Way to keep your cool running a Raspberry Pi 4 (2019) | Hacker News Reader

Way to keep your cool running a Raspberry Pi 4 (2019)

jeffgeerling.com

137 pointsguiambros99 comments

vertical orientation also seems to impact temperature

""" Simply moving Raspberry Pi 4 into a vertical orientation has an immediate impact: the SoC idles around 2°C lower than the previous best and heats a lot more slowly – allowing it to run the synthetic workload for longer without throttling and maintain a dramatically improved average clock speed.

There are several factors at work: having the components oriented vertically improves convection, allowing the surrounding air to draw the heat away more quickly, while lifting the rear of the board from a heat-insulating desk surface dramatically increases the available surface area for cooling. """

https://www.raspberrypi.org/blog/thermal-testing-raspberry-p...

It's not that simple in the general case though: putting a board vertical will cause the top of the board to be hotter than the bottom, possibly a lot hotter than it would have been in a horizontal position even if the average temperature of the board is a lower one!

Looking through a thermal imaging camera gives all kinds of interesting insights.

This is a useful article. Thank you.

Thermal engineer here. There absolutely no need to use fan cooling, humongous heatsinks or harebrained temperature controller schemes to cool the raspberry pi 4.

All that is needed is a small heatsink to slightly decrease the thermal resistance to ambient air, to stay below the thermal throttling threshold.

Anything more complicated than that is a pointless exercise in pointlessness.

I kind of agree. I think in most situations, a Pi is going to sit idle most of the time, so it won't need active cooling.

What if I want to use pi 4 as a building farm? Especially now with the 8gb version out, it seems to be suitable for building haskell programms

Well then I'd say sure, of course you need cooling. But I'd also guess that you're within <1% of Rpi users who actually need cooling, and I'd guess that's also why Rpi doesn't come with a radiator by default - vast majority of users simply don't need one and it would increase the cost.

I use my pi 4 as part of a build farm with a 2cm tall passive heatsink. It never goes above 65c.

64-bit kernel, or 32-bit?

It's a really, really slow computer. The Pi 4 is something like 10x faster than the original Raspberry Pi, but that was also a really, really slow computer. I remember not wanting to set up a cross compiler once, and recompiled ntpd on one. It took more than 24 hours. The same build runs on my desktop in seconds.

It is good for things like reading a sensor and sending it to some server, but so are ESP32s which are cheaper. You are paying a heavy price to get video output and Linux compatibility. In the current era, it's a bad microcontroller (expensive!) and a bad desktop computer (slow!). When it came out, most people were using $25 MCU boards, so $10 extra to run Linux was a great deal. These days, a massively more capable MCU dev board can be had for $4 (if you want free 2 day shipping, less if you can wait).

It's relatively cheap, and it seems to be supported fairly well (loosely, "good"), so that's two points of the triad. The Law says we can't also have "fast", but that's a reasonable trade-off for those who prioritize "cheap" and "good".

I mean, if you want a build server that runs at Core Duo speeds, you can probably find a used Core Duo workstation or laptop on eBay for less than what a Raspberry Pi costs. While we often talk about how Moore's Law is dead, go back several generations of computers and you'll be impressed at how slow they are. (I'm regularly impressed with how slow current generation computers are!)

Would you describe that flirc case as complicated? To a naïve layman, it seems like the smallest all-metal case that'll fit the rpi4, made so that the metal acts as a heatsink. Which may be more heatsink area than a 15W CPU needs, but not bigger case area than is needed. Is there something I'm missing?

I got a pi 4 around 6 months ago and got the flirc case based on the recommendation of TFA. I've been very pleased with it - makes the pi look nice; I no longer worry about other people touching it; and of course it cools well (the very thick thermal pad connecting to the case is a bit unusual, but it works).

I got the flirc usb dongle for remotes at the same time, since I thought I'd want it (my pi is a media center) and there was a discount at the time if you bought both it and the case.. I've been far less impressed. It's a great idea in theory, but in my experience there was enough flakiness/delay with each button press, especially with Kodi for some reason, that it was basically unusable. Also, their utility for programming the dongle did not run well on the pi itself (it stores configuration on-device so you don't need to program it from the device you'll be using it on), and the install process involved setting up a repo that is served over http and does not have an associated gpg key, which gave me a bad feeling (although overall it felt more like a startup flying by the rest of its pants than anything malicious). I am somewhat hopeful that with more time/polish these issues will be fixed in software, but can't currently recommend it.

