How can I prevent my work computers turning my home into an oven?
[Edit] Details on the plan as it stands are here, potentially using one of these heat pumps.
Looking for advice before the weather starts to warm up! I'm running multiple GPUs for dev work in my small home office, and it's pretty much equivalent to having a fan heater running all day. Right now that's actually a bonus, but it really won't be in a couple of months.
The big heat generating components are all water cooled - partly just to fit them in a sensible amount of space, and partly because I figured I'd end up with exactly this problem and being able to physically pipe the heat elsewhere (ideally outside) would probably be necessary. The bit I'm trying to figure out now is how to actually make that happen...
Ideas so far:
-
Stick an air duct on the back of the radiator and hang the duct out of the window: straightforward but messy, may be counterproductive depending how hot it is outside and how well I can rig up some kind of baffle between the open window and the duct.
-
Put the whole radiator outside and drill a couple of small holes in the wall for the pipes: this was my first thought, but PC radiators and fans definitely aren't rated for outdoor use, and I'm not sure where to start looking for something that would be designed for that while still being suitable to hook up to the computer waterblocks. I'm also concerned about condensation on the electronics if the coolant gets below indoor ambient temperature overnight.
-
Set up some kind of actual exterior radiator (or heat pump?) outside, and use a heat exchanger between that and the PC cooling loop: seems more like the "proper" way to be doing this, but it's well outside my area of expertise and feels like there would be a lot of potentially expensive stumbling blocks. Also still has the condensation problem, I think.
-
Just crank up the air conditioning: I'm not actually sure if the unit I've got has enough capacity, and it definitely seems wasteful to heat up the air and then use more energy cooling it again rather than dumping the heat directly outside, but maybe I'm wrong there!
I'm in a kind of awkward middle ground: I'm running enough hardware that this is getting to be an issue beyond what you'd get with normal end user setups, and I'm willing to put some money into fixing it (it's affecting my job and my home, after all!), but I'm self employed and nowhere close to the industrial or datacenter scale that tends to come up when searching for solutions.
Has anyone dealt with this themselves, or come across small office/homelab scale solutions that might work?
I don't think you should do this and instead just run these workloads in a datacenter, but I did think of a fun janky solution that may work.
You're already water cooling, so you can just buy a small split ac unit. These systems generally have the compresor in an outside unit and an evaporator/air handler in the inside unit. Put the outside unit outside, take apart the inside unit and ditch the fan/air handler components. Attach the evaporator loop to a heat exchanger block for your water cooling system. Attach the coolant loop between the two units
Then viola, you have a refrigeration unit that pumps heat from your PC outside. It has a thermostat controlled by a remote to keep the components whatever temperature you like, and it will keep them frosty no matter what the temperature is like outside.
I think I've found the missing link: swimming pool heat pumps! They look pretty much identical to the condenser for a split AC, similarly affordable, intended for outdoor installation, but they're already designed to circulate temperature controlled water rather than air.
Turns out most of them have a cooling mode for things like spa plunge pools as well as the traditional pool heating mode, and some also have a constant temperature setting that auto switches between heating and cooling mode if required. Just need to set that a degree or two above indoor ambient and it looks like this could rather neatly solve all my problems here.
I might even go as far as saying it'd be a fairly professional setup - similar to the water chillers for laboratory lasers I've seen a few times, except the pool version is outdoor rated and costs £500 rather than £20k for the lab version.
Big thank you for your help on this one, the combination of the AC idea and the liquid heat exchanger below put me absolutely right on track here. Still some more research to do, but assuming no major pitfalls I'm thinking pool heat pump outside, and then still go with the dual loop and liquid heat exchanger (with water in both sides) just to make it easier to disconnect the external side in winter and deliberately dump the heat inside instead.
Nice! Please post some updates when you get started. I'm invested in how this goes now!
Absolutely will do! Might end up being a month or two before I actually get the hardware in place, but I'll make sure to show off the results and/or any particularly amusing failures along the way.
I really, really, don't recommend doing this. Pool heat pumps are extremely expensive; doing a quick search had me find one for $4500 if you do one that both heats and cools. In the meanwhile if you get a regular mini split you can afford to have a cassette in every room of a mansion and still spend half that. I'm really not sure where you found a 500 Euro one because on a search I couldn't find anything under $2000.
Even though a regular AC for the room might be less efficient overall you do get the benefit of keeping your house climate-controlled, so it's not like you're really losing out on the proposition. The only downside you have is the noise, but if you use a pool heat exchanger you'll have significantly more noise because you'll be dealing with a compressor! I assume you'll be putting it outside but no matter what you do you'll be dealing with a lot of effort for something that can be accomplished a lot easier with reliable off-the-shelf solutions.
This is my current front runner, although I’m still researching: https://heatpumps4pools.com/product/thermotec-mini-heat-pump/ - looks like just the job unless I’m missing something? There are others in that price range under the brands Flotide and SunSpring, as well as quite a few generic ones on AliExpress (but they seem to be heavy enough that once shipping is factored in it’s a better bet to buy locally from a supplier that imports them in bulk).
