Not-Quite-So-Hot Stuff: A Thermal Exam On The Latest Raspberry Pi

When the Raspberry Pi 4 was first launched, one of its few perceived flaws was that it had a propensity to get extremely hot. It’s evidently something the Pi people take very seriously, so in the months since they have addressed the problem with a set of firmware updates. Now they’ve taken a look at the effect of the fixes in a piece on the Raspberry Pi web site, and it makes for an interesting comparison.

The headline figure is that all updates together remove about a watt of power from the load, a significant quantity for what is still a board that can run from a capable phone charger. Breaking down the separate parts of the updates is where the meat of this story lies though, as we see the individual effects of the various USB, memory, power management and clocking updates. In temperature terms they measure an on-load drop from 72.1 °C to 58.1 °C, which should be a significant improvement for any Pi 4 owner.

There is a debate to be had over in what role a computer such as a Pi should serve. As successive revisions become ever more desktop-like in their capabilities, do they run the risk of abandoning the simplicity of a cheap Linux box as a component that makes us come back for more? It’s a possibility, but one they have very well addressed by developing the Pi Zero. They have also successfully avoided the fate of the Arduino — inexorably tied to its ATmega powered original line despite newer releases. As we have frequently said when reviewing Raspberry Pi competitors, it’s the software support that sets them apart from the herd, something this power-draw story demonstrates admirably.

53 thoughts on “Not-Quite-So-Hot Stuff: A Thermal Exam On The Latest Raspberry Pi

  1. Using those numbers, the theta JA (thermal resistance between junction to ambient) is roughly
    (72.1C-58.1C)/1W = 14C/W Since the 1W figure is not a measurement of actual power going into the SoC, take the value with a bucket of salt. Likely it is at the input to the entire Pi, before any power supply regulators.

    1. I’d say it’s more to do with the Pi people being a small organization, albeit one with good resources. That they get it out at all is amazing, and that they fix things afterwards is a HUGE bonus.

      1. They’re not that small an organisation – I count over 150 people on their “meet the team” page. That’s plenty to be able to take a product like to market (I’ve worked in a company half that size which was world #1 in their field with a not vastly dissimilar product, born from similar roots). In addition, I think they’re in the luxurious position of completely owning their market – customers are (rightfully) loyal, and will accept the next Pi whenever it arrives (cf Apple), so there should be more priority to it being ‘right’ rather than ‘released’.

        Boards getting hot to the point that they get throttled by 50% after a minute is pretty inexcusable. It must have been noticed from the day the first prototype arrived in the office.

        That the number of fixes are many, and released repeatedly and quickly shows that the product was not ready, rather than that they’re great people for fixing things (which shouldn’t have been broken). However, that doesn’t take away from the fact that, when pushed, the engineers _have_ fixed them quickly.

        1. Seconded! I worked at a semiconductor company which, at the time, was 35 people, and they managed to ship some pretty sophisticated systems-on-chip that had a whole lot of software and firmware wrapped around them; not to mention all the electrical design, manufacturing, test, and other disciplines that go into shipping silicon in volume. That same company sold for close to half a billion dollars after growing to the monstrous size of maybe 80 people. It doesn’t take an army to do good work.

        2. While you are not wrong the Pi folk seem to prioritise stable and functional over optimal in theory – better to thermal throttle but work than have weird issuses. So internal builds probably had all the firmware updates going around enough in advance that they expect it all to be ready for release and know it works.
          As there is nothing wrong with the hardware it is actually ‘ready to ship’ because its stable not because its perfect (the usb power issue for me doesn’t count as who uses such expensive supply for a Pi, and this wifi jamming thing seems like its not a pi issue but a cheap HDMI cable in proximity issue – though doesn’t seem well enough understood yet).

          One of the things I like most about the Pi over any other SBC is the support and functionality. The hardware just works without compiling odd drivers, or needing really out of date kernels.

        3. Also inexcusable is not following the USB C power circuitry specifications 100%, which caused the Pi 4 to have issues with good USB C cables while working fine with cheap cables that didn’t fully meet specifications.

        4. Most of those 150 people work for the Raspberry Pi Foundation: writing educational resources, training teachers, running Code Club and Coder Dojo, delivering our parts of the National Centre for Computing Education etc. Of the 70 people who work for Raspberry Pi (Trading), roughly 30 are engineers, and roughly half of those worked directly on Raspberry Pi 4.

