Your computers and this hot weather

It’s been a long time since I’ve had a computer struggle in this weather, but my Pi 3 staring having stability problems.

With this little bodge, RO is nice and stable again.

How are you all fairing?

Wrong hemisphere for me. However I fully approve of type cooling solution and have used it many times.

Semi related, it was a constant struggle to keep the egg incubator below 39*C for a few months with the cooler on, a fan pointed directly at it, and those silver bubble wrap windscreen heat shield things on the windows. By extension a lot of electronics fail every summer.

By extension a lot of electronics fail every summer.

In my experience, electronics have gotten a lot more hardy in the past 7 years. I guess RPi problems are because it’s not very well engineered, but it’s cheap enough. You get what you pay for.

I guess RPi problems are because it’s not very well engineered, but it’s cheap enough. You get what you pay for.

Engineered with two aims in mind

1. Be cheap enough to enable almost anyone to afford it
2. Be simple enough to hack around and yet have quite a lot of features

The drive was to get people doing low level computer electronics, the initial target was schools but that widened. I think it succeeded, at least with the interest level.

I suppose the real question is how many users that have a unit that has problems throw it away and get another because it’s cheap and how many poke at the components trying to fix it while using another one as their main unit because they are cheap.

Those poke around characters are the real success.

I suppose the real question is how many users that have a unit that has problems throw it away and get another because it’s cheap and how many poke at the components trying to fix it while using another one as their main unit because they are cheap.

Unfortunately, it is my main RO comp so I’m stuck with poking it (although I hate throwing things away if they could be fixed). If I can scrape together the pesos, I’ll probably build a titanium based rig, as it seems quite well engineered.

Unfortunately, it is my main RO comp so I’m stuck with poking it (although I hate throwing things away if they could be fixed).

It will come as no surprise that I had you labelled as being in the second (poke & fix) group. Probably a high number of those around here anyway.

It will come as no surprise that I had you labelled as being in the second (poke & fix) group.

Thank you :)

Probably a high number of those around here anyway.

I tend to agree.

I guess RPi problems are because it’s not very well engineered,

I wonder if it is a wider problem of the problems associated with having memory stuck on top of the SoC. Makes it simple to connect the two, but I can’t imagine heatsinks being terribly effective at cooling the core when attached to the memory…!

That said, my Pi doesn’t run that hot. I don’t overclock it or anything. For the most of the time, there’s no real difference between 700MHz and higher. I’m typing this, a billion instructions of “nothing happened” pass between each keystroke.

The Vonets WiFi, on the other hand, runs hot. When I probed the thing last September, it shot up to ~73C (and climbing) when powered up, case open, on my keyboard. A hot day in an enclosed space, it’s no wonder it cooked itself. Likewise, one of the power adaptors gave up in the heat. So now I have a 2A capable supply running a ~1A (or less) load. Run things at their capacity, they’ll heat up.

As for the Vonets, here’s my bodge:

(full size – 2560×1440, ~128K (QHD screenshot from FHD video))

I had taken the fan out of my case (with a Pi Zero in) as a bit of an experiment and never put it back in. On Friday it hit 51°C and then froze.

Fan back in and it never got above 33°C all Saturday.

I don’t think its the Pi thats generating all the heat though…I have my suspicions that it might be a less-than-efficient voltage regulator…

That said, my Pi doesn’t run that hot. I don’t overclock it or anything.

Me neither. It’s on a lot so I need the chip as stable as possible, and not looking for a shortened life. Mine is the Pi 3, though.

I’m typing this, a billion instructions of “nothing happened” pass between each keystroke.

True, but overclocking does speed up SD read & write speeds on RPi.

On Friday it hit 51°C and then froze.

Is that case temp?

I have my suspicions that it might be a less-than-efficient voltage regulator

They do get quite hot.

Is that case temp?

Yep, I have a thermal sensor from CJE that is wired to the GPIO and sits next to the Pi Zero…its close to the die but certainly not close enough for me to claim I was measuring the die temperature. Still without a fan the whole thing would have probably got to a similar sort of level. The case itself is magnesium alloy so was doing a reasonable effort of radiating the heat and up to Friday had worked just fine!

In warmer weather my Pi 3 can’t manage tasks like decompressing a file without getting rather hot. When it’s temp gets above the high 80s I start take action to reduce load.

It’s common for the ADSL router to become unstable and crash during the warmer months, especially the day. I have a pile of fallen soldiers that didn’t stop working completely but started to display erratic behaviour and couldn’t be rectified.

I’m not crazy on the RPi Foundation’s stance of the Pi being fine if the temp stays below the failure point of the silicon.

Glen, re the voltage regulator, it’s entirely possible. However the design was improved a lot over the B. I found my Zero generates a lot of heat too.

I think heat is an SBC problem in general. The AllWinner chips get toasty on my OPi whatevers. The H3 without a heatsink has to throttle back and shut down cores to stop itself from cooking on boot! Way better with a heatsink of course. Just saying putting a lot of fast semiconductor into a small space will cause a lot of heat. It’s not so much a RPi design issue. Just something that can’t be avoided. The cases look nice but they aren’t great for what’s inside.

IME case temp can be as much as 20 deg. C less than die temp as measured by an app such as CPUTempMon. I use the latter (my Pi2 is mildly overclocked). In summer die temp tops out at 60-65 deg. I made additional ventilation holes in the plastic case. It seems quite stable.