I emailed the supplier of that one above this morning basically saying “here’s my idea, is it broadly sane?” so I’ll see what they think when they reply, but if you see any pitfalls then absolutely let me know, because I definitely don’t want to end up making a mess of this or wasting my money.
[Edit] Another similar option I hadn’t spotted earlier: https://www.tradeinn.com/swiminn/en/poolex-nano-action-a5r-2020-5kw-2.5-m--h-inverter-heat-pump/139816348/p - broadly similar, but this one has more detailed specs about cooling capacity on the datasheet rather than just a vague reference to “cooling mode”.
Here's the thing: Heat pumps are relatively simple machines. It's really hard to justify why they are so expensive in my opinion, so it doesn't surprise me that you managed to find cheaper versions of them. I will say that you should be careful about what you buy because that first one you linked only did heating, not cooling.
Don't expect a good answer from the suppliers. They sell them and they should know about how to use them for their intended application, but they aren't going to necessarily be able to give you any hints outside of it's intended application. Depending on the supplier, they might not even know how it works.
I don't think what you're doing is too dangerous to your health, but the condensation and risk of leaks are the biggest thing, but the power safety circuitry should (very theoretically) keep you safe. It's just that it's not a "reasonable" thing to do. If you're approaching this as a hobby project and are willing to throw money on it and potentially lose on it, that's fine.
That being said, I have pool industry experience that will help you. Every heater will have some sort of safety switch either built into it if it's not intended to be integrated into a larger system. That first one, for instance, has a flow switch. It's probably more common these days to have pressure switches. That is probably going to be your biggest problem, as you will need to have enough flow or pressure to activate that switch in order to activate the heat pump.
DO NOT CIRCUMVENT THE SAFETY SWITCH. And if you buy one that doesn't have one for whatever reason, add one. It's there to prevent the heat pump from overheating/freezing, because they are designed to run when water is flowing through it. I was going to suggest replacing the switch with an adjustable pressure switch, but that's probably a bad idea too, because having the thing freeze on you would be a very bad thing.
You've absolutely got more experience with these units than I do, and it's pretty clear you don't think it's a good idea, so I'd really appreciate digging a bit deeper if you don't mind. Want to figure out if I'm missing anything important, or if maybe I just gave the impression it's going to be more hacky than it is with all the various threads here where we were floating early ideas.
I'm not planning to modify the heat pump at all - if anything it seems to me like I'm using it for its intended purpose: keeping a container of water at a relatively constant temperature, with a constant flow running through it. I've been cross referencing manufacturer docs, AliExpress listings, etc. so I think I've built up a decent-ish idea of the standard capabilities, but it's absolutely not obvious from every listing! You have to go to the datasheet to get actual confirmation that the first one does cooling, for example; other than that it's just a one line reference to "cooling mode" in the manual.
To an extent I am treating it as a hobby project, and the money involved is a good amount less than the alternative so I'm willing to take some chances, but ultimately I do want this to be a genuine solution to my problem and I'm happy to scrap the idea if that's not going to happen.
The full plan so far:
Mount a liquid:liquid heat exchanger on an interior wall and run a couple of PVC pipes from one side of it to the outside. Two 28mm (~1") holes through the exterior wall should be the only permanent changes needed, so that's minor enough not to worry me.
Run a pipe from the heat exchanger into the heat pump: at this point it seems like this is the expected mode of operation from the heat pump's point of view, assuming the circulation pump later in the loop is specced correctly? On that note, thank you for mentioning the pressure/flow safety switch, that's exactly the kind of unknown-unknown I'm very grateful to be aware of! Absolutely wouldn't plan on modifying a safety feature of whatever setup ends up coming together here.
Pipe the outlet of the heat pump into some kind of container to act as a reservoir, to buffer any temperature fluctuations and give air bubbles somewhere to go. I'm thinking just a cheap insulated beer cooler, like you'd take to the beach - drill a hole in the top for an inlet pipe, one in the bottom for an outlet pipe, job done. Size is an open question; larger would be better (compressor needs to kick on less often, operation is more similar to the intended use case where the "reservoir" is a pool or hot tub), but I don't want to take up the whole balcony here. I was thinking somewhere around 40 - 60L, but that's a guesstimate based on what's available and wouldn't be too obtrusive, so I need to figure out whether there are any other limiting factors here.
Run the outlet of the reservoir into a circulation pump, and then back into the heat exchanger to complete the outdoor loop: this is one of the parts I haven't figured out in detail yet. Need to consider flow rate, continuous duty cycle, etc. so it's high on my list to research further, and I'm very open to suggestions!
The existing computer watercooling loop then just plugs into the other side of the heat exchanger in place of the radiator it has now. I'm planning to put flow and water temperature meters on both loops with an automatic shutdown on the computers if either fails, tie the heat pump into Home Assistant via bluetooth to keep it set above indoor ambient, and already have an IP66 rated power connection on the balcony I can use to run the outdoor components.
Thank you to anyone who's made it this far into my stream-of-consciousness planning process! So, what have I missed?
[Edit] Cc'ing @papasquat, @vord, @derekiscool, @ChingShih, @vektor, and @Weldawadyathink in case anyone has input, since you also all had thoughts on the heat pump idea specifically - thanks all!
