Useful processing power often isn't in a parallel form. If a task is highly parallelisable, you probably haven't optimised it very well... Still, high operation per joule figures get me kinda hyped ngl
high processing power per watt, but power is 20c/kwhr not 200/kwhr therefore a single second hand laptop is better. Its mainly autism + OCD but each to their own
@@MichaelKlements I don't really understand the purpose of this, because buying a micro motherboard with a modern CPU would have been surely way better use of money and time...
For anyone who knows about water cooling, having it in series is the only way to go. That water flow is usually fast enough to warm up evenly so there is no noticeable difference between each node. Usually what you have is the loop thermal balance. jayztwocents did a good video about it.
Yup - you maintain the temps of the loop as a whole. That's all you do with watercooling. Flow rate is the main concern here to be honest due to the amount of blocks that restrict flow by design.
2 components are a lot more different than 8 components in a loop with pressure drops over every connection. it will matter and the later ones will be hotter. maybe not enough to thermal throttle, but they sure as hell wont run as cool as the first ones
i suspect the blocks are simple in design and lend themself to slow flow rates rather than the high turbulence/high flow of a higher end pc type block with all their micro grooves etc. over and over real world tests in pc world have shown loop order makes no difference in temps. ie sending GPU hot water to CPU doesnt make the CPE hotter because everything equalises. having 8 equal boards rather than CPU + GPU maybe the 8th board will run hotter for a while but not over the long term and i think this is a system to be run 24/7 long term personally id run a parallel water system on what im guessing is a slow flow rate just to be 100% sure everything heats at the same rate. even if the last board is hotter for a bit longer eventually i will shorten its life especially if this isnt a 24/7 system
If anybody is wondering about these two comments above me, yes, they are bots advertising a website that is specifically designed to take your money and info. I’m starting to see these bots more and more so be careful.
Clustering Learning Try out different technologies (SAAS/SARS or container swarms for example). As a test environment to not have to run tests in production..... And much more ;)
I guess these "build a RPi cluster" videos have the same reason... find somebody who will figure it out what is it good for.. or atleast i hope somebody will one day :D
First of all, that´s an amazing job thank you so much for sharing with us... but i´ve seen in the video few air bubbles flowing into the circuit and it may be a problem because you lose eficiency in the heat transfer and air also affect to the pump, making it to cavitate. A posible solution, make sure that you don´t feed the pump with any air bubble just by setting it vertical, or rotating 180º the radiator to keep the air trapped in the upper part of the radiator. Thank you for the time you spend making this video, and greetings from Spain. Keep it up!
Hi Jose, Yes, there was still quite a lot of air in the system during the video as it was still only the first couple of minutes of the system running, so the system was still working the air out of all of the cooling blocks and radiator and that. Yeah, ideally the reservoirs should be vertical as well, so that the air pocket is as far away from the pump as possible. Thanks for the tips and the support!
Looks brilliant and love the black and red. Only comment with the actual build is I would have rotated the Pi's on the right by 180° so the coolant was on the right and mirror the coolant on the left.
@@r3d_panda Good use case. Could you share your cooling solution on your raspberry pi 4 ? I see many discussions about raspberry pi 4 could go really high temperature like 80 deg C and get throttling. Wonder how your case running so good.
@@AndrewTCW Honestly i don't have any long term data yet but i was barely able to let my pi reach 50 °C during stresstesting. I'm currently using the argon one v2 case. The complete upper part of the case works pretty much as a giant heatsink and the only time i ever saw the fan kicking in was the test mentioned above (there is software for this case available to setup a very basic fan curve). Keep in mind i don't have a lot of permanent workload tho. Most of the time my pi idles with a workload of 0.3 (htop) and temps around 42 °C (pihole) with my current setup (13 Containers running in docker on a headless host with raspberry pi os lite installed). So i guess this highly depends on your use case and if you are willing to use at least some sort of cooling solution (passive is probably enough).
They're a relatively cheap way to learn and experiment with clusters and distributed computing. They're not competitive in performance, but when you want to learn how to configure scaling systems with load balancing and redundancy etc. it's cheaper than building or renting a cluster with actual PC hardware. Plus it's fun.