It’s common for the ADSL router to become unstable and crash during the warmer months, especially the day.

I’m sure your right. I tend to avoid consumer-grade equipment. When you’ve worked as a net/sysadmin, the last thing you want to do is faff around with cheaply made and badly engineered networking equipment. I used to use Cisco until BT upgraded to 21CN, then I switched to OpenBSD-based router.

I think heat is an SBC problem in general.

Actually there are a ton of excellent SBCs. I think the problem is development boards. They are boards “that just work” most of the time so you can develop your applications and design a similar SBC for your own purposes. They are not hardened devices. The problem is RO survives on development boards (IGEPv5 and arMX6). This is why I was so excited about Elesar’s Titanium board.

In summer die temp tops out at 60-65 deg.

That’s incredible. Do you have air con or live in Greenland or something. :D

I made additional ventilation holes in the plastic case.

I am tempted to cut a hole and wire in a 5V fan intake. Keeps cool and creates positive pressure. I don’t think my RTC has a temp sensor. If it did, it’s would give false readings because it’s so close to the heatsink.

Do you have air con or live in Greenland

No and no (Buckinghamshire is about 54deg N, I believe). The Pi is in a small plastic case, liberally drilled top and bottom with c.8mm holes, attached via a VESA mount to the back of the monitor.

die temp tops out at 60-65 deg.

That’s what it’s been recently: from memory it never exceeded 70deg last year.

the last thing you want to do is faff around with cheaply made and badly engineered networking equipment. I used to use Cisco

Not exactly a shining example of build quality

In summer die temp tops out at 60-65 deg.

That’s incredible. Do you have air con or live in Greenland or something. :D

333.15 – 338.15 suit you any better? :)

liberally drilled top and bottom with c.8mm holes

I keep meaning to figure out a way to cool the bottom. The RAM chip gets quite warm.

The idea behind holes top and bottom was to allow air flow /through/ the case, rather than simply allow heat to escape, but I have no idea if that is what is actually happening: I suppose I could tape over the holes selectively and see if the temp changed, but life’s too short… If your temps are higher that’s probably down to usage.

I don’t know if the Pi has any support for auto-throttling the CPU?

I just discovered today that one of our HTPCs had overheated and frozen (its a Pi 2 or 3…can’t remember which way round I have them at the moment). It lives in a cupboard with an AMD AM1 based computer that is the satellite decoder/server.

The cupboard can get quite warm and I do intend to drill some ventillation holes in the back but I have found that the AMD server does an excellent job of automatically slowing down the CPU when things get warm and has thus far been incredibly stable for consumer grade stuff (several years of near continuous operation and I’ve not exactly been kind to it with the amount of load I’ve put on the CPU!)

I’m guessing that since ARM have up to this point been relatively efficient that there hasn’t been a great deal of time invested in clever ways to manage the thermal performance?

I’m guessing that since ARM have up to this point been relatively efficient that there hasn’t been a great deal of time invested in clever ways to manage the thermal performance?

You would probably want to read the documentation on the Portable module but there are current utilities that help manage the settings.

ARM SoCs have generally featured at least some form of CPU throttling for at least the past 10 years. The race to make thinner yet more powerful smartphones means that managing heat dissipation is a serious concern; any high-end smartphone is unlikely to be able to manage running at full power for more than a minute or two before it’s forced to drop the clock speeds to avoid cooking itself.

You also need to remember that heat is typically wasted energy. Less heat = less power = more energy efficient, and energy efficiency is one of the main driving factors behind ARM’s designs. So even before overheating became an issue they were focusing on generating low-heat designs (right back to the ARM1).

As to how all this impacts your situation…

• RISC OS has no builtin temperature-based throttling. Only some simple throttling based on CPU usage, which is managed by the Wimp. However there are utilities which will implement temperature-based throttling (a combination of CPUClock and CPUTmpMon?)
• The Pi’s firmware has builtin temperature-based throttling. Unless you’ve disabled it in your config.txt, when it kicks in it will override whatever CPU speed the ARM is requesting. And if the temperature remains too high then it will show a warning icon on screen.
  • If necessary you can also manually specify alternate min & max CPU speeds in config.txt, which will then be picked up by both the firmware and OS. http://elinux.org/RPiconfig#Overclocking_options (take the defaults listed on that page with a pinch of salt, each model has different defaults, for the CPU at least)

Only some simple throttling based on CPU usage, which is managed by the Wimp

Wouldn’t the kernel or module do a better job?

Wouldn’t the kernel or module do a better job?

You’d have thought so, as like that it’ll be able to work all the time instead of just when the Wimp is running (and not blocked by a busy app). Plus as long as TickerV still works (ie mouse pointer moves etc) the machine will still be able to throttle itself if something goes horribly wrong and the user end has frozen up… Or its sitting unattended with an error box on the screen, etc etc.

more powerful smartphones means that managing heat dissipation is a serious concern;

Certainly my S7 doesn’t run the primary ARM cores at the 2+GHz, quite a bit slower. The M0 cores run at a different speed depending on whether two or four are in use. I’m not sure it’s possible to actually run this thing with all cores at max speed.

Doing intensive tasks such as navigating with Google and GPS in a weak signal area – you find out quickly that the fancy metal edge between the two bits of glass isn’t there for pretty, it’s the heatsink!

Or its sitting unattended with an error box on the screen

That happened to me and found the box locked up. Make sense now. Any chance of dropping cooperative multitasking in favour of preemptive?