Hmm. Actually the heat exchange might make it work. I was most worried about the pressure blowing your pc fittings. Do keep in mind that you will need to also have a pump for the heat pump in addition to the pump for your pc’s cooling, and having a long loop for your PC cooling might mean needing to upgrade that pump as well for the extra head that will introduce. Any angles or restriction on the pipes are also going to add head. Also the heat pump line pump is probably going to need to run 24/7; it’s generally bad to power cycle them too often.
I will however reiterate that this is going to be much more expensive, complicated, and failure prone than simply installing an air conditioner. This is the last time I swear.
Since I’m not sure if people get notified about edited in mentions, Cc'ing @papasquat, @vord, @derekiscool, @ChingShih, @vektor, and @Weldawadyathink. Sorry for bugging y’all if you already got this.
Much appreciated, and I'm glad to hear that it's not sounding like quite such a bad plan anymore! Very good points on loop length and pump configuration - I haven't figured out what to do about the exterior loop pump at all yet: whether that should be something from a pool supplier, or an aquarium submersible pump in the outdoor reservoir, or something else entirely. That's my next todo item for research, so it'll be interesting to see what comes up.
In terms of just using air conditioning, fair enough, I'll give you that one in the general case! For my specific (small) living space layout and for my own comfort with temperature / humidity / air flow / noise, I'd rather keep the two separate if I can - physically piping the computer heat outside just seems more appealing as long as I can do it in a way that's not totally mad.
I do already have A/C installed, and it'd definitely need an upgrade to deal with summer heat and computer heat simultaneously anyway; it was struggling a bit last year and I was drawing a good bit less power back then.
An air conditioner is piping the heat from your computer outside though. It’s just also piping out the ambient heat as well.
If you’re worried about efficiency then you would just install an air conditioner for that specific room rather than a central unit. A mini split would be perfect for such an application.
Fair, I should’ve said piping it directly outside - as in, not using the air I’m sitting in as a transfer medium. My place is mostly open plan and superimposed on an older, converted building so climate control is a bit of a delicate balance at the best of times; I’m hoping that the extra complexity with the water loop pays off by avoiding adding a big thermal gradient across the space.
Edit ping notifications do work! I don’t mind the double notification though. This part is far beyond my expertise, but it has been fun to read and learn.
You might be able replace the outdoor setup with what is essentially a larger cooling block designed for hydronic heating/cooling systems. Something like this
That's a good shout, actually - similar to the "whole room watercooling" setup but a good bit more robust! My main concern there would be the outdoor loop getting too cold when the temperature drops quickly; if it's consistently cold outside I'll have the loop running inside for heating anyway, but if there's a quick shift from warm humid day to rainstorm to cold clear night it seems like it could pull the interior loop down below the dew point.
I'd use a DC hot water circulating pump, where you can more easily adjust the flow rate. As well as adjusting the active fan speed on the outdoor chiller as well.
If you speed/slow the flow rate based on diff between indoor and outdoor temp that should resolve. Pump the water faster, but outdoor fan off so it spends less time outdoors when it is colder.
I'm also presuming that this whole setup gets shutdown at first frost.
Got you, thanks! Will keep all of that in mind when I’m figuring out pump setup.
I know you pitched this as inherently not the right idea, but I do kind of like it! The integrated thermostat and secondary loop solve a lot of potential problems with pumping the actual computer coolant outside, it's comparatively cheap, the condenser is already designed to work outside... It has enough potential that I'm going to do some more research in this direction at the very least - I appreciate the suggestion, even if you regret me taking it seriously now!
Gonna +1 to this recommendation. This is going to be your most efficient solution. The only comparable option that's cheaper would be a window AC unit, but those have their own drawbacks
Actively cooling PC components is a bad idea because if it's the coldest thing in the room it'll from condensation. Just like a glass of icewater.
(@Greg to insure visibility)
I think the thermostat can take care of that concern. If you're setting it to something like 80 degrees F I don't think you'd have that problem except potentially if you shut the PC down before you shut down the AC unit
The AC coils cool much lower than ambient temp when it kicks on though. Like easily 20F lower than ambient, even if the thermostat was wired hot.
The thermostat basically just determines how long it runs at that temp.
True, but wouldn't the fact that he'd be doing heat exchange between coolant and water via an external heat exchanger mitigate most of that and smooth out the temperature by the time it got to the water block?
I'm thinking especially if the thermostat is measuring the temperature of the water, not the coolant, and the heat exchanger has a decent amount of thermal mass, those sharp valleys in temperate drop wouldn't be an issue.
Too hard to say from remote unseen adhoc engineering, but my gut says "maybe."
I'm still unconvinced this would be more efficient than just installing a vent in the ceiling. Less power draw, added bonus draws up cold air from below.
That said I do like his pool heat pump idea.
Cheers for this - definitely had condensation up there in my mind as a concern, but there's every chance I would have missed some nuance around the cooling setup and ended up causing it anyway. Now the pool heater plan is coming along it looks like a lot of them have integrated bluetooth, so with any luck I can get that talking to Home Assistant and have it automatically track the interior temperature.