Kinda wish the top of the FLIRC case wasn't rubberized. My older one is sticky now, but that might be more of an issue in Florida humidity. Still got one for a Pi 4 and it ran well enough.

I believe the FLIRC case is custom designed for the RPi4, not just "happens to fit" as you imply. Otherwise you have it correct; it's an aluminum case that mates down onto the CPU for direct passive cooling.

Note that it's clearly an injection-molded aluminum top-half and the bottom-half is plastic, just used to keep the dust out.

I would describe it as simple and elegant; fit for purpose.

Otherwise I'm quite happy to have a full passive cooling config, as currently my RPi home server must live in the kitchen/dinning room area due to space constraints.

I did not mean to imply that. Rather, if anything, that it seems to be based on an insight that a metal shape that's enough to enclose also works as cooling.

What a pity that there's plastic on the bottom.

Can you qualify your statement (please)?

In what case should I care about fan cooling then? How are electronic components affected by temperature in general?

This is basically an appeal to authority. You haven't explained what property of the Raspberry Pi 4 means that a tiny heatsink is sufficient. Why can't I just replace the words "raspberry pi 4" with anything else and conclude that I never need fans or large heatsinks?

Because you can see in the graphs that the bare pi doesn’t get that hot. If you allow just a bit more heat to escape you push the graph down to where the peak sits just below the throttling point.

For home/hobby use this doesn’t matter too much but you typically want to run as hot as allowable. It’s a win-win since small heat sinks are cheap and don’t have any moving parts.

Running 32-bit raspbian... Try that with a 64-bit OS, running 64-bit compiles/etc.

I have a huge heatsink on mine (40mm cube), and even without the overclock it will crank up to 80C doing long running compiles. Running it at 80C for an hour or two, the temp of everything on the board starts to approach 80C. Touching the usb ports at that point will give you a surprise.

So, I added a small 20mm fan, and set the profile to kick on at 60C, which generally keeps the soc thermals in the 60-70C range. No throttling, and still no noise because with that large of a heatsink it doesn't take much air to cool it.

Add in a couple hundred Mhz of overclocking and that fan stays on pretty much all the time.

Fan cooling a Raspberry Pi 4 is such an odd thing to do. You are essentially taking one of the pros of the Pi - quiet and reliable operation - and replacing it with a moving part that will not offer too much benefit and will eventually wear out or start having issues.

As someone who uses pis and MCUs, I don’t find anything that relies on a micro sd card to be reliable.

You can USB or network boot a Pi. Also in my experience over 9 years: just don't use Sandisk cards and you're fine. Samsung (Pro|Evo) branded cards have worked for me.

Network boot adds even more points of failure. USB boot is very attractive, but also very new. I’ll play around with it once it’s out of beta and the setup procedure is established.

I use Samsung Evo cards and haven’t had an issue with them either. My problem is one of confidence. I have no data to suggest that these cards won’t just hard fail at some point in the next five years. Simply doing better than their competitors in the micro SD card space is not good enough.

All of the dead cards, if I open my drawer where I store them marked with a black cross-mark, are samsung evo cards. All my Sandisk Ultras (the real ones verified on Sandisk's chinese website) are still running fine after ~3-4 years of usage.

Anyway, Samsung Evos didn't hard fail for me. They just stopped executing writes (without reporting any errors). So your system seems to work (for a while), but after reboot it's back to where it was previously. It's a pretty stupid way to handle end-of-life condition for the card. I wish samsung decided to report write failures instead of simply ignoring the writes and reporting success.

Officially there is no support for USB boot on Pi 4 yet.

The current supported solution is boot from read only file system on SD card, overlay file system with writes to the USB drive.

The foundation plan to add support back but for now only supported on Pi 3 and earlier.

You don't need to use an overlay filesystem (didn't know that was even a supported solution), you can move the rootfs to the usb device and point to it. See https://gist.github.com/lucabelluccini/9a11c48dcf1d627bbcbd8...