Wow Maker Michael!! This is totally neat. I was wondering about surface mounting the radiator but then saw you had cut a hole through. I would love to know how the thermal performance the individual nodes. Look forward to seeing what you are planning to do with this neat cluster. Keep up the good work and congrats on the 10K subscriber mark!!
Thank you Deech! I'll probably do a follow up video on the cooling performance at some stage as well. I first need to figure out how best to log and display it. I did do a basic test, running all of them at full load for a minute or so and there was around a 3 degree difference between the first and last Pi in the loop and the system was only about 6 degrees above the ambient temperature. But this will probably change for longer runs and once overclocked. Thanks for the great feedback and support!
Nice work! Really love how it looks. Please do cover all the software that you will be using on this in a video! Looking forward to watch it! Take care and be safe!
@@TotesCray The bigger issues is using pi's for mining eth is like asking a child to pick up a hot tub. Any way you look at it, it's just not practical.
A custom 15x40x40mm copper fin heatsink with a decent TIM adhesive will sufficiently cool a 2.147 GHz OC'd RbPI. If boards are under high utilization, a 120mm fan can remove excess heat from 5 boards. This together makes a high density, modular platform
One mistake is that pump and reservoir is on bottom, should be on top because you have residual air in the loop and it will be trapped in the top PI cpu block. Puting the reservoir on top the air will come back to it it would be no problem. Other then that i dont see anything else wrong here, everything is beautiful, great work.
I dont think there is much fins in the block at all, so the bubbles would flush out 1 sec after the pump is on, also remember we are talking about 10-15 watts of heat total here. Air will normally not work its way to the top past the water in the tubes, in pc water cooling the top radiator or a cooling block will not fill with air even if it is over the water tank air level, only if the pump is sipping air and pushing it through the loop. When the pump is at the lowest spot, the pump and loop will work until the entire loop is empty. If the pump is higher, the pump will pump air before the system is empty. But it really does not matter, i would put cable ties on all the hose connections though.
I fully expected that to end with,"and next week, we'll ask it how it feels, we connect it to the internet and I accidentally create skynet." Seriously engaging video, you built that amazing machine with the care of a craftsman. Thank you for filming it, as well as building it.
Yeah this is correct. Same as with a water loop in a PC, the loop with eventually reach its equilibrium and no individual component will get hotter water into its block than any other. However it should also be noted that should one component be running at max and generating a lot of heat then all other components will receive the heated water from the loop.
This is just simply incorrect. what youre stating is that no heat would transfer to the water from progressive heat sinks. youre using the words "saturated" to mean its at the same temperature as the water which would mean no heat transfer takes place and is therefore useless.
Good project. If you are concerned about water temperature, might I suggest you run the radiator and fan from the rear of the display and insert some thermal barrier to isolate the heat from the cold. Likely you would not require much but can likely cool the water by a degree or two under load. Also adding longer tubes would allow the water to cool more before it even reaches the radiator. I would have gone with a circular arrangement myself by that is just personal taste.
@@MichaelKlements Yes please, I would love to learn more about the Rasberry Pi 4 GPU power and if it would even be a viable in profiting relatively quickly compared to previous versions that would only need 1000 years to make $5 through CPU mining Monero.
Since its one Thermodynamic System, the temperature difference in the cooling Water will even out over time. Giving enough runtime every Pi will run at the same Temp. Sorry, 4 the spelling I'm German.
Looks really nice! I think turning the pi's on the left 180° would have looked a tiny bit better though - the network cables and tubing would be symmetrical then. But I guess that was not possible for space problems?
It is possible to do, the blocks aren’t in the centre of the Pis though, so then they’re slightly offset on one side. It probably would have looked a bit more symmetrical your way.
awesome build! though you might wanna reconsider the zip tie on the leaking inlet. zip ties in water cooling, if used, are usually used to hold the tube in place, so it doesnt slip off of the fitting, not for ensuring a tight seal.
