built a picade with a Pi4 and ran it in a cool room for around 2 hours, the device kept crashing when playing various emulators. After the final crash, it didn't start up again. Opening up the back was like opening the door to an oven and the Pi4 was just steaming. Nothing was obviously melted, but it didn't work again :(
This seems conflicting with what I read about its behavior. It should heavily throttle (underclock) itself, but shouldn't destroy itself. Maybe you got a defective unit or the hot thermal environments caused issues elsewhere (again maybe due to some defect).
I had a power supply with similar problems which actually started on fire - I had been restarting it repeatedly after it kept (supposably) hitting thermal or some other shutdown condition.
Lesson: just because there are protections for going past limits doesn't mean that everything is fine when you are constantly hitting them. (applicable to our current government problems sadly)
Oh, I'm sending it back to be looked at. The hat that comes with the picade doesn't have space for a heatsink or fan under it (as the instructions are put together to assemble it) so I'm also suspecting that its early doors for the Pi4 and picade still.
"The heat will be spread out, but not moved away from the device, so once it's hot, it will likely radiate that heat around the board."
The author doesn't seem to grasp the concept of convective cooling or how heatsinks are supposed to work.
I think that comment was in contrast to what they said about the fan:
> This option does apply a decent amount of cooling at idle and under load, it also moves the heat away from the board unlike the heatsink.
I think the comment makes sense with that context, compared to a fan, the heat from a heatsink will stay near the board and make future cooling less efficient.
The article is basically just a list of some options. Even for the most advanced option the author writes just about temperautures in "mostly idle" setup. Anyone has a link to an article with actual measurements?
Yeah I was rather disappointed as well. Could have run the same benchmark over some length of time recording the temp at intervals to see which is the best cooling option.
So many ad-hoc solutions, but I can't seem to find the maximum power consumption/TDP of the RasPi4, which would inform the kind of dissipation you'd need.
Best I could find was the BCM2711 (the Broadcom SoC at the heart of the RasPi4) "datasheet" (a frickin' Word document) [1] that says the board requires a PSU capable of providing 3A at 5V.
RPi power consumption and power supply is a bit more complicated than it seems on the surface.
A large problem is that even “3A rated" supplies suffer large voltage droop under load. This can be seen pretty easily using a cheap USB DC load [0]. As load goes up, voltage tends to drop slowly at first, then heavily at a certain point.
Several I’ve bought drop as low as 4.1V at just 2A load which will almost definitely make any Pi very unhappy. I haven’t found an authoritative source but generally the Pi’s power management system starts complaining (flipping the “undervolt since boot” bitflag [1] in “vcgencmd get_throttled”) around 4.6-4.7V [2]
To monitor the throttled bitflags, which also report thermal throttling, I wrote a basic shell script that just converts the hex to bin then returns a Grok-ready string with the flags as booleans. I've already got Telegraf on the RPis sending basic host metrics to an InfluxDB instance so that script just gets called by an exec input to generate and persist the data for monitoring or analysis.
Eventually I'd love to write (or contribute to) a proper Golang Telegraf plug-in that just reads from the VC mailbox directly but that's still on the back burner.
As far as actual power consumption during use, I’ve recently started running tests using a decent USB power meter (WITRN/Qway U2 [3] and X models, in particular) to monitor load over time while running various workloads on a Pi Zero W, a Pi 3B+ and a 4GB Pi 4.
Power usage at idle is usually low but even that's unpredictable. Daemons are gonna daemon regardless of the user load. Generally though the load goes up substantially when powering peripherals - Ethernet, USB, BCM VideoCore enc/dec/CSI/SSI, SD card R/W, HDMI, fans, HATs, GPIO peripherals, etc. CPU load definitely factors in heavily, of course, but it’s only part of the picture.
For reducing power usage, some surprisingly simple things can help. As a very minor example, shutting off the activity LED [4] can save a few dozen mA. :)
According to [1] a 1 meter USB cable can have a resistance of 286 milliohms, for a voltage drop of 0.8v over a meter (taking the resistance of four contacts and two lengths of cable into account).
So if you have a '3A 5V' power supply without a captive cable, you can still get 4.1v at the useful end of the USB cable.
The official RPi chargers include a captive cable, presumably to stop users substituting such standard USB cables.
A lot of the USB chargers most people see are way lower than the rating marked on them. That is why you should buy a special charger for your stuff, not the generic cheap stuff at the supermarket.
Haven’t reached any degree of publishing quality yet, but it’s something I’m aiming to do if time allows.
