No, it's not viable...it's one of the least efficient ways to cool something, and they're not coolers/heaters really ~ they're heat-pumps.

As they're heat pumps, you have to keep the 'hot' side of the peltier cooled (ambient)...ie; you're taking the heat from the CPU to the 'hot' side of the TEC, and you dissipate that energy with a heatsink/fan assembly.

To do that, you need to supply the peltier module with power ...let's just say 12v & 4A ....this is 48watts of power your PSU is going to need supply... + the 3-5watts the cooling fan is going to need...

...fact of the matter is, that 3-5watt cooling fan ontop of an i486 heatsink, is enough to keep the CPU cooled sufficiently by itself (presuming 25C ambient air temp) ...ie; saving you 48watts of power =)

If you're going to do it anyway.... the construction of the peltier modules is such, that the contact surface of whatever's being cooled/heated, must totally cover the area of the TEC's ceramic plate. In many cases, you need a heat spreader to move the heat evenly from the (typically) smaller CPU top face, to the usually larger ceramic plate of the TEC.... for example, around 50% of a module's junctions, are around the periphery of the cell...

....and that's why you need to cover the entire plate, with the source material being cooled/heated. Same goes for the 'hot' side ~ you have to use a bigger heatsink than you typically would for convection/forced air cooling, purely because you have to dissipate more heat energy, faster, as a consequence of the TEC's operation....thermal dynamics, physics ...really bite =) This is why these aren't a thing (still) in the computer hardware market (caveat industrial and other specialized uses, where typically they've had to (begrudgingly) use peltiers due to some size constraint and bemoan the additional power required to run them =)

One supposes, given where CNC machining has got to, you could cut some 'compound' heatsink arrangement, with an outer set of fins, and peltier in the middle with another heatsink ontop of it, connected to a fan at the top which forced air past both the outer fins and the inner stack on the TEC.... ie; self powered fan forced heatsink ....in the meantime, we just use the onboard or other power connector to power our 5watt fans... =^)

I'll have to dig out my 'portable peltier generator'... and take piccy

Chilled water sounds like the best option to me. Sadly, I mistakenly left my chiller behind when I moved home 12 months ago, but I was planning to replace it with a more powerful chiller that can cope with several CPUs worth of cooling (and probably a couple of high end GPUs), so no major loss. I did enjoy when I had a simpler chilled setup with it 😀

The general problem with peltier cooling is their tendency to die abruptly, leading to death of what they were pumping heat from a second or two later. There's usually no reserve heat capacity to allow graceful shutdown when failure detected, it's phut(pelt)phut(cpu)phut(mobo)bleepbleepbleepbleepbleep(smoke alarms) before you know what's going on. A chilled loop probably gets you a minute between cooling failure and CPU overheat.

I was a fool to expect anyone to read past the title.

I am aware of the greater risk in case the Peltier element fails - which is exactly why I wrote of my intention to, if possible, only have the element active during overclocked benchmark sessions once in a blue moon.
Everyone knows of the drastically worse power efficiency - which is why I said I'd watercool the module, and also, again, why I'm inquiring here about the possibility of only using the Peltier element while it is actually needed, and not to have it powered on during normal operation.

So first of all: here is a person successfully using this cooling method on the same model CPU I'm using:

https://youtu.be/LI1_RlVLhu8

And here is my setup so far, minus assembly and most parts of the water loop:

So as you can see:
- the Peltier element is fully covered on both sides: by the CPU and by the water block
- the power supply can easily handle both the rest of the system and Peltier element at the same time and has all methods of protection you'd expect of a modern PSU in case something goes wrong

Now back to actual question:

Suppose there was a powerful water loop connected to the waterblock on top of the Peltier element that is situated on top of the CPU.
1) Would the CPU survive everyday operation, in other words: running at default clocks, despite the Peltier element being powered off (but the water loop still powered on to cool it and thereby indirectly the CPU)?
1.1) Only if not, how could I switch between the Peltier element running at the minimal needed voltage (thus reducing risk of failure and loss of efficiency for most of the time I'm using the system) and full voltage (for whenever I want to run the system overclocked)?

Edited to fix 2 minor typos.

It depends on a particular CPU, but I assume you want to cool the 3V ones which absolutely have to be heatsinked or they'll cook to death sooner or later. Peltier itself makes for a pretty poor heatsink and has significant thermal resistance so I don't think you want it completely powered off and sitting there, but you can possibly power it from a lower voltage to make up for its thermal resistance which would let the actual heatsink part do its job with CPU not cooking underneath. Easiest way is 5V to try, maybe it'll be enough. You can wire a switch that gives the peltier 5 or 12V depending on what you want to do. There's even an option to stick it between 5 and 12V for effective 7V but that makes things more complicated, all else will require active bits like some switch mode regulator with adjustable voltage.

The undriven or just died peltier is like having a slab of glass in there, I wouldn't even want to put a square of glass on top of a CPU that doesn't need a heatsink.