Так он именно как кластер собрал, наверное как лабораторный стенд для распределённых вычислений. А такие заморочки с водным охлаждением чисто для прикола делаются, есть видео где люди просто собирают кластеры из малины без водяного охлаждения и все у них ок.
Yeah a lot of people worry about it in regards of cooling CPU's and GPU's and at your first bootup, or while enterring a game, it seems like there is a difference, but when the water reach that temperature equilibrium, the orientation doesn't really matter anymore. Let it run untill it no longer increases in temperature, then run the tests. (Which is why techtubers HATE testing water cooling gear)
@Michael Klements I'd advise you to think about putting the reservoir vertically next time, because if you mount it horizontally, it might run dry so the pump might die
Awesome build! One thing I'll say, in case no one else has, the series run of each Pi in the water loop won't really end up increasing temps for the ones down the line as much as you might think. The thermal capacity of water and a 120mm radiator is going to be far higher than even 8 pis will overcome. Certainly the water temperature will increase as you add heat to the loop as a whole, but there shouldn't be appreciable difference between series vs parallel. As more heat is added to the loop as a whole, the water temp will increase, but the ordering doesn't really matter, in practice. th-cam.com/video/RnPB_q51iVk/w-d-xo.html for more info and testing. That video is using much higher wattage parts, as well, the 1080Ti consumes (and thus, converts to heat) 200+ watts, and the 6900K CPU is rated for 140 watts (before the increased voltage and 600MHz overclock he'd put on it.) Even if you're overclocking the Pi 4, it's still going to be dissipating less than 10 watts under load, so the total for the 8 Pi cluster will be well below 100W total. Your 120mm radiator is not going to have any problem at all with that heat load. Thanks again for the videos! You've made me want a laser cutter even more than I did before!
Thanks for the detailed explanation John. This was pretty much my thinking when putting it together and it seems like there are a few videos out there which have tested different configurations with similar results. A couple of people have mentioned watching JayzTwoCents's video. I must admit that when I bought the laser cutter I didn't think I'd use it nearly as much as I've landed up using it! Thanks again for the input and support!
@@MichaelKlements I just left that exact comment and was about to mention JayzTwoCents etc. (on a side not LinusTechTips did one on radiator stacking, unrelated but useful to know if you ever wanted multiple rad's) Yes your build is perfect, Good Job. I'm looking forward to a future where liking these together with PCIe is a thing (like what's happening in the datacentre's) then the true scalable ARM supercomputer will be something anyone can have (Pipe dreams i know)
I was gonna say that too, eventually your loop will reach equillibrium. I'd imagine none of the water in that particular loop is going to be much more than ambient. It really wouldn't have mattered either way you set it up, so long as everything was getting adequate flow.
You are right. In addition; the water has so much flow (even on the lowest pump rpm) that the delta temp between each waterblock is zero, microscopic at worst.
That's just not true. The temperature of the coolant out of the heat exchanger will be lower than on the return. As OP mentioned - the first Pi will experience colder coolant, then warmer, and warmer etc as you go down the chain. Once back to the H/E, the coolant will be cooled again and the process repeated.
@@robertshort9487 too many variables in that test and also not all the same components. The OG of the 8x Pi's video needs do a temp comparison of each of the Pi's CPU's and the coolant temp at the in/out of each block and the reservoir to make this a valid test and argument.
@@robertshort9487 also that link you sent, he is making a lot of assumptions. "If you were to...." - "you would probably...." etc etc. No actual proof - just his assumption. Fact is, if the radiator/Heat Exchanger is efficient, then the placement of the blocks in series vs parallel will matter. First proc is getting 20C coolant, 8th could be getting as much as 40C coolant. Core temp of the first will be much cooler than the 4th. It just will be.
I think most people (myself included) build them to learn about cluster computing, they're a great learning tool. Long-term I'll hopefully be able to get it running some simulation software packages.
@@TmanT321 You can see he uses 2 splitters on the back side (in black). I found that curious because they didn't make the parts list. Normally I would overlook that, too, except he is EXCEPTIONALLY meticulous about naming all the parts down to the screws and even the machines he used to create them. Also, the budgeted wattage seems a bit low for that many pi's. I'm sure they will power up, but I wonder how they will fare under load.
