> "set fan to speed needed for temp z"(pid control)

Why though? I generally don't care about the specific temperatures of my CPU and GPU, just that they don't get too warm, so for the CPU (AIO) I basically have "0% up until 45C, then increment up to 100% when it hits 90C" and the same for the GPU except it's always at 10%.

I guess I could figure out target temperatures, and do it the other way, but I'm not sure what the added complexity is for? The end results (I need at least) remains the same, cool down the hardware when it gets hotter, and for me, the simpler the better.

I also have two ambient temperature sensors in the chassi itself, right at the intake and the outtake. The intake one is just for monitoring if my room gets too warm so the computer won't be effective at cooling (as the summers here get really warm) and the outtake one is to check overall temperature and control the intake fans. In reality, I don't think I need to do even this, just the CPU+GPU temperature + set fan speed based on that feels simple enough to solve 99% of the things you'd like to be able to do here.

In my case the gpu fan was not changing speed automatically, (something, something, linux) which I discovered after an embarrassingly long time of poking at other things trying to figure out why my system was crashing so much. And now instead of doing the sane thing and figuring out how to get the fan speed map to work, I thought about it and came to this conclusion "I want my temps to be set value, say 45C, why not use a closed loop controller to find the fan speed necessary for that temp at any load?"

And now am about halfway through building pid fan control software and a janky gpu temp simulator so I can get some intuition on tuning the pid parameters before I set it on my actual gpu. you know, the fun part of computing. But now I am worried that perhaps there is a real reason nobody does it this way.

Aren't you ultimately gonna have to figure out why you cannot change the speed on the GPU, regardless of the approach? And then you're just back at square one?

I think no one is doing it that way, because there is simply no need for it. Sure, when I'm 3D printing some material it sometimes need the heatbed to be exactly 45C or whatever, but why would I care about the specific temperature of my GPU? As long as it's not throttled when GPU utilization is at 100%, I'm good to go.

KISS :)

This could be total nonsense and I am definitely out on a limb here, but I remember from my amateur crypto mining days (almost 15 years ago wow) that keeping everything relatively consistent is better for the longevity of your hardware (GPU/CPU mainly). Some people obsessed over getting temps as low as possible but that’s not very productive, diminishing returns and all of that. There is just generally a healthy temperature to keep things operating at most of the time, coupled with how hard you push the component in question (while not putting too much strain on your fans either). Basically: swinging/fluctuating is not good, consistency is good.

Again though I could be totally off. I just remember that being spread around as “conventional wisdom.”

You're speaking the truth, as this is not nonsense - it is correct. At the micro level, thermal expansion/contracting is occurring across the variety of the circuit board materials and components. Hot is fine (not too hot), but it is the consistency that makes for longetivity.

I limit power consumption profiles and clock speeds unless higher power is required, and combine that with an oversized cooling system - keeps regular temps consistent.

Yeah that was my understanding as well. Basically it comes down to how much you are warping/straining the materials. It all seems rather intuitive to me but this is not my area of expertise and just because something makes sense doesn’t mean it’s correct lol

> Question: is there a reason pid type control is never a thermal option? Or put another way, is there something about the desired thermal characteristics of a computer that make pid control undesirable?

I've only every seen something like this in really high reliability equipment because they're worried about repeated thermal expansion causing cracks in the boards/solder joints. There is, often, heaters available for use if the temperature gets too low. For most equipment I think that the juice just isn't worth the squeeze so it isn't done.

PID loops work best when you have active control for both heating and cooling. PIDs are also best when you have a single optimal point you’re aiming at, there’s a kind of fundamental mismatch here in that there isn’t really a “too cold” for your system, you’re just aiming to keep the system below an upper limit while hopefully keeping fan noise to a minimum (otherwise you’d just send it and run at max 24/7).

A simple PID wants to keep it at a setpoint, but it’s easy enough to layer a target change on top of it, some hysteresis so it’s not pulsing on and off, perhaps with control authority depending on the the target, etc. but then that gets complicated and usually manufacturers aren’t going to expose those sorts of tunings to the user.

For most people a fan curve is more obvious to work with and it’s largely good enough without the irritating behaviors insufficiently tuned control loops can exhibit.