Yeah I think a split AC system is the only thing that's really going to appreciably lower OP's room temperature and use that increase in electricity cost in the most effective way (as apposed to getting a window AC unit or something like that to cool the existing air). Then, if OP's building allows it (although they mentioned it's a first floor apartment), they could offset some of their energy costs with balcony solar. I'm not really sure that non-HVAC people should be messing with a configuration as potentially risky as all this, but it's always sounded like a fun project to me.
I'm also concerned that any solution that's not what papasquat suggested is just moving inside air to the outside and outside air to the inside of the apartment. Any cool air that's created through a closed-loop HVAC system and exhausted to the outside is turning the entire dwelling into an open-loop system. The air from the apartment that gets warmed in a computer and vented outside has to be replaced with air from somewhere else. Generally that's going to be through the gaps around doors, door frames, windows, etc. You will find out real fast how well sealed and square your dwelling is when trying to vent a lot of heat. The split AC system essentially keeps outside energy outside all the time, which should keep it might nicer inside.
I had completely failed to think about the necessary inflow of air if I were venting hot air out of the window with a reasonably well sealed duct, that's a really good point. Probably still worth it on balance assuming the computer heated air were hotter than the outside air, but it's a strong point in favour of something a bit more detailed - thanks for pointing it out!
If it helps at all, while looking how you could do something like up I did find these refrigerant to liquid heat exchanger that you could, in theory, get fittings for and replace your radiator in a liquid cooling setup with, and then likely replace the evaporator coils in the air handler with (you'd still need the electronics from the air handler, because thats where the thermostat and control circuitry live. Also, I'm not an HVAC guy so I may be completely talking out of my ass). I wish you the best of luck!
Ohh, I kinda like that. Cut out the middle man, makes the AC not have to work nearly as hard.
re: #1-3, I asked something similar here.
Probably a dumb question, but have you already undervolted all of the GPUs to shave off the stock "50-100W for 5% performance"?
Thanks for that, I definitely wouldn't have found the conversation in that giant thread otherwise! Seems like the consensus there leans towards the hot air out the window idea, which is certainly the lowest overhead option too, so I may end up trying that as a first and easiest path - although I'm definitely still wondering if there's a more robust answer we haven't found yet.
Totally reasonable question about undervolting, but it comes back to the scale issue: saving a few hundred watts here is good, but it leaves a couple of kW still to deal with one way or another. It's a lot to be running at home, to the point I feel a bit silly for having this kind of setup at all, but the wonderful incentives of modern capitalism mean this is by a pretty wide margin my best option right now...
What type of dev work requires multiple GPUs?
While I'd imagine the answer is a resounding no, would doing some of this work in cloud compute through AWS or similar be cheaper than potential home modifications?
Machine learning research work - I do also run some stuff in the cloud, but I've been expanding the amount of things I'm running locally in the last few months specifically because of how much it was costing me.
If you compare what a datacenter-grade GPU would cost to own vs rent in the cloud, and you're running it consistently 24/7, it's favourable to own within a couple of years (makes sense, cloud providers need to make their margins too, and a lot of times the ability to quickly scale capacity up and down is worth that premium). If you can get away with running your dev work on consumer GPUs rather than datacenter ones, which are often 10x cheaper for effectively identical silicon, that payback period gets much, much quicker and suddenly breaking out the power drill looks like a very attractive option!
Yeah, I figured as much. Just wanted to double check.
I'm not running that much hardware, but I have two computers close to my body - my main PC with a decent GPU that puts off heat, and a work laptop I've mounted under my desk that runs hot.
My solution has been running a fan under my desk to help move cooler air over my feet/legs, and push away more of the computer heat. It's a Ryobi one that's powered by a power tool battery, so it's pretty compact. That's made a significant difference vs running a fan elsewhere in the room along with A/C. It's worth trying, as they're only ~$20.
Well, if it's workloads that don't need to be done locally, the option isn't just local or cloud. You can just rent some colo space near you and run your own rackmount server. Some of them will rent space by the rack unit for as low as 30ish bucks a month. You can stuff 4 GPUs into a 2u chassis.
Ultimately you're doing data center work in an office, so it's probably always going to be hot or loud or both unless you spend a lot of money and time making it not be. Why not just run it in a datacenter? If you're making money off of doing this stuff, it makes sense to invest a little of it to not be miserable and not eat up a ton of your time trying to solve this problem.
The very honest answer here is that this is hardware that's jankily been built, rebuilt, tinkered with, and upgraded over the course of a couple of years and I assumed any colo provider would laugh me out of the building before considering letting it anywhere near a rack with other people's machines in it.
In reality I probably could neaten things up, get a proper rackmount case, some high flow fans, a couple of PCIe risers, decent air cooled heatsinks (I'm assuming even in the best case they don't want my DIY watercooling loop threatening to leak on things, however much confidence I have in it personally), etc. and it may actually be the right call in the end. It's absolutely an option that's still on the cards, although it'd probably cost more just to set up than your less-serious split AC idea below!