The way I'd want to include an occasional use peltier, would be a triple waterblock setup, radiator to peltier cold side block, that outlets to cpu block, which outlets into peltier hotside block, which outlets back to radiator. Well you need a pump in there one side or the other of course. You can either stack that up on top of CPU, or just cpu block and peltier stack beside it. You should have the peltier, pump and radiator fan on a separately switched circuit from motherboard power of course, because if you are trying to get speeds the peltier is absolutely necessary for, then it won't boot until after a few minutes of pre-chill cooldown. If it will boot from room temperature, then this whole setup is unnecessary, and regular water loop is adequate.

It is just fine to use Peltiers for cooling and long-term usage.
link <- 386
link <- 486 (also, check the following discussion), this exactly your case

Danger Manfred wrote on 2024-01-02, 22:14:

2) If not, do you have any idea how I could change the Peltier element's voltage for normal operation and overclocking mode, so that in normal operation it doesn't go below ambient or at leat not below the dew point?

Quite simple - measure the ambient temperature and relative humidity, from that you can calculate current dew point.
Measure temperature at the cold side of the peltier... and control the input power by PWM regulator to keep the temperature above dew point. Should be doable with Arduino or something similar.

Many years ago I made peltier based chiller. So two separate water cooling loops - one with huge radiator to cool the peltiers and second to cool the CPU. I also intended to control the power to prevent water condensation. In the end I never did, because even 3x150W peltiers were too weak to reach that low temperatures (with reasonably powerful CPU) so water condensation was never an issue.

Anyway - for your use case, I think it would be best to just assemble the PC with peltier directly cooling the CPU, put some insulation inside the socket, around the socket and possibly back side of the board. Run the benchmarks you need... and then put regular air cooler back on. It is too problematic to have peltier installed permanently.

I had an Am5x86 running with a Peltier element but the moisture was not feasible to keep track of.
For experimental purpose a nice option though... sandwich was cpu -> peltier -> heatsink

I just remembered I also have two old commercial Peltier CPU coolers for the socket 478.
The ActiveCool AC4G-D, I bought them as new old stock for 15€ each because I saw them and thought what cool Oddware that is. Apparently they poll the temperature of the coldplate between CPU & Peltier element with a probe 40 times per second, and adjust the Peltier element so the temperature doesn't fall below 28°C. Of course that's too much for good OC results, but I wonder if there's a way to change that number, either by making the probe report an offset number (dirty solution) or finding out where that exact number is stored and changing it to 10 or so.
Since I have a spare, and I think there's even still some stock where I bought it, I'm not too worried about breaking one for science.

Danger Manfred, up to early Pentiums and P-equivalents most of the motherboards and their BIOSes (socket 7 with AT or AT+ATX PSU) support "turbo" button function (I have one PII/III slot 1 mobo that also support it). Maybe you can figure out something to use the "turbo" button for CPU O'C plus Peltier element - on.

Turbo switches really need to be renamed the slow and unslow switch, because they're backwards to what everyone thinks they do.

Hm. I vaguely remember that on my father's 386DX-40 PC, the seven segment display would say 40 if the button was pressed and something lower if de-pressed (25 or 33). 🤷‍♂️

BitWrangler wrote on 2024-01-04, 02:52:

Turbo switches really need to be renamed the slow and unslow switch, because they're backwards to what everyone thinks they do.

It's just on-off toggle switch. Which position is which depends on the used switch itself (NO vs NC contacts). And I think there was an internal BIOS setting (non accessible in BIOS user settings menu) that also alters the on/off "tubro" jumper/switch. I don't remember if on my AT PC case this switch was pressed or depressed for the "turbo on" mode, but it can be used for something else i.e. relay switching.

Yes the switch is just a switch and maybe supplied as a 3 pin in a case, such that you can use the normally open or normally closed side. BIOS support of it is a bit irrelevant if you just want to repurpose the switch. You can plug it to a FSB or multiplier jumper block to go between two settings that are one jumper different.

Involving the motherboard's turbo function in an overclocking setup is problematic though. Late AT class boards usually have 8mhz, ref/2 or sysclk/4 outputs they switch to when the "turbo" feature is active, which is the deturbo where it sets the slow "compatibility speed", this is baked into the clock chip. In 486 class this is "too slow to 486" so really annoying to try to use as everyday speed vs the overclocked setting. If however the late MMX/PII turbo method is used which is turning off L2 or L1, this is not de-overclocked, but overclocked but also slow, which may still leave things hot or crashy.

I mean, it's possible to re-wire "turbo switch" to control some DIY PCB with relays, which controls the Peltier element and the hardware side of CPU's overclock (for example - changing resistor combinations for PLL chip). It will be motherboard specific mod, but seems doable.

Check out the construction of the Coolermaster V10. That might inspire some custom with/without active TEC cooling. In short, the TEC is mounted on it's own heat pipes alongside the normal radiator ...