          You can do a lot with 15 exceptional engineers, but it’s unrealistic to expect the platform to be in some arbitrarily defined “complete”, never-to-be-improved state before launch. We launch products when they provide a good experience for the majority of users, and then polish them in software until they provide a great experience for everyone. That is a feature, not a bug. We are *still* polishing Raspberry Pi 1, which we released in 2012.

          There are getting on for two million Raspberry Pi 4 units (out of 30 million total units) in the field now. People seem to like them. And they’ll still be getting software upgrades in the mid-2020s.

      1. Not quite sure how that’s a useful comment, but yes, I’ve been there, done that (several times). And yes in some of the cases, the company has overriden what’s ‘right’ in favour of what’s timely.

        Sometimes they’re right, sometimes they’re wrong.

  2. The least they could do is put the SOC on the other side of the board, so it would be easier to heatsink when using the expansion connector on top.

    Ditto the Beaglebone AI, which now runs so hot it needs a heatsink too. Oy vey.

    1. Probably would make routing harder.
      I would say solder the 40-pin connector from the other side, and change those two flat cable connectors to 90 degree ones, and if possible move and align them to the borders.

      Even so, most of these “software fixes” seem like those for the intel security problems : you fix the problem, but decreases the performance.

      1. Guess you didn’t actually read the article?
        Or maybe you just didn’t understand what they said?
        Each of the firmware improvements INCREASED the amount of time BEFORE the processor started to throttle it’s clock due to temperature.
        In other words, performance IMPROVED with each fix.
        In the final test, using a more common real-world example of compiling a kernel, the rPi4 didn’t throttle at all during the compile whereas the rPi3 started to throttle immediately, taking nearly twice as long to compile.

      1. Wow, you’re just a wealth of useful information in these comments!

        Care to explain what PCB designs and thermal calculations will prove when it comes to the question of the size of heatsink one can use when one connects a HAT to the 40 pin connector? Seems blatantly obvious that if one is using a HAT, the size of heatsink you can use is severely limited.

        But when you feel like putting someone’s comment down without any real argument, barely tangential questions are probably your best bet.

        1. The armchair “engineers” claim that they have the fix for everything. do this, blah blah, I’d do it this way, blah blah.

          Without any calculations, BOM, circuits, layouts to actually prove it.

          For example, If I did the layout, I could reduce heat by 200%, thermal throttling would be zero.

          Furthermore, the designers of the SoCs don’t know what they are doing. If I did the SoC design, I could get 16 cores, at 4Ghz, for the same price!

          And again, those idiots compiling the kernel, I have 64 bit linux running for ages.

          Just don’t ask me to post anything.

          1. Go ahead.
            Build an SoC, and a board create a community and launch it all for less than $40 a board.
            We’ll all wait . . .
            . . . .
            . . . .
            . . . .
            . . . .
            . . . .
            You done yet???

          2. Therein speaks an armchair engineer. These fixes are self evident by them being done. My beef is in shipping the product before they’re done. I am not disparaging the engineers – I’ve worked with some of them.

    2. Fan, the AI needs a fan. (Or a huge ass extra heatsink)

      I’ve run their OpenCV example using a webcam in headless mode, and the BBAI shuts down from overheating after some 30 seconds+. From what I read on their forum, running a desktop can cause the same.

      It’s a nice board, but it’s got a heating problem. In contrast, I just leave my PI4 out on my desk, in the box it came in (opened) and it runs just fine, and has done so from the start (I’m using it as my main computer and I’m posting this with it). It’s shut down exactly zero times, and hasn’t given me overheating warnings since I installed the USB fix (and then it only happened once). Not that the Pi and the AI are trying to solve the same problem, but just to show how other SBC-makers are handling things.

      1. The overheating and price aren’t exactly related. Don’t let my lack of coherency fool you. The price to enter with the Pi 4 has gone up drastically (depending on what parts you have lying around). Most people would associate a higher entry price with a “better product”. It kinda sucked that dropping that money ultimately led to a device that didn’t last for more than a few days.

        Not to mention, the dual HDMI outputs didn’t allow the display to power off without loading a beta firmware update. Not willing to do that, I was stuck with a monitor that would never turn off.

        I love the Pi, but I’m not sure this thing was ready for release.

        1. Just and FYI for other people thinking about getting a pi 4. My actual cost break down on getting my pi 4 up and running.

          USB C to micro USB – $6.99
          2x micro HDMI to HDMI – $7.99
          Raspberry Pi 4 – $46.99
          32GB SD Card – $8.28
          ___________________________
          Total $70.25

          Now I have a $46 paper weight…

      1. You can with a charger which outputs enough current without negotiation, but unfortunately gone are the days you could more likely than not get by with any crappy old phone charger.