Do you happen to know what kind of rules and restrictions there generally are on what you can put in a colo rack? I've only ever worked with nebulous cloud VMs or hands on local hardware, so I'm not at all familiar with the process for putting my own hardware in someone else's datacenter.
They generally don't care unless it's a hazard. Water cooling and obvious fire hazards would be a no go. The racks have a certain amperage they're rated for also, so you can't exceed that.
You'd have to call around because not all colos rent space by the U. Many of them will only rent space by the rack or even in groups of racks, mainly because it's basically impossible to physically secure and segregate space a U at a time, but some do it.
You'll mainly want to find out what cross connects are available for network providers, or if you can get internet as part of the colo space. The latter would be more ideal because then you don't have to mess around with ISPs. Other considerations you might want to ask about is remote hands availability and how physical access to the colo works.
Good to know and very much appreciated - now that there looks to be a viable outdoor water chiller option on the cards that doesn't involve me hacking up an AC condenser I'm strongly leaning towards that, it looks like it'll work out best of all worlds if I haven't missed anything major, but I'm filing this away as plan B if that does fall apart for any reason.
Should you decide to go the colo route, it looks like you’re in the UK… I can heartily recommend Jump (https://www.jump.net.uk) for London based colo. They are small and aimed at enthusiasts and smaller businesses and are very very competent and easy to deal with, as well as renting by the U and being open to “weird stuff”. I used them happily for years.
Simple, no-marketing-BS website that has all the info I need and still has references to Ubuntu 11 in the FAQ? Yup, these are clearly my people - thanks for suggesting them!
Looks like it'd definitely add a solid few thousand a year to my costs, though; no criticism to them for that, everything there suggests to me that their pricing will be very fair, but I guess that's the nature of doing things properly rather than hacking them together like I'm doing right now.
For the moment, being the one person operation I am, this is encouragement to get the DIY version working for a one-off few hundred quid and hopefully keep that extra ongoing cash in my pocket. In a year or two if it turns out I'm scaling significantly beyond that (fingers crossed!), it's really helpful to know where to go and what to expect when I'm past the hands on hackery stage.
I'm struggling to picture the exact setup, so to clarify: all the gpus are linked to a single cooling system? In that case you could try using one of those window seals that have a flexible AC duct link, and vent the air outside. You maybe want a fan in the duct at some point, unless the cooling system one is enough.
That would be the cheapest, least work intensive option, I think.
Yup, multiple GPUs running into one giant radiator - I'm actually in the process of adding capacity and reworking things a little (some new hardware, some waterblocks still in shipping, that kind of thing), which is why I asked the question now, but keeping all of the heat exhaust localised in one place as far as possible is definitely part of the goal. I may end up with a second radiator, but having two side by side shouldn't dramatically change the answers, I don't think.
To a reasonable degree I'm alright with a bit of cost and effort if there's a nice option here that wraps everything up neatly - these are the tools of my trade and I don't mind spending where it's worthwhile - but so far most of what I'm seeing in my searches is industrial equipment that's orders of magnitude above my level in capacity and in price, so perhaps keeping it simple and just getting some nicely fitted air ducting will end up being the right answer after all!
This sounds like the perfect use case for Whole Room Watercooling! Get some longer hoses and put those radiators outside. Probably want to go for the second radiator since the temperature difference will be lower outside, so more area will help. And definitely want to make sure it is shaded. But still should be worthwhile.
Why am I thinking of a nuclear power plant style cooling tower outdoors now? Not that you suggested it, but it might just work: Whatever water your heatsinks dump their heat into, you pump that stuff into a cooling tower. In there, you simply spray the water around, collect it back up, and pump it back inside. I hope your neighbors like their balconies humid...
I am, by the way, imagining the jankiest possible solution: Big plastic bucket for the tower structure, some plastic pan as a sump, aquarium pumps everywhere, and a shower head to produce the droplets.
Given Greg just took a very janky idea a bit too seriously, I'm afraid to submit this thought...
I like that idea, and I hate that I'm going to say this, but I think the cheapest option is simply to run water into the liquid cooling system and then ... dump the hot water down the drain. Water is incredibly cheap and most nuclear power plants don't waste much time or expense on trying to recapture all the steam from the turbines, they simply dump it back out into their local water source (which is why so many are built on lakes or along rivers or are ocean-adjacent). If OP has to dump a bunch of BTUs somewhere, dumping that down the drain and utilizing the already fairly good thermal capacity of cool water is just a cheap one-size-fits-all solution.
That thought had actually crossed my mind, although I don't think I could ever actually bring myself to do it (and I know you weren't seriously suggesting it)! It's probably a little silly of me to worry about given that my local water supplier loses hundreds of millions of litres from poorly maintained infrastructure while giving huge piles of public money to its investors instead, but hey, we draw our lines where we can.
Yeah, it really depends. If you're in France and your electricity is mostly nuclear or solar/hydro, then you're already using a lot of water to create that electricity, so maybe it's more efficient just to use the supply that's created at a national/regional level. If you're in a coal-burning area, then you're using a lot of everything else (not sure how much water coal plants actually use), so using water instead of electricity is the most environmentally-friendly option ... even if we're all running out of fresh water on a human timescale.