I remember 10 years ago there was this trend on TH-cam where everyone was buying inflatable toy boats to mount floor, electric engine and expensive furniture on it and they sailed with entire familys while filming TH-cam videos. These were absolutely overkill mods for toy boats, and extremely dangerous rides. 10 years later and I see similar trend with Raspberry Pi's and small microcomputers. It looks extremely cool and its fun to watch. Just like these inflatable made in china Intex toy boats 10 years ago.
You made a statement which has been debunked many times when it comes to water cooling. The statement saying that having the loop in series will make the pi's have hotter water the further you go in the loop. This has been proven by top water cooling pros to be a total non-issue. The water in the loop moves much too quickly for it it become noticeably warmer from when it gets from component to component or in this case from pi to pi.
its only 8 pis so I doubt it can do anything useful. Its a cool overkill project for learning and putting on youtube for ad revenue but its probably not very practice for anything in real life. Id love to be wrong though so if there is actually something useful that can be done with it that would be cool
I can definitely look into that. I was planning on packaging them together with the cooling block. International shipping becomes expensive (or takes forever) for smaller items, but if you can package a few together then it might be worthwhile.
Looks awesome! Super curious as to what you are using this for. A small suggestion, get some 90 degree connectors for your monitor so its not sticking out as much.
Thank you! I built it primarily to learn about cluster computing, but I'll hopefully be able to get it to run some simulation and rendering tasks eventually. Yeah they would make the cabling look a bit better. I actually debated soldering the cables to the underside of the Pi and the monitor so that they're completely hidden, this was just quite a lot of work!
Currently working on watercooling my pi just for fun. I did find a way to mount a small waterblock without needing a spacer. I just used a water cooling block that is typically used for the north bridge I found on ali express and 3d printed the bracket to attach it. wondering if water cooling will provide any better performance for overclocking!
Thanks for sharing, I'll have a look at the north bridge cooling blocks available as well. It definitely keeps the Pi cooler than any other solution I've tried, but for the price and complexity you're probably better off with something like an Ice Tower. You can still overclock the Pi to it's maximum without overheating and it costs around a quarter of the price. I did a comparison video a few weeks ago - th-cam.com/video/O6O_hcckXrg/w-d-xo.html
@@MichaelKlements I had already used the Ice tower and got to 2.4ghz stable. Just trying to see if I can get to my goal of 2.5. I also 3d printed/designed a simple mount.
If you go to 2.5Ghz, does the Pi overheat with the Ice Tower? The water cooling system would definitely keep it cool at 2.5Ghz, but you're likely to start running into power related issues rather than heat.
that cable management is just next level.
That looks gorgeous. It's appropriate that if there's a coolant leak that the whole thing will look like it's bleeding. :O
I bet even a person who has no clue what Raspberry PI is going to love this video. Well done mate.
@@pow274 Are you drunk? Wtf are you talking about?
@@pow274 bruh wtf? u high or smthing?
Love these types of vids, HATE when they use tools most everyday people do not have access to.
Would love to see how much processing power per watt you can squeeze out of it.
Useful processing power often isn't in a parallel form. If a task is highly parallelisable, you probably haven't optimised it very well... Still, high operation per joule figures get me kinda hyped ngl
high processing power per watt, but power is 20c/kwhr not 200/kwhr therefore a single second hand laptop is better. Its mainly autism + OCD but each to their own
This is why I still love IT and tech after studying/working with it for over 50 years ! Yep learned fortran as a 15yo :) Loved this video.
Thanks Paul!
Nice! I'm 15 and I learned C++
I'm at script-kiddie level at the moment though
GREAT BUILD!.... I actually would love you to cover the software isde of the cluster.
Thanks!
Thanks Miguel, I’ll definitely do so once I get it all up and running.
@@MichaelKlements cool. Looking forward to watch this. :)
+1 on this
Great build. But why not add an additional radiator and vent after right four Pi's, and you mostly have temps under control :)
@@MichaelKlements I don't really understand the purpose of this, because buying a micro motherboard with a modern CPU would have been surely way better use of money and time...