Ideally I’d prefer to Yep people run their own testing and monitoring. This stuff isn’t rocket science, just requires a bit of knowledge, inexpensive tools, and a bit of logical framing. :)
I guess Raspberry Pi 3 B+ (or CM3+) was close to the sweetspot in terms of thermal management vs performance tradeoffs, for many applications. Unless you really need more RAM or higher USB speed.
I hope the next two main steps for the next Raspberry Pi (4+ or 5?) and the corresponding compute module will be:
a) including a SATA controller (SD cards are not... ideal.)
I don't think a SATA controller makes sense for them. Keeping the price low is an important design goal, and adding more chips goes against that. An external hard drive with USB3 is probably the best option for a while.
I doubt SATA is that expensive, and honestly they could alternatively throw in PCIe instead of SATA if that's less expensive, and it would also be way more general. I'm considering getting a Rock64Pro because it's reasonably well supported and has a PCIe port, which means I can buy a SATA add-on board.
I wish Raspberry Pi had better support for fast storage.
Remember that the new Pi 4B has USB3, and according to the benchmark posted on their site it can read from an SSD over USB at 363 MB/s, and write at 323 MB/s. That's pretty good, I don't think you'll get much better than that with a SATA chip connected to the same SoC, so a regular external hard drive with USB3 is probably the best option for most Raspberry Pi customers.
The Rock64Pro is probably a great board too, and it also has USB3, so it can do both options. There's some Odroid boards that have built in SATA as well.
I'd start looking at actual volume prices for controller chips first, before dismisssing this, but that's just me. I have no idea what the cost is, but I could imagine it to be like maybe 5-10 cents by now. Just a guess. It's that kind of old but still very useful tech by now.
Btw: Since we needed to interface to SATA reliably via USB (from a raspi CM.)
We ended up identifying a particular Taiwan-designed JMicron-based SATA/USB2 controller that worked really reliably with Linux, unlike, say every single PRC-designed SATA/USB2 controller chip we tested. It was a pretty painful experience going through all of that testing.
We bought 120 retail packages of that JMicron-based product, directly from the Taiwanese producer, just to be sure, to guarantee our first-gen production run.
I think many RasPi users want a reliably high-speed, low-cost storage device. It doesn't make much sense for all of these users to have go through this very annoying process. It should just work. Since the chip cost for this by now is likely very small in the raspi kind of volume, why not?
Nothing of that complexity is 5-10c in volume (even huge volumes). I would guess somewhere in the 1-2$ range, which _is_ a big issue for them. For reference, 5-10c won't even buy you a middle to low performance op-amp.
The Pi became possible because of integration, where the SoC supported all the required interfaces internally so very few external ICs were required. That's what allows them to keep the price so low. Additional chips add significant additional cost. That's why there have always been compromises like the USB-connected ethernet. They can only do what the SoC supports natively.
Keep in mind that this is a 10+ year old problem by now (talking about USB2-to-SATA). They've gone through all of the cost cutting methods possible by now.
I can buy that chip that also happens to come with a PCB, a very nice extruded and anodized alumunium case, plastic end-pieces + screws, indication LEDs, voltage regulators, 10-20 pieces of small passives (redistors/caps), a screwdriver, 2-3 pieces of instrution manuals/warranty notices, two layers of very nicely designed paper packaging + shrinkwrap... all for like $5 in retail. The cost to the retail store for these items is probably closer to $2.
(Oh, and there's also the assembly cost for that PCB inside the case.)
This is my job. I design consumer electronics that ship in volumes larger than the raspberry pi. You've probably used something I've designed before. The backseat engineering by software devs in these threads is annoying and presumptuous. The pi designers aren't incompetent, they've probably rolled up dozens of feature BOMs to find what they can afford to ship.
Now, I wouldn't be surprised if it was 50c instead of 1$, but _nothing_ is 5-10c, so you're off by an order of magnitude. Even a single transistor in a package is a few cents. Additionally, if they add the IC they still need a connector, and those are some of the more expensive components in a device like the pi, probably 25c-1$ depending on quality.
> I can buy that chip that also happens to come with a PCB...
You're right that the economics don't seem to make sense, but you made several mistaken assumptions:
1) The price of that device is _almost entirely_ driven by the cost of the IC. Those designs have almost no supporting electronics, and the PCB itself will be 10-20c.
2) You're overestimating the margins. It's probably not delivered to the retailer for 2$, closer to 3$. Generic electronics are not high margin.
3) In cheap, commodity devices the manufacturer can use grey-market ICs, but the pi designers can't get away with that. That easily doubles the IC price.
> This is my job. I design consumer electronics that ship in volumes larger than the raspberry pi. You've probably used something I've designed before.
Fair enough. Thanks for your explanation.
> The pi designers aren't incompetent, they've probably rolled up dozens of feature BOMs to find what they can afford to ship.