I guess if you wanted to get really out-there you could capture grey-water from washing your hands in the sink and use that to fill your open-loop liquid cooling system. Then treat the water with a drip-feed of chlorine to keep the insides of the tubes clean.
Argh, so many layers of consideration! You're doing a very good job of spotting the nuances I hadn't considered here... I'm in the UK nowadays, so looks to be primarily wind and natural gas - my gut tells me that with heat pump levels of efficiency it'd be better to just shift the heat outside that way, but I have nothing to back that up. Honestly, regardless of actual solutions I've kind of been nerd sniped here and I feel the need to figure out a back of the envelope estimate at some point now.
5 Euros or so per day just for the water to cool the setup, and that's when we're willing to discharge at inletTemp+40°C. That's not nothing. I'm not liking those rates, but yes, it is kind of remarkable how cheap that actually is, considering the wasteful use of resources.
Yikes! I didn't know water was that expensive in the EU! And thanks for doing the math!
Where I am in the US the cost of water is around 1/2 of that of some (median?) EU cities (although OP is in the UK). I just wanted to see this in numbers: At 3.5 Euros/m3 of water and converting it to a nominal 1,000 cubic feet of water (a baseline amount in the US), that would be about EUR100 or $108. I'd pay around $50 if I used that much water and I don't use half that much water in a 3-month period.
Apparently even water in San Francisco costs less than it does in Denmark. Yeesh!
I was going to post this as a top-level musing, but it's not top-level worthy, so I'm saddling you with this response for no particular reason. Congrats xD
I've always wanted to do a mineral oil-submerged PC build, but repairing/replacing components is a hassle. It got me thinking about different ways to better utilize liquid cooling setups or heatsinks in general. For instance, why use a copper heatsink-to-air/liquid when you could run pipes directly into your home/apartment foundation? Geothermal on the cheap! (And maybe some building code violations, too! So fun!) I don't know what the thermal capacity of the average concrete foundation is, but I'm sure it's enough for a few hardware geeks to exploit here and there.
Another idea, and one that's more friendly to those of us wanting to recapture heat in the winter, is to expand the liquid cooling reservoir itself (in OP's case between the system and the minisplit AC) and to use a silicone oil or something as the coolant to maximize efficiency. Most consumer liquid cooling systems that I see don't even have a spare reservoir (which makes topping up closed-loop systems that much harder). So running part of the cooling loop into a large fish tank would give additional thermal capacity to the system, but also allow the operator to take off the top of the fish tank and exhaust some of that heat during colder temps -- including optionally at night. I think it would be a fun project to have exhaust fans on a timer to open up either into the home or out of the home at night. Refilling the reservoir would also be easy and maintenance on the PC hardware wouldn't be as troublesome as it would be with a submerged mineral oil system.
Funnily enough, I did see this video in my feed the other day...
I like this thinking! You wouldn't be worried about corrosion and whatever other outdoor hazards could end up befalling the radiators running it that way?
Full disclosure, it wasn’t my idea. Whole Room Watercooling was from LTT trying to do exactly this idea. It ended up going poorly, but I don’t think you would have the same issues. They were trying to cool 6 editor computers in a single loop, and allow the computers to be quick disconnected and run independently. Their issue was some obstruction in the blocks on a few computers. It was some sort of algae growth, if I remember correctly. They didn’t have any issues with the radiator being outside, although they did build something to protect from the sun. They also used water resistant fans.
Since you are doing this on a single system, I don’t see it being an issue. Functionally it’s just a large watercolor loop. If you have to go a long distance to get it outside, you probably want to spec up your pump a bit. Also, if you have to run it vertically, it will be difficult to prime. In a closed and primed loop, head pressure of the pump doesn’t matter (in theory). But before priming it does matter. However if your workstation is on an outside wall already, it should be simple. I would also recommend going with waterproof fans. I can recommend Noctua’s iPPC line. You can even get the high rpm ones and run them fast to get lower temperatures since noise won’t be an issue.
Besides the waterproof fans, I don’t see any issues with having radiators outside, as long as they are protected from direct sun and rain.
To be honest, I am a little jealous. It sounds like a very fun project.
Interesting, I hadn't come across that video - I'll give it a look and see what I can pick up from their attempt. If it could be as easy as sticking some water resistant fans on an off the shelf radiator and building it a little birdhouse type enclosure that sounds totally doable.
I'm still a bit concerned about the potential for condensation on the GPUs if the outdoor part of this hypothetical loop gets too cold overnight, but maybe doing it as a double loop with a heat exchanger, but using an extra set of straightforward PC radiators rather than hacking up air conditioners could solve that. Either way, I am absolutely expecting a fun project out of this and it's a big part of why I asked here! Very much hoped I'd get some good discussion like this.
It was back when they were in the Langley house, so very old. If you search YouTube for « whole room watercooling » it’ll give you a playlist.
I hadn’t thought about the condensation, but that is definitely a possible issue. One possible solution would be to insulate the computer like what you need to for sub-zero cooling, but that is a ton of work. Going with a heat exchanger might be a good idea. Or just make it movable and bring it inside in the winter. Also, if you have an attached garage, you could put the radiator in there, as long as you don’t mind the garage getting a bit warmer than the house. Or, if you have an attic, you could put the radiator there.