For anyone who knows about water cooling, having it in series is the only way to go. That water flow is usually fast enough to warm up evenly so there is no noticeable difference between each node. Usually what you have is the loop thermal balance.
jayztwocents did a good video about it.
Yup - you maintain the temps of the loop as a whole. That's all you do with watercooling. Flow rate is the main concern here to be honest due to the amount of blocks that restrict flow by design.
was ready to make the same comment, thank you.
2 components are a lot more different than 8 components in a loop with pressure drops over every connection.
it will matter and the later ones will be hotter. maybe not enough to thermal throttle, but they sure as hell wont run as cool as the first ones
but a rail system looks much cooler ;)
i suspect the blocks are simple in design and lend themself to slow flow rates rather than the high turbulence/high flow of a higher end pc type block with all their micro grooves etc.
over and over real world tests in pc world have shown loop order makes no difference in temps. ie sending GPU hot water to CPU doesnt make the CPE hotter because everything equalises.
having 8 equal boards rather than CPU + GPU maybe the 8th board will run hotter for a while but not over the long term and i think this is a system to be run 24/7 long term
personally id run a parallel water system on what im guessing is a slow flow rate just to be 100% sure everything heats at the same rate. even if the last board is hotter for a bit longer eventually i will shorten its life especially if this isnt a 24/7 system
Nice build! looks really nice, def love the red color!
Getting ideas? lol, I know I am.
If anybody is wondering about these two comments above me, yes, they are bots advertising a website that is specifically designed to take your money and info. I’m starting to see these bots more and more so be careful.
def love_the_red_color():
As a man of culture I cannot like a comment with 69 likes
this is really good stuff
No clue what this is for but looks really good
same feeling , like and want it but dont why, a made up excuse is coming for sure ;-)
Same here!
Clustering
Learning
Try out different technologies (SAAS/SARS or container swarms for example).
As a test environment to not have to run tests in production.....
And much more ;)
I guess these "build a RPi cluster" videos have the same reason... find somebody who will figure it out what is it good for.. or atleast i hope somebody will one day :D
useless lol just cute
that's probably the coolest pi cluster out there.
First of all, that´s an amazing job thank you so much for sharing with us... but i´ve seen in the video few air bubbles flowing into the circuit and it may be a problem because you lose eficiency in the heat transfer and air also affect to the pump, making it to cavitate. A posible solution, make sure that you don´t feed the pump with any air bubble just by setting it vertical, or rotating 180º the radiator to keep the air trapped in the upper part of the radiator.
Thank you for the time you spend making this video, and greetings from Spain. Keep it up!
Hi Jose,
Yes, there was still quite a lot of air in the system during the video as it was still only the first couple of minutes of the system running, so the system was still working the air out of all of the cooling blocks and radiator and that.
Yeah, ideally the reservoirs should be vertical as well, so that the air pocket is as far away from the pump as possible.
Thanks for the tips and the support!
Looks brilliant and love the black and red. Only comment with the actual build is I would have rotated the Pi's on the right by 180° so the coolant was on the right and mirror the coolant on the left.
Really interested to see the software site of this project.
It looks like a breaching charge that a SWAT team would use, looks awesome.
This is one of the coolest things I've seen in awhile!
Baaadoom pssshh
#1: Outstanding project, definitely an award winner.
#2: Excellent video editing, very well paced.
Thanks for the great feedback!
Next: "What to do with a rasbperry pi 4 cluster"
Indeed!
I agree! I think Raspberry Pi's are so cool, but I just don't know what the heck I would do with them...
See, I don't know what half of that is to know what I would do with them, lol.
@@r3d_panda Good use case. Could you share your cooling solution on your raspberry pi 4 ? I see many discussions about raspberry pi 4 could go really high temperature like 80 deg C and get throttling. Wonder how your case running so good.