There is post that at least describes in detail how to enable this beta feature: https://www.jeffgeerling.com/blog/2020/im-booting-my-raspber...

Same, have some units that have been running nonstop for the past 5 years without having to touch them at all, using Evo cards. Also, I don’t think I mounted var/logs as a tmpfs.

My issue with heat management is dust (on fan), also I live in the equator line area that makes it very hot in the afternoon (makes heat sink less effective).

That makes me think a lot before choosing passive or active cooling.

The Flirc case-as-heatsink one seems good tho, and on-line with your statement

Somebody's never overclocked their pi before...

With an overclock, even a FLIRC would throttle when running our full unit test suite. A 30% reduction in unit test runtime is nothing to sneeze at and I certainly didn't want to downclock to 600-800MHz due to throttling. An ICE Tower was a lifesaver.

by overclock, do you mean full-out running at peak all the time? or letting the CPU governor still work and just bumping up the limits?

one makes sense, the other is hubris.

In theory, TDP scales linearly with clockspeed. In practice, power increases higher than that rate. 30-35% increases in heat to dissipate is significant.

I'm old school when it comes to overclocking. Stability and sustainability matter to me. I don't always run all cores full-on for hours, but when I do, I want a cooling solution that's up for the job.

Did you read the article? He tested a heatsink. I'll quote the relevant part.

> Heatsinks will generally shave 5-10°C off the highest temperatures you encounter, which is significant, but you still need to have air moving across the heatsink for it to do anything. This means either a fully open Pi (great for testing or tinkering, not so great for a semi-permanent installation), a well-ventilated case (with slots or holes in the bottom and top to allow natural convection), or a fan. So not a bad choice, and they're so cheap you should probably stick them on any Pi you can... but heatsinks alone won't solve Pi cooling problems.

He doesn't have much of a methodology and seems to have missed a key point of the heatsink: orientation.

If he'd mounted his Pis' GPIO parallel to the ground, heatsink fans perpendicular to it, he would have found his missing convection with or without a heatsink.

This is what I do, and mine haven't reported >45`F ambient under peak load in a passively ventilated area. There is no issue with heat removal with enough air space.

The article is dated November 22, 2019, which makes me wonder whether it was before the Raspberry Pi 4 firmware upgrade that reduced heat generation quite a bit.

My two 4GB Raspberry Pi 4s have had no thermal issues. Running the same measurement script now, will report back the results.

Edit: Ok, done with the tests. Both RPi4s are bare and should be running exactly the same versions.

RPi4 #1 started at 51C, peak 79C, no throttling at any point. Ambient temperature 24C.

RPi4 #2 started at 54C, peak 82C at which point it started to throttle. Ambient temperature 27C.

Are you talking about the update described at https://www.jeffgeerling.com/blog/2019/raspberry-pi-4-might-... which is mentioned in the article? Or something else?

Possibly.

I haven't closely followed RPi4 firmware development, but I did notice positive changes since June 2019. I think there have been at least two occasions that dropped the (idle) operating temperatures.

In any case, I'm happy with their thermal performance.

I've also a feeling there were two things they changed but I don't recall what the first one was. And I don't think there's been any more improvements since November so I think this article includes everything so far.

When the pi 4 first came out, mine throttled when driving two screens and that was depressing. I've not tried that setup again since but I really should.

TDP theoretically scales linearly with clockspeed and scales worse in reality. Most Pi 4 can sustain 2.1GHz CPU and 650-750MHz GPU. That's a 30-35% increase in clockspeeds. If I'm taxing the system, even a FLIRC case will throttle on my pi. ICE tower has done an amazing job and the sound isn't noticeable on 3.3 volts. In fact, you can simply unplug the fan and passively cool (though it'll throttle after a few minutes at peak clocks).

I used the pi for dev work for a few weeks as an experiment. Overclocking is the single-biggest improvement you can make. An overclock combined with the IPC improvements from a 64-bit OS, and load improvements via SSD, the system is probably 75% faster which is a huge deal at that performance level (the difference between extremely frustrating and just mildly annoying at times).