Whatever you choose, let me know how it turns out. I am curious how well it will work. It’s one of those things that should work great on paper, but to never know until you try it.
Oh wow, the playlist is from a decade ago, that's a great throwback! I kind of love that their super giant mega radiator array for six top end editing PCs is a bit smaller than one of the (admittedly absurd) commercially available radiators you can get now.
Just watched the first couple of videos there and it's also reassuring me that even the jankiest ideas we've been floating here are a good bit more robust than what they seem to be putting together! Which is almost definitely because we've now got the luxury of all these extra commercially available bits and pieces (plus decent 3D printers, the jank levels in my setup would be so much higher if even an amateur like me couldn't just model a perfectly fitted custom part in 30 minutes when I need one), but it still made me smile.
I'll finish watching those tomorrow, it'll be interesting to see how they get it all finished up with all the DIYing they're needing to do, it'll be well worth seeing what pitfalls they hit so I can avoid them, and I'm enjoying the earlier-days-of-YouTube vibe too - thanks for suggesting it!
Front runner for my actual build now looks to be a swimming pool heat pump, so that'll hopefully tie in with any PC-related tips I can get from those videos and end up being a fairly neat solution.
Perhaps a few more details are in order about your home: Anything to do with heat/cold is potentially relevant. How is your hot water prepared? Is there a substantial heat store there, e.g. a hot water tank? How your home is heated is, I believe, mostly irrelevant for the question of how to cool it down, but might not hurt for completeness' sake. How is your home cooled? And most importantly (IMO): Is it a necessity to sit your sweaty ass right next to your multi-GPU workstation, or could you put it in a server rack in the utilities room, interacting either via remote-ssh or via periphery cables run through the walls? Also, just how many watts of compute are we talking about here? Also also, what's your home like? Single family house? Any utility spaces that you could use and/or pollute with the extra heat?
Counter-questions aside, here's the high-effort, high-reward solution. I don't think most of this exists as off-the-shelf components, but who knows.
Put your machine next to wherever your hot water is made. Suppose you have a storage tank for hot water next to the rig. Whenever you run your rig, cold water from the utilities is brought in, cools your machine (should only really be at a trickle; water can soak up a lot of heat.) and is thereafter stored in a tank. That tank is either your primary hot water storage tank, or is used as the primary source of fresh water for your actual hot water storage tank. Yaaay, your hot water in the summers is now completely free. Did you want to take a hot bath?
Admittedly, this solution makes the eco-nerd in me very happy, but I don't think it's very plausible. One, depending on your compute needs, you might struggle to use up as much heat during the day in the summer. Running a few GPUs hard might get you into the 1-2 kW range, at which point we're talking about the electric power of a kettle that can boil a liter of water in a few minutes. Kinda hard to use up that much hot water during a hot day. If you end up going for something like this, I'd say you definitely need to talk to a plumber.
Perhaps a bit more realistic is to find a more passive heatsink. Fewer moving parts and such. Maybe a concrete floor in the garage or something... But I can't quite put that together without a bit too extensive renovations. Give your garage underfloor heating, and hope that it soaks into the ground mostly? Seems a bit too high-effort for the problem.
I guess if I want you to take anything away from my ramblings here, is that there might be space for a good solution if you move your machine away from your desk, but keep it under your roof. Even if it means that you let it heat your garage up to slightly-uncomfortable temps.
Very good questions, many of which envisage an awful lot more space than my tiny first floor flat in the middle of a dense city provides! I'll get back to you about those garage-based ideas if a lottery win comes my way, but I do have the luxury of a large balcony which is where I figure the bulk of the heat probably needs to end up one way or another in the summer.
No hot water tank, unfortunately, although I like where your thinking was going there - water and central heating is all on-demand with a combi boiler. Cooling is a fairly cheap and fairly old mini-split AC unit that'd probably need upgrading if I want it to fight with the computer heat all summer.
I'm running somewhere over 2kW all in, so it's more than enough to heat the whole space even if I'm not sitting next to the hardware - right now it's 7°C outside and I'm sitting comfortably warm with a window open! I do like the water heating idea in concept though, and ultimately there does need to be some kind of big heatsink even if that just ends up being the air outside, so I'll do some more investigation as if I were dumping the heat into a hot water tank and maybe that'll set me off in the right direction. I'm already thinking maybe a heat exchanger from an indoor loop to an outdoor loop where the outdoor loop only moves if it's above indoor ambient could be interesting...
Yeeeeah, ok. I was picturing a bigger home there, so most of my ideas are probably scrap. Most of those simply aren't an option.
I've done some math. With 2kW, you'd actually need a fairly decent tank to store it all. If you're just working with, say ~40°C deltaT, then you can expect a daily cubic meter of hot water. You'll get three times more mileage if you're also using the melting point of water (that alone is a cool 300kJ/l), but at that point the complexity of the system explodes. And with the amount of energy you'd need to store in a year, building a seasonal heat battery is out of the question, unless you want to convert the entire flat next door too.