@@AndrewTCW Honestly i don't have any long term data yet but i was barely able to let my pi reach 50 °C during stresstesting. I'm currently using the argon one v2 case. The complete upper part of the case works pretty much as a giant heatsink and the only time i ever saw the fan kicking in was the test mentioned above (there is software for this case available to setup a very basic fan curve).
Keep in mind i don't have a lot of permanent workload tho. Most of the time my pi idles with a workload of 0.3 (htop) and temps around 42 °C (pihole) with my current setup (13 Containers running in docker on a headless host with raspberry pi os lite installed). So i guess this highly depends on your use case and if you are willing to use at least some sort of cooling solution (passive is probably enough).
That thing is ridiculous...
I love it! ☺️
Amazing setup!! Definitely deserves more views!!
Thank you!
That sir is a work of art as well as functional.
“I’m sorry Dave, I’m afraid I can’t do that.”
That line still gives me shivers. Awesome movie!
@Hashir Easa 2001: A Space Odyssey. It might be old but that movie is gold!
I thought, i was the only one with Hal9000 flashbacks
@Hashir Easa *LOL*
Pile On ;-)
*"I don't know the reference ;(" -Hashir Easa*
Oh no
This is absolutely sick. Great job dude. Hand and technical is an awesome combo. Keep it up!
Thanks for the great feedback!
I'm sorry if this question is inappropriate: For what are Raspberry Pi clusters used for?
They're a relatively cheap way to learn and experiment with clusters and distributed computing. They're not competitive in performance, but when you want to learn how to configure scaling systems with load balancing and redundancy etc. it's cheaper than building or renting a cluster with actual PC hardware. Plus it's fun.
Bitcoin mining
@@lassmicharztichbindurch5319, with a Raspberry Pi? I reckon that would be quite inefficient.
This is seriously cool. Great work!
Loop order doesn't mater for temperature; only the thermal capacity of the radiator.
my main reason for subscribing is the cable management
Wow Maker Michael!! This is totally neat. I was wondering about surface mounting the radiator but then saw you had cut a hole through. I would love to know how the thermal performance the individual nodes. Look forward to seeing what you are planning to do with this neat cluster. Keep up the good work and congrats on the 10K subscriber mark!!
Thank you Deech! I'll probably do a follow up video on the cooling performance at some stage as well. I first need to figure out how best to log and display it. I did do a basic test, running all of them at full load for a minute or so and there was around a 3 degree difference between the first and last Pi in the loop and the system was only about 6 degrees above the ambient temperature. But this will probably change for longer runs and once overclocked.
Thanks for the great feedback and support!
Michael Klements I’ll be looking forward to the results video or whatever neat idea you want to show us all next.
Hell with the cluster. I want this guys tool shed. Damn.
Man this video makes me miss using laser cutters all the time. They're so nice!
Superb Michael!
Thank you!
Nice work! Really love how it looks. Please do cover all the software that you will be using on this in a video! Looking forward to watch it! Take care and be safe!
I definitely will do, thank you!
Epic build for the PI! Thanks for sharing!
cool build, the quesion is whats the use for a pi cluster ?
I built this mainly to learn about cluster computing, but hopefully I'll be able to get it to run some simulations and rendering tasks.
@@MichaelKlements etherium node?
@@TheSaintP1800 Wouldn't one Pi be enough? You'd likely want the 8Gb model, I suppose, but still. Bigger issue is the 32 ETH to stake.
@@MichaelKlements Kubernites?
@@TotesCray The bigger issues is using pi's for mining eth is like asking a child to pick up a hot tub. Any way you look at it, it's just not practical.
Just wondering if it keeps everything cool and what temp numbers you get while under load?
I did a follow up video where I tested the temperature under load - th-cam.com/video/3kghCftw_CA/w-d-xo.html
You should get a 90° micro hdmi cable in red for the master
A custom 15x40x40mm copper fin heatsink with a decent TIM adhesive will sufficiently cool a 2.147 GHz OC'd RbPI. If boards are under high utilization, a 120mm fan can remove excess heat from 5 boards. This together makes a high density, modular platform
One mistake is that pump and reservoir is on bottom, should be on top because you have residual air in the loop and it will be trapped in the top PI cpu block. Puting the reservoir on top the air will come back to it it would be no problem. Other then that i dont see anything else wrong here, everything is beautiful, great work.