That's a borderline straw man argument. You're seem to be implying that I either wrote (or think) that they are incompetent, while none of those things are true.
> Now, I wouldn't be surprised if it was 50c instead of 1$, but _nothing_ is 5-10c, so you're off by an order of magnitude.
Okay, let's say it's 50 cents. I would still think it makes sense to include this feature. Also, that's "half a magnitude off" compared to 10 cents, not a magnitude off.
That would certainly be a way for the foundation to go all Apple, with 3-4 different storage amount levels. I don't like it.
Caveat: I think it makes sense that all RPI5 boards (except the lite Compute Modules) would have like 8-16GB onboard flash, via eMMC. If it makes sense, economically speaking. I think it could. The SD card aspect of the RPI sucks in so many ways.
Except that MMC versions of RPi already exist and it didn't really bring the world down. They're also significantly more reliable due to lack of horrible SD cards.
Those MMC versions are called RPI compute modules. They didn't "bring the world down" because they required a separate board for them to be mounted on.
Good luck! I'd like to hear how Arch works out for you. I used to use Arch before I switched to macOS. If I ever get around to using a Pi as a desktop, Arch will definitely my first choice.
GP is not talking about the Arch Linux distro, but rather switching to using a distro compiled for the 64 bit ARMv8 ISA. It’s not a completely accurate comparison, but a fair analogy is armv7h is to aarch64 as i386 is to x86_64.
I just put my stack of Pis in front of a Noctua NF-S12A and it cools them all very effectively, even under heavy load, without the need for any goofy hardware that blocks additions like the GPIO extender or any other HATs I may want to add. I bought an adapter to power it from one of the Pis. It’s an easy solution that works for many of them with negligible effort.
I have one of these cases[1] for my Pi4 4GB from Amazon. Comes with heatsink and small fan (very quiet). Right now running "watch -n 1 vcgencmd measure_temp" puts it at 36C idle. Well I guess it is actually running k3s with no deployments/pods, so fairly idle.
I do remember it running a bit hotter before I updated to the latest firmware via "sudo apt full-upgrade" a few days ago.
I've got three 4GB pi4's running. Two are in FLIRC cases which do an Okay job (and they look good :-) They transfer heat to the case and the Pi peaks out on my loads at about 55C. I've also got a cheap fan+heatsink solution (https://www.amazon.com/gp/product/B07TZQHXZ6/) which does a better job and keeps the Pi under 50C on all loads (it doesn't throttle at all).
I recently put my Telegraf configs/scripts for monitoring RPi thermals and throttling into a git repo. These might be helpful for anyone doing their own comparative testing of cooling soliutions and/or power sources.
Has anyone tried immersion cooling (i.e. submerging an RPi in mineral oil)? Tempted to try it myself; just need some spare time (that I don't squander by commenting on Hacker News ;) ).
Yes it has been done, and works[0-3], it's just a matter of effort vs reward. Generally a simple heat sink and/or small fan setup works fine for nearly any usecase the Pi is designed for, and if you really do need to do heavier loads that require that level of effort to cool, then you probably shouldn't be using a Pi in the first place.
Outside of maybe the ICE cooling tower I suspect that all of these heatsink/fan solutions are woefully under specified if you happen to load the system or overclock.
I grabbed one of my little 20mmx20mm heatsinks which are generally good enough for devices of that size that come passively cooled. In the RPi's case that thing got hot enough to burn me even with a small fan. So I repeated this process a few times with stuff from my junk box trying to find a passive solution eventually I ended up with a 40x40x30mm heatsink (looks sorta like this https://www.amazon.com/Karcy-Aluminum-Heatsink-Cooling-Raspb... only that one says its 11mm) and a small fan.
The thing is huge, but the temps stay under control and it over-clocks @2Ghz now, but I wouldn't run it like that for long even with that heatsink. I've got mine pinned at 1.75).
One of my disappointments with the Pi 4 is that the official case for it got rid of any ventilation. Previous cases had vents and a removable top plate. I've found that the Primori fan shim works really good, but only with the cover removed. Will have to try drilling some vent holes in the cover and the back side of the case, and see if I can make it look good.
I also bought this one because I wanted a passive solution. Temperature looks fine (45°C in Idle). Although I think there are different versions and some of them don't fit perfectly on the RPi4. I got mine from amazon and it cools the three most important chips but the position where it touches the chips isn't perfect.
Note that the main issue here is to avoid cpu throttling due to excessive heat. IIRC you can allow a higher cut-off temp (100ºC instead 77ºC) by setting a config parameter. This will possibly reduce the chip lifespan from whatever it is to decades! But make sure no one can accidentally touch the chip.