(Edit: As a point of reference, Germany subsidizes low-energy homes. One class of low-energy housing, called KfW40, requires at max 15kWh/a/m^2. At 100m^2 of space, that's 170Watt. To heat a home. That's it. You're dealing with 10x as much heat, no wonder you're struggling to get rid of it.
Yeah, out the balcony it is I believe.
BTW, what you're doing right now, perpetually open window in the winter? I've heard that sarcastically called "heating for the air force".
Hahaha that definitely puts it in perspective.
@Greg you may need to look out that they won't raid your home on suspicion of running a grow op. Hot, energy sucking houses are prime targets for an impromptu quality assurance check on your front door. I'm only halfway joking.
It's a thought that's very much crossed my mind! I don't know that there's much I can preemptively do to avoid drawing that kind of attention beyond maybe a nice reassuring sign on the door saying "Lots of computers, definitely not weed, pinkie promise", but at the very least I've got offsite backups of all my data and a pretty iron clad paper trail of where all the power is going.
I wonder if literally popping town to the local police station and asking them might at least put you on their radar so they maybe double check before sending a raid?
(This assumes the police operates for the benefit of society, which I would argue is a bit of a stretch or complete fabrication in many places)
I could imagine even starting the conversation with “hey I’m concerned that from the outside, my home data centre might look like a weed growing operation, could you please send someone around to have a chat and explore my options so that we avoid an expensive misunderstanding?”
It's a solid suggestion - the police where I am aren't the best, but they're far from the worst and it might well be worthwhile. I'll give it some thought and maybe see if any friends locally have advice either way...
Meanwhile as I dive further into planning this cooling system, it seems like half the searches for detailed info are turning up reddit results for either hydroponic circulation pumps or questions about cooling high power LED arrays for self-described "agricultural reasons", so perhaps I really should get ahead of the issue before I start buying things!
Oof, I knew I had a pretty ridiculous setup but that really brings home the scale here. Sure, I use a renewable electricity provider, but there's some real discomfort in this being the most viable way for me to make a living that's a whole other topic of its own...
This, however, definitely made me chuckle!
To be fair, you have customers willing to buy it. So whatever it is you're doing, there seems to be enough demand for it to justify what you're doing. I'm in the same general line of work, just slightly different methods. Right now the AI space is a bit overrun with so many ventures and we haven't really figured out which ones deliver real value long-term, but there certainly is enough signal there in the noise, of projects that bring real benefit to real people. In the healthcare space, those benefits can be worth 10s of thousands of dollars per patient, so it's not like we're turning electricity into heat for shits and giggles. Even those of us not working in domains that so readily show their net benefit to society.
And to be fair, those KfW40 houses are quite efficient. Almost a solar array and a ventilation system away from passive houses. Buut, doing the math helped me square away how a seasonal heat store for you seems so unreasonably large, when it's doable for some low-energy homes. If you cut the yearly energy store by 10, triple the energy density because you're running your thermal battery from -1°C to 40°C, then you'd end up needing ~12 cubic meters of water to heat a KfW40 home through one winter, and in the summer that's your heatsink for AC. 12 cubic meters is still a lot, but it starts to get reasonable.
I appreciate the reassurance, and discomfort aside I do appreciate the sense of scale too, it's interesting to visualise! I'd certainly like to think there's some potential for real benefit from the work I'm doing - one of the potential applications is making prediction models for things like weather and climate work a lot more compute efficient, ironically enough - I just wish I could spend more time on those meaningful applications and less on the adjacent finance-related ones that pay the majority of the bills.
If you're cool with drilling holes and you're on the top floor I'd consider adding a bathroom vent fan on the ceiling, if you're feeling extra fancy pair it with an intake fan near your HVAC to avoid negative pressure. Since you wouldn't have moisture issues could just vent direct into attic. Pair it with a nice insulated cover for when you don't want the heat to escape. Pay the premium and get a silent model.
Do you have another room you can stick the development PC in and make it a headless machine you remote into? I do that with my work PC and there's no noticeable lag over Ethernet.
The workflow is already all SSH, so that’d be no problem, but unfortunately my living space is small and the power use is large! Vektor did some calculations above and it’s more than enough to get an entire home to uncomfortable temperatures, so some way of getting the heat out of the building entirely is definitely needed.
I run a heavy GPU load as well, for the same reasons. At max load, it's almost 3kW across two circuits.
In the winter, I have the machines inside, because hey, free heat.
In the summer they get excommunicated to the garage. Simple, and it keeps my cooling load reasonable. Can you just move yours to an outbuilding, garage, or different room in the house?
I’m in a flat in a densely populated city, so I’m doing this on hard mode! Sticking it out in a garage would be ideal, but all I’ve got is a balcony, so it’s going to be a matter of shifting the heat out there without leaving the GPUs themselves outside exposed to the elements and/or any passerby with a long enough ladder.
That is hard mode! Your best bet is probably simply exhausting to the outside, then. No need to get too fancy, and since you're in a flat, you don't have to worry as much about energy loss from the air being pulled in to replace the vented heat.