  3. We might be seeing the next step in miniaturization. Minicomputers replaced mainframes, and microprocessors replaced minicomputers. AMD’s latest chips are SoC designs. It could be that SBCs will become the normal “desktop” in a few more years.

    1. You can almost see that now with the ‘smart-phone’ . I bet a lot of people now use it in place of a PC. I think the PC downward trend is showing that. I see a future where you ‘dock’ your device and get access to your keyboard/mouse/hires screens/local storage/etc. Voila, its a desktop! Now where the PC is still dominate is the Gamers/Engineers/Designers/Programmers/Scientists/etc where absolute performance is desired. I personally always would want to ‘pick’ my own components like memory/processor/storage/Operating system/etc. But for majority? Not so much.

      As for the RPI4, I have three now. All work just fine. Two have fans. I run my 5V fans at 3.3V which seems to be enough cooling and no fan noise. One has a metal case that acts as the heat-sink. Hat usage is either riser pin extenders, or a ribbon cable. I can deal with that.

    2. Aren’t plenty of laptops essentially an SBC these days? Plenty of low-end Chromebooks have a non-socketed ARM CPU with memory soldered to the board. Some use USB-C for charging.

      The boards come in funny shapes and sizes, sure, but your first keyboard, monitor, and case are included in the original purchase!

      But I guess $200-300 is still more expensive than an average SBC, and they don’t usually let you swap microSD cards to use a different firmware. Maybe we are moving in that direction, though.

    3. I highly doubt it. The miniaturization process you’re talking about happened a decade ago, when desktops and laptops were both replaced with iPhones for the majority of users.

      The remaining users of desktops are using them because of the features of desktops – expandable storage and memory, customizable price-to-performance through component selection, support for high-throughput expansion cards like GPU’s, capture cards, accelerators, hundreds-of-channel audio interfaces, network cards (including networked processing and network storage host bus adapters)…

      I could see amd64 SOC’s being replaced with arm64 SOC’s, but given the price breakdown of boards like the macchiatobin (an ARM64 ITX board with PCIe, SATA and DDR4 dimm slot), I think we’re a long way off yet.

  4. If I get one of these I’ll wait for one that fixes the USB power issues, is redesigned to allow larger heat sinks, and also has the WIFi – HDMI interference issue fixed.

    1. Why? Works good right now. Cooling solutions are already in place. The USB power cords I use works great, so no problem there. Then if you use a monitor, don’t use the problem resolution. Or if you ‘really’ need to run at that resolution, plug in a wifi dongle if that is ultra important to you. Simple. For $35 – $55 you still have a perfectly working computer to have fun with. You’ll be ‘waiting’ forever if you use ‘minor’ problems as an excuse to not buy one :) … There will be always problem areas — and in this case, none of them are show stoppers.

  5. Nice article linked to:
    “USB improvements”: shows HDMI ports
    “SDRAM improvements”: shows power controller (I think, not wearing my glasses)
    Aaargh, want to be taken seriously? Did that get fixed before print?

  6. Look when pi started it was suppose to be a cheap computer around 30 dollars now it’s almost a100 and I can find cheap window pc that will run more and better then this your still keeping your cost low but charging the end user more

    1. True.. Surplus “Refurbished” towers here in Los Angeles start around $50 to $75 with the OS installed.
      Intel i7 towers, 8 gig ram, 2 TB hard drive, DVDrw Windows 10, $175 tops.
      Dual Xeon HP Z800, @3.50 Ghz, 24 gig ram, 2 TB drive DVDRW, Win10, $200 tops.
      Laptops too.. A nice HP “Ultra Book” with Intel i7, 8 ram, 256 SSD, $175
      I did buy the “Chip” prior to the meltdown.. Cheap enough.

  7. This past weekend, i grabbed one of my Pi B+ boards, burnt Raspian Buster onto a 4GB SD card, stuck it into a repurposed plastic case, did some simple configuration… and now I have a headless platform for a MQTT server to run home automation. This was easy to do, worked right away, and there’s buckets of useful info on the web.

    This ease and dependability is what keeps me loyal to Raspberry Pi’s. I would be less loyal to it if current or future versions require extra cooling or other kludges or add-ons to work reliably in a very basic application. Just a heads-up to Eben and the great rPi team.

Leave a Reply to Eben Upton Cancel reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.