News 64-bit Raspberry Pi 3 launches as first Wi-Fi-enabled board

Well, now, that's not actually true, is it? The bundled education software is pretty much all designed to be run on the Pi itself, and the majority of it runs in the desktop. Stuff like Scratch, Wolfram Alpha, BlueJ - these all absolutely *chug* on the original 700MHz single-core Pi. They're usable, just about, but frustrating - and especially so if you're running a web browser in the background for reading tutorials or help pages. The Pi 2 and Pi 3 bring more grunt to the table, and that makes everything a lot more responsive - which means fewer frustrated kids clicking a bajillion times and making something break.

Then there's people doing compiled languages: compilation takes about a tenth the time - or less - on a Pi 3 than on an original Pi. Sure, you can cut it down considerably further by running a cross-compile environment (or qemu or similar) on a desktop, but at that point you might as well just be using the desktop to learn programming and save yourself the £30.

 

So, I've been playing with the Pi 3 for a couple of months now, but I've only just got around to sticking it under the thermal camera ('cos I've got a Hobby Tech column due tomorrow.) This is the result:



Yes, that's a chip temperature of 97°C in a 16°C room - a delta-T of a ridiculous 81°C. Without a case. That's at full CPU load but minimal GPU, so if I were really stressing it we'd be seeing worse.

If you're buying a Pi 3, do not put it in an existing case - especially compact fully-covered slice-style cases like the Pimoroni PiBow - without a heatsink, and even then make sure you're keeping an eye on the temperatures 'cos it's likely you're going to need to make a cut-out so it can breathe. (Incidentally, that's exactly what Pimoroni has done: the new PiBow has a hole over the SoC and you can be a heatsink that pokes up through said hole for an extra quid.)

 

85°C; it's supposed to throttle at 80°C, but the internal temperature sensor was convinced it was below that even as I was taking the photo; roughly five minutes. In that order.

 

Right, so, I'm bepuzzled. I have two Pi 3s. They're both V1.2, but from different batches: one is a pre-production sample with debug header on its arse, the other is a boxed retail RS model. I can make both reach temperatures well in excess of where it should throttle using SysBench. Both show a very different reading when I read them externally than from their internal temperature sensors when under load (though the readings match at idle). I've verified my findings in three ways: a cheap IR thermometer smushed up against the SoC to account for the 8:1 measurement area growth ratio, which reads higher than the internal sensor; a cheap K-type contact probe connected to a multimeter, which reads higher than the internal sensor; and a really chuffin' expensive, fully calibrated Flir thermal camera, which reads higher than the internal sensor (and yes, I do know how to operate it and have correctly set the distance, reflected temperature and thermal emissivity values. I also pointed it at the kettle with a strip of masking tape on the side to make sure it hadn't got busted somehow. It's fine.)

I also poked the chip with my finger, and it felt unpleasant enough that I'm willing to believe my measurements.

Then there's this guy, who also has two Pi 3s from different batches. His don't go above 60°C and the external reading is never more than a few degrees away from the internal sensor. As I did, he's used multiple methods to confirm that - including poking the chip, which was 'ow' but not 'argh.'

Then there's the Foundation, which has been doing its own testing. Their test Pi, running the same benchmark, hits 83°C externally-measured - lower than mine, hotter than Make's - but correctly throttles as a result, dropping to 77°C before climbing back to 83°C to throttle once more.

So, explanations on a postcard, please.

 

Just give it a slap across the chops with this at the terminal:

Code:

sudo apt-get update && sudo apt-get install sysbench -y
nohup sysbench --num-threads=8 --test=cpu --cpu-max-prime=10000000000 run &
watch cat /sys/class/thermal/thermal_zone0/temp

If it reckons it's sub-80, stick yer finger on it and see if you agree.

Every home should have one!

I'd say so, aye.

 

The Parallella runs Ubuntu, from memory, so you've got your full-fat OS there.

It's a good time to be alive if you like playing with small, low-power, affordable computing hardware, innit?

 

Got an explanation for the guy from Make seeing a peak of 60°C - if you feed the Pi 3 an inadequate power supply or a crap USB cable, it throttles down to 600MHz regardless of temperature. The only people I'm seeing saying theirs hits anything less than 70°C unmodified are suffering from that problem - and while the guy from Make hasn't responded to my messages, I'm willing to bet that was the case with his Pis as well.

Still trying to find people with accurate means of measuring the external temperature of the chip. I can't be the only Pi 3 owner with a thermal camera, surely?!

 

You should do, but at least one person claims they had no rainbow (but did see the power light flash, which is the other warning of low power).

Yeah, it'd have been closer to 60 if it were running at 600MHz. You can check easily, though:

Code:

sudo apt-get update && sudo apt-get install sysbench -y
nohup sysbench --num-threads=8 --test=cpu --cpu-max-prime=10000000000 run &
watch '(vcgencmd measure_temp; vcgencmd measure_clock arm)'

That'll run a CPU stress test while printing the reported temperature and clock speed of the chip. If you're seeing it drop to 600MHz, that's power throttling; thermal throttling should see it drop in increments, not by half.