I have one case with a fan and one case where part of the case body attaches to the chip top with thermal goop. The fan works better.
I read about the heat issues, so when I bought my pi4 I also got an aluminum fin cooler case with 2 fans built into it. Thankfully the fans are totally silent. I half expected it to sound like half a drone.
Now I can crank it up and rip the knob off, and the temperature doesn't get anywhere near throttle point.
Why do companies insist on shipping something that otensibly requires a heat sink, without the heat sink itself?
If I shipped out a fully-powered MK-R LED without a heat sink, saying "Oh it'll work fine!" I'd be getting chewed out within two minutes of the customer turning it on.
Maybe for the same reason that you can by a CPU without a heatsink - in order to provide flexibility. They assume that the number of people that will want to specialize and customize their heatsinks/fans will be so high in proportion to the people that will be just fine using the standard sink, that it is not worth it. There are very cheap easy to mount (like 2 seconds easy) heatsinks that will do the job (probably with thermal throttling though).
The problem for me is I attach a card to the top of the Pi for robot circuitry. That makes putting a heat sink or fan on the board awkward. I feel like having to do this really compromises the versatility of the Pi. For now I am sticking to 3s until I hear the temp issue is solved.
The Pi 3 has the same thermal throttling issue as the 4, and the 4's performance under throttled conditions is much higher.
Though IMO even calling it an issue is overblown, most phones and laptops cannot maintain their peak clocks under load for extended periods of time either. Intel calls it turbo boost so it's a feature instead of a bug.
Really? Because I run a Pi with a servo board on top that gets quite hot (too hot to touch) and it seems like CPU is performing time sensitive calculations (walking kinematics) adequately. It could be that the calculations are simple enough that they are insensitive to the throttling down. Do you have a suggestion how I could write some Python code to check if the 3 is throttling down? (Maybe a link to an example?)
Maybe it wouldn't be an issue on the Pi 4 that I think it is. In that case I have been holding back on switching to a 4 for no good reason. This is very helpful, thank you.
You certainly don't _have_ to do anything, and under a fixed computational load the Pi4 should run cooler because of the better core architecture.
However, you can get better performance by cooling. A fan off to the side of the board blowing between the two would probably work well if you want to keep it on top, but you're probably fine doing nothing as well.
You can do that, but many boards are really designed to sit on top of the card. In a robot that walks you don't have that much space and you want to mount everything as firmly as possible to avoid a bad connection from repeated vibrations. It is possible but so far it hasn't seem worth the hassle.
There’s something to be said for having the cpu on the back side of the board, like some of the NanoPi models. Easier to spread the heat with no connectors in the way.
Most don't require cooling, but will have better clocks. The 4 itself throttles under heat (above 60C and heavily over 80C), and seems to have good enough behavior even if not actively cooled.
Probably either the Odroid N2 or HC2 or Nano Pi M4.
The one feature all these boards have in common is their CPU is mounted on the _bottom_ of the board, allowing a much larger heat sink to be mounted without obstructing anything. The lack of this feature in on the RPi 4 is a fatal blunder IMHO and why I chose not to buy it.
I currently use a Nano Pi M4 as my main desktop computer. It has no active cooling, only a big heatsink which covers the entire rear of the board. As I write this, I have three browser windows open containing a total of 30 tabs while streaming a 44 kHz stereo 128 kbps audio stream. Its CPU temp is holding at 32 C.
Even Pi 4 B doesn't seem to require active cooling. It will do fine-ish (sitting at barely under the temperature cutoff limit of the thermal throttling, under load, by using a bigger type of passive heatsink), which is still quite a feat. To have a cheap totally silent computer with the specs that 4 B provides is still a great achievement.
My Pi 2 stays comfortably cool, but it's rarely under heavy load. It serves as a Gopher file server for a bunch of old computers and occasionally a music player, so I think the heaviest thing it does at any time is to decode mp3/ogg.
I'm not sure why, but my RPI4 become broken 4 times in a row. I returned it 3 times, but finally I gave up. Every time it eventually stopped to boot up and there was solid green and red LED light.
There was original raspberry PI case, modified to mount Noctua 5V fan, with additional holes to allow air circulation. There was original power supply.
It was only connected with ethernet, this fan through GPIO and HDD drive through USB.
Now I have HP T620, can be bought cheaply, it's definitely bigger than RPI4, but finally it just works.
We spend a ridiculous amount of money testing various cases from China. In testing we found that the heat sink style[1] cases with fan work best.
Having said that, for most use cases you just need to move the hot layer of air away from the SoC, so even a low speed fan will do the trick.
[1] https://raspberry.piaustralia.com.au/collections/cases-enclo...