Yeah that's consistent with what I saw when I looked at the upgrade. Two degrees gained or so.

I gather the main temp gain isn't the CPU though but rather on the USB charging circuitry.

Nice test, it could be cool to have people repeating it to see how much the "silicon lotery" influences the temps.

Also there a few model of cases on ebay/aliexpress that would be interesting to test such as https://m.aliexpress.com/item/4000509636456.html (with a heatsink added)

I have one of these:

https://cdn-shop.adafruit.com/1200x900/4340-08.jpg

It never gets hot that I can't touch it.

https://www.aliexpress.com/item/4000392602236.html

Anyone has links to similar articles with regards to either

- SD card reliability

- techniques for reliably using what we have

Given that people are using these for all kinds of stuff there must be some advanced techniques being used (or some basic I have missed).

I've been running a raspberry 3 with a cheap sandisk sd card for years now without any SD Card issues. It runs a DNS based ad-blocker and a vpn.

I am logging everything to ram instead of writing it to the card and that's about it. I never encountered any issue with SD Card reliability, despite me sometimes just pulling the plug instead of properly shutting it down.

Here is a simple guide: https://raspberrypi.stackexchange.com/a/62536

I do have a couple of other raspberries for other uses (e.g. a small rc car) which are also running on their first SD card. Might be just luck, but I do think that it's not necessary to go to extreme lengths to have them run reliably.

Many problems with SD cards are actually rooted in power issues (bad power source or usb device that eats power). Ever since I took proper measures to ensure stable enough power I've had no issue (Pi 3B here)

I can second this. My previous company runs a few hundred Pi units across loads of separate locations; the locations where we had constant corruption issues were also the ones experiencing dodgy site power-supply issues. We did try various alternatives to cheap SD cards but ultimately running them off PoE from a switch attached to a UPS was a pretty good solution to the issue.

Can you elaborate on these measures? Sounds interesting.

> Many problems with SD cards are actually rooted in power issues

Does anyone have any info on this that's more substantial than hearsay? Experiments, measurements or at least a more technical explanation than "bad power source"?

I'm persistently hearing this story that power supply quality affects SD card life. As an electrical engineer, and given what I know about R. Pi design, I fail to see how an SD card could get physically damaged by any reasonable power supply that would otherwise run the rest of the R.Pi.

I suspect people are confusing filesystem corruption due to brownouts/OS not shutting down cleanly with physical damage to the flash (i.e. unreadable sectors on SD card)

> I'm persistently hearing this story that power supply quality affects SD card life

Not SD card life indeed (in my case at least), merely IO failing, hosing the filesystem long term because of b0rked writes. The SD card is fine, it's just the data that eventually becomes inconsistent garbage.

I noticed the power LED showed the undervoltage behaviour, measured the power output which indeed showed the voltage not being stable on power spikes, so bought a better power supply, problem gone. Got the same kind of issue with greedy USB devices, so either went the powered hub route or used self-powered hard disks.

Regarding microSD card reliability, one option would be to simply forego the microSD card completely and boot the Raspberry Pi 4 via an SSD connected to one of the Raspberry Pi’s USB 3.0 ports.

I was originally going to recommend checking out this¹ article for instructions on how to do it. However, much of the instructions are no longer valid (as of two weeks ago!) since they assume that the Raspberry Pi 4 doesn’t natively support booting from USB, which finally isn’t the case (although only if you use the current beta bootloader²). I’d still recommend reading it though, as it also mentions known working 2.5″ SATA to USB 3.0, M.2 NVMe to USB 3.0, M.2 SATA to USB 3.0, and mSATA to USB 3.0 adapters. In addition, I’d recommend checking out this³ article with some recommended SSDs; in particular, this⁴ 120 GB Kingston A400 SATA 3 2.5” Solid State Drive coupled with this⁵ StarTech 2.5″ SATA to USB 3.0 Adapter, which also has the added bonus of giving you far better speeds than a microSD card would! As the author of that article mentions:

“The Kingston A400 drive performs really well in the Raspberry Pi Storage Benchmarks⁶. It’s a great drive and is cheaper than many mid-range MicroSD cards.