I dont think there is much fins in the block at all, so the bubbles would flush out 1 sec after the pump is on, also remember we are talking about 10-15 watts of heat total here.
Air will normally not work its way to the top past the water in the tubes, in pc water cooling the top radiator or a cooling block will not fill with air even if it is over the water tank air level, only if the pump is sipping air and pushing it through the loop.
When the pump is at the lowest spot, the pump and loop will work until the entire loop is empty.
If the pump is higher, the pump will pump air before the system is empty.
But it really does not matter, i would put cable ties on all the hose connections though.
I fully expected that to end with,"and next week, we'll ask it how it feels, we connect it to the internet and I accidentally create skynet." Seriously engaging video, you built that amazing machine with the care of a craftsman. Thank you for filming it, as well as building it.
Thanks for the support and great feedback!
His raspberry pi setup is much better than my computer setup lmaooo
Thats sad.
@@TheChrimboEffect Why would it be sad?
Hi,you made my day. I love your works. English isn't my native language so I can't find words to explain how excited I am.keep on.
Thank you Nima!
Concerning your water temperatures: flow order does not matter and once the loop is saturated, all Pi's should run at practically the same temps.
💯
Yeah this is correct. Same as with a water loop in a PC, the loop with eventually reach its equilibrium and no individual component will get hotter water into its block than any other.
However it should also be noted that should one component be running at max and generating a lot of heat then all other components will receive the heated water from the loop.
This is just simply incorrect. what youre stating is that no heat would transfer to the water from progressive heat sinks. youre using the words "saturated" to mean its at the same temperature as the water which would mean no heat transfer takes place and is therefore useless.
@@t_c5266 you're wrong. But `okay`
@@Personalinfo404 prove it
Good project.
If you are concerned about water temperature, might I suggest you run the radiator and fan from the rear of the display and insert some thermal barrier to isolate the heat from the cold. Likely you would not require much but can likely cool the water by a degree or two under load. Also adding longer tubes would allow the water to cool more before it even reaches the radiator.
I would have gone with a circular arrangement myself by that is just personal taste.
Id love to see some CPU mining with this set up. Maybe do a price vs profit breakdown? Love your content!! Amazing work
I’ll have a look at this, a lot of people have asked about it.
@@MichaelKlements Yes please, I would love to learn more about the Rasberry Pi 4 GPU power and if it would even be a viable in profiting relatively quickly compared to previous versions that would only need 1000 years to make $5 through CPU mining Monero.
Great Build!!! Very Good!
Since its one Thermodynamic System, the temperature difference in the cooling Water will even out over time. Giving enough runtime every Pi will run at the same Temp.
Sorry, 4 the spelling I'm German.
i believe a reverse return piping system would fix your balancing issues cool idea looks great !
Looks really nice! I think turning the pi's on the left 180° would have looked a tiny bit better though - the network cables and tubing would be symmetrical then. But I guess that was not possible for space problems?
It is possible to do, the blocks aren’t in the centre of the Pis though, so then they’re slightly offset on one side. It probably would have looked a bit more symmetrical your way.
That looks fantastic and what a fun project !!!
I have no idea why you did this! lol! But somehow I am pleased you did :)
Really nice job!
awesome build! though you might wanna reconsider the zip tie on the leaking inlet. zip ties in water cooling, if used, are usually used to hold the tube in place, so it doesnt slip off of the fitting, not for ensuring a tight seal.
Should be fine, if only because the inlets are barbed.
Awesome project!!!
Красиво . На малинка нужна для экспериментов с железом. А для мини сервера лучше использовать mini PC .
Так он именно как кластер собрал, наверное как лабораторный стенд для распределённых вычислений. А такие заморочки с водным охлаждением чисто для прикола делаются, есть видео где люди просто собирают кластеры из малины без водяного охлаждения и все у них ок.
Excellent execution
Many thanks
@@MichaelKlements Many welcomes
Finally a way to keep the sahara heat the Pi produces out of my room...
