Do not trust any readings from those thermal images; whoever took them doesn't know how to use a thermal camera.
The short version: different surfaces have different "thermal emissivity" values, which you can think of a bit like reflectivity: if you look at a flat black surface you can see it's black, but if you look at a mirror all you can see is your own reflection. Thermal emissivity is that, but for infrared radiation rather than visible light - and effectively the same, in that shiny light-coloured surfaces have lower emissivity than matte dark-coloured surfaces. (That's a simplification, but it'll do.)
What that person has done is taken an untreated Raspberry Pi, which has a range of surface finishes and materials, and pointed a thermal camera at it. See how the system-on-chip, one of the hottest parts of the system, is a dark blue? That's because it's covered in a shiny silver heatspreader - a material with much lower thermal emissivity than the rest of the board. See how the PCB around it looks much hotter? It's not.
What you have to do before you can actually use a thermal camera properly is ensure everything you're pointing it at has the same, known thermal emissivity. In the case of a PCB like the Raspberry Pi, that means coating it in something like zinc oxide. This has zero impact on the temperatures, but provides a uniform emissivity for all components - making it possible to get actually usable data out of your efforts, rather than just a pretty but meaningless picture.
Here's what a Raspberry Pi 4 looks like under a thermal camera when it's done properly:
That's thermal data from a FLIR ETS320 inspection camera overlaid on an edge-enhanced visible light shot for detail. See how the SoC's heatspreader is the same temperature as the PCB around it? See similarly how the smaller shiny metal components are also not showing as "cold" when they're actually hotter than the PCB? And how the small micro-HDMI connectors are hotter than the shields around the Ethernet and USB ports? That's all stuff that wouldn't show without the coating to standardise the thermal emissivity. Spot readings (as in the picture you shared) would also be wrong for any surface that doesn't match the emissivity set in the camera.
Didn't mean to rant for quite so long, but it always annoys me when I see broken thermal imagery used to prove something or other!
Statistics: Posted by Gareth Halfacree — Tue Jan 28, 2025 10:12 am