My computer was overheating, and all I really wanted to know was: "are the fans actually working properly?". Or how AI still manages to blow me away every single day...
After 8 years of intensive use, the fans in my desktop computer, a Surface Studio 2, started making an annoying rattling noise. Original parts were nowhere to be found, so with ChatGPT as my guide, I searched for a Chinese aftermarket alternative. My computer was already crashing fairly randomly from time to time, but I thought: "oh well, these things happen, probably nothing serious, restart and off we go again."
The new fans were silent, but the crashes kept happening, more and more often, until the computer became almost unusable. The outside also felt very hot. So: opened it up, cleaned out the dust and started monitoring temperatures with a free tool (HWiNFO).
Unfortunately, cleaning out the dust alone did not solve it; heat remained a problem. With ChatGPT as my virtual PC engineer, I ended up diagnosing ageing thermal paste, which is supposed to ensure that heat is transferred efficiently from the processor to the heatsink. Picked up isopropyl alcohol at Hubo, again on ChatGPT's advice, removed the old paste and applied new. Temperatures dropped significantly straight away. ChatGPT for the win, again!
Still, I kept wondering: are those fans spinning fast enough? HWiNFO shows a thousand and one data points, but whether, and how fast, those fans were spinning was - strangely enough - not displayed for my computer. Other specialised tools could not detect the fans either.
So together with GPT-5.6 Sol, I started figuring out how Microsoft controls this device. And that is where my simple question escalated rather quickly... bear with me.
We started with Windows devices, drivers and services. With some IT experience, you can still follow along. But then came ACPI tables, hardware IDs and Microsoft's Surface Serial Hub. Until we started dissecting Windows files that show nothing but strange characters in a text editor: the firmware of the embedded controller in my computer.
How did I know that? I didn't, but GPT did, after doing its own research. It dug into machine instructions, hexadecimal byte patterns, ARM code and memory addresses. All to figure out: "What does this actually mean?" I was just standing there watching...
Still with me? I had long since lost the plot. But GPT-5.6 Sol was remarkably good at keeping track of it all.
And then we found it, or rather GPT did: the commands to read the fan speeds. Including the logic Microsoft uses to control the cooling.
It was no "walk in the park". There were dead ends, hypotheses and checks. GPT kept going long after my technical knowledge had run out, deep into machine code that a normal person would barely be able to make sense of without reverse engineering.
All because I thought: "Are those fans actually spinning fast enough?"