Water-cooling is actually better at transferring heat from the rPi to the room. So if anything it'd be the opposite.
More like, A way that keep a Sahara heat in my room unless the heat exchanger is outside the room
The Pi 4 actually does need active cooling and this many individual fans would be much much louder than this setup.
the cabling is pretty much perfekt nice work
Because after a few minutes all the water will equalize in temperature there actually is no reason not to run them like that.
yep
Thanks Turbo Dakota. This seems to be the most likely result. I'll be doing a test video to demonstrate this in future as well.
Yeah a lot of people worry about it in regards of cooling CPU's and GPU's and at your first bootup, or while enterring a game, it seems like there is a difference, but when the water reach that temperature equilibrium, the orientation doesn't really matter anymore. Let it run untill it no longer increases in temperature, then run the tests.
(Which is why techtubers HATE testing water cooling gear)
Great build!
This is absolutely insane, could you run some Spark job on it? :)
I'll have a look at that, thanks for the suggestion
@Michael Klements I'd advise you to think about putting the reservoir vertically next time, because if you mount it horizontally, it might run dry so the pump might die
Awesome build! One thing I'll say, in case no one else has, the series run of each Pi in the water loop won't really end up increasing temps for the ones down the line as much as you might think. The thermal capacity of water and a 120mm radiator is going to be far higher than even 8 pis will overcome. Certainly the water temperature will increase as you add heat to the loop as a whole, but there shouldn't be appreciable difference between series vs parallel. As more heat is added to the loop as a whole, the water temp will increase, but the ordering doesn't really matter, in practice.
th-cam.com/video/RnPB_q51iVk/w-d-xo.html for more info and testing.
That video is using much higher wattage parts, as well, the 1080Ti consumes (and thus, converts to heat) 200+ watts, and the 6900K CPU is rated for 140 watts (before the increased voltage and 600MHz overclock he'd put on it.) Even if you're overclocking the Pi 4, it's still going to be dissipating less than 10 watts under load, so the total for the 8 Pi cluster will be well below 100W total. Your 120mm radiator is not going to have any problem at all with that heat load.
Thanks again for the videos! You've made me want a laser cutter even more than I did before!
Thanks for the detailed explanation John. This was pretty much my thinking when putting it together and it seems like there are a few videos out there which have tested different configurations with similar results. A couple of people have mentioned watching JayzTwoCents's video.
I must admit that when I bought the laser cutter I didn't think I'd use it nearly as much as I've landed up using it!
Thanks again for the input and support!
@@MichaelKlements I just left that exact comment and was about to mention JayzTwoCents etc. (on a side not LinusTechTips did one on radiator stacking, unrelated but useful to know if you ever wanted multiple rad's)
Yes your build is perfect, Good Job.
I'm looking forward to a future where liking these together with PCIe is a thing (like what's happening in the datacentre's) then the true scalable ARM supercomputer will be something anyone can have (Pipe dreams i know)
I was gonna say that too, eventually your loop will reach equillibrium. I'd imagine none of the water in that particular loop is going to be much more than ambient. It really wouldn't have mattered either way you set it up, so long as everything was getting adequate flow.
You are right. In addition; the water has so much flow (even on the lowest pump rpm) that the delta temp between each waterblock is zero, microscopic at worst.
It looks like a small army of laser tanks.. Boston Dynamics liked this video
running in series is fine, as long as the flow is above nothing. it will get into equilibrium.
Amazing!!! Great vision!
If you have all the pi's in a series, then once the water gets up to temp, they will all be the same temperature no matter what order they are in.
That's just not true. The temperature of the coolant out of the heat exchanger will be lower than on the return. As OP mentioned - the first Pi will experience colder coolant, then warmer, and warmer etc as you go down the chain. Once back to the H/E, the coolant will be cooled again and the process repeated.
@@5UPRAH it's been proven time and again. Once the loop gets up to temp it stays the same.
@@5UPRAH th-cam.com/video/RnPB_q51iVk/w-d-xo.html
@@robertshort9487 too many variables in that test and also not all