The 2.5″ SATA to USB adapter above allows us to do this. There is no power adapter needed as SSDs are low power and are powered by the Pi through USB.”

――――――

¹ — https://jamesachambers.com/raspberry-pi-4-usb-boot-config-gu...

² — https://www.tomshardware.com/how-to/boot-raspberry-pi-4-usb

³ — https://jamesachambers.com/raspberry-pi-storage-benchmarks-2...

⁴ — https://www.amazon.com/dp/B01N6JQS8C/

⁵ — https://www.amazon.com/dp/B00HJZJI84/

⁶ — https://storage.jamesachambers.com/

Ah, just saw that Jeff Geerling (the author of the linked article in this HN post) also has an article about booting the Raspberry Pi 4 from an SSD located here¹ (HN thread here²).

――――――

¹ — https://www.jeffgeerling.com/blog/2020/im-booting-my-raspber...

² — https://news.ycombinator.com/item?id=23471729

My original pi has been corrupting SD cards all the time, especially now that I am running pi hole on it which seems to need more RAM than is available on my old board.

For 2-3 years I had a tower (1m height) with 8 RP3s, 4 thinker boards and about 12 Orange Pis (different models). I did run different BOINC projects on it. So the load was 100% 24/7.

The whole time I had 1 or 2 broken SD cards. Temperature was never an issue.

I put on all SBCs just a passive heatsink (about 30x30x20mm). I installed packs with 4 boards vertical. On the bottom I had 2 large fans for all the SBCs. (sorry no picture)

The easiest is to eliminate the SD card by either booting from network, move the rootfs to a USB drive, or boot from a USB drive (now available for Pi 4 with beta firmware). These are documented at https://www.raspberrypi.org/documentation/hardware/raspberry...

A good approach used in industrial/embedded applications: use a read-only root volume. If you don't write to your SD, you don't wear it out and have no risk of file system corruption either. It's quite a lot of work though, since most packages assume they're installed on a writeable file system.

Debian wiki has a (pretty bad and outdated) page on the subject https://wiki.debian.org/ReadonlyRoot#Enable_readonly_root

Either network boot or SSD boot. The SD cards do eventually die. I've lost two already

At the bottom of this thread the official response from SanDisk support is that all cards except the High Endurance are not supported for Raspberry/Linux:

https://www.raspberrypi.org/forums/viewtopic.php?p=1158010

I've been using them for the last couple years and haven't had any dead Pis, unlike what happened with a couple other "regular" sd cards.

The reasonable answer seems to be the flirc case.

No noise, no moving parts, excellent cooling.

Up here in NH it gets a little chilly in the barn loft where the server rack is hiding inside an old credenza. A chicken egg incubator on a heat switch keeps the spinny disks on the NAS above 35F and it turns off at 45F. It only turned on a few times during the coldest days last year. I dont know for a fact whether spinny disks like to be above freezing but anecdotally someone said yes and it was 12 bucks to set it up.

When we had some very cold winters in the UK I had to wait for the room to warm up for our 8 inch floppys to work in our DEC PDP.

The flirc case is a giant heat capacitor, not a heat sink.

Notice that it has the only temperature curve where the temperature does not go back to baseline after the stress test. It needs a lot longer to get to equilibrium, and the final temperature is almost certainly hotter than what's shown on the graph.

exactly. For passive heat sinks, you need big surface areas. The FLIR case doesn’t do that and is arguably one of the worst options esp considering its price.

$16?

I have been running one for a few months, and it does get consistently warm, but does not throttle the CPU (at least for my workloads). Will be setting up a second one as a “workstation” soon.

There's still a good increase in surface area for convection it's definitely a heatsink. The graph does a poor job showing this but the start and stop idle time is 5 minutes. That's a good amount of time for temperature to settle.

You can additionally stick a cpu heatsink on it (like prop. 2 in this article). This will help dissipate heat from the case.

But overall, this case is still better than bare pi, or regular plastic case (the heat capacitor has much larger surface than the cpu of the pi).

I own one. To some extent you're right. My pi runs mpd and pi-hole at all times and under that relatively low load, the case is constantly warm to the touch. But it has enough surface area that I've never been worried about overheating. Refer to TFA -- a tiny little heat sink on the cpu was all that was need to keep from throttling.

I don't have the hardware to give a proper surface temp reading. I can measure the pi's reported temperature / test for throttling under extended load later, but for now: I stuck an "instant"-read thermometer intended for cooking into one of the usb ports, which was the hottest place I could find on the case. It measures 36.1°, compared to an ambient temp of 23.4°.

My PI is too cool. It freezes.

I have playing with a PI Zero W over the last weeks. At first everything worked as expected.

But now it developed some strange "hangings" I cannot explain. It kinda randomly freezes in time for a few minutes every now and then.

Sometimes it takes several minutes to ssh into it. Sometimes, the terminal just freezes for several minutes.

Neither top nor iostat seem to show any elevated activity during these periods. Since they hang throughout the freeze, its a bit hard to say though.

The strangest thing: Pings to the PI always come back fast. So its probably not that the hardware really freezes completely.

disk issues?

Have you tried replacing the SD card with a fresh one with a fresh install and checking if this still occurs? I've had a somewhat similar problem and it turned out to be a dying SD card.

I will try that. At first I would like to see if I can find the problem. I would have thought that when the SD card is not responding, it would show up as "wa" in top and "%util" in iostat -x. But both are zero for the next output after a freeze.

Is your hostname added to /etc/hosts?

I had a similar problem. For me the reason was that I used a cheap charger with too less power.

Just exchanged the charger .. the hanging still occurs.

Could strace on the ssh server help you? You would know if it's a disk issue.

How would that work?

I've had this sort of issue caused by bad SD cards. Basically anything that touches the disk can block for several seconds sometimes.

Sounds like a crappy sdcard to me. I tossed an older one out recently that would randomly stall during io operations.

I'm using mine primarily as a streaming device. In my experience a heatsink is more than good enough. Depending on your workload all a fan will do is make a bunch of noise and spread dust around.

Folks may also enjoy "Cooling off your Raspberry Pi 4" from my blog (also in 2019) https://blog.alexellis.io/cooling-off-your-rpi4/

I compare a number of the different fan / heatsink solutions including pro/cons along with a hacked Noctua fan

I've been doing voice + screencasting on a modern MacBook Pro. The fans seem to intermittently kick at full blast while connected to a 4k external display, or sometimes it's just Dropbox doing CPU intensive things. Either way it's disruptive to the sound, and background noise removal in post doesn't completely fix it.

I've never had to think about heat dissipation so much. Might try a laptop cooling pad.

This is timely for me, I finally am upgrading my original A to a 4! Another advantage of the Flirc case is that it hides the blinking lights which are so tempting for my toddler!

I’ve tested a heatsink case on a Pi4 with and without fans. Without the fans at max load (all cores) it didn’t pass 60 degrees Celsius (20C ambient). So fans are not required.

Same, I hit 59 degC max running the script from tfa. I'm in a house with the A/C on set to 80 degF.

Edit: this is my case

https://chicagodist.com/products/black-raspberrypi-4-armor-a...

I have a bunch of the FLIRC cases for my various versions. They are relatively expensive for what they are but they work well.

I've had a little fan on an old rpi3 I had and it was annoyingly loud.

Why do people use cases on a Pi? I’ve never ever used a case.

My Pis are always out of the way and in a shelf somewhere, untouched until I make modifications to the setup, in which case the case is often coming off anyway.

There are "hardware project RPis" where you're always futzing with the gpio header, and then there are "appliance RPis" that live on top of the router they're connected to via a six inch Ethernet patch cable. A case is common for the appliances.

So the cleaners will touch the case, not the electronics. Also, the bare board doesn't offer a convenient way to fix the Pi to a wall IIRC.

> Also, the bare board doesn't offer a convenient way to fix the Pi to a wall IIRC.

My Pi 3 has little circular holes that I assume are meant to mount into a case, but which I have successfully used to just hang the board on a hook on the wall.

Dust and protection from casual accidents. Having a naked board on any kind of desk is just asking for trouble...

I'm gonna go ahead and be the shallow person here: I kinda like the look of the cases. It's a shame the official case has the worst heat problem, because it looks the coolest to me.

Reply on news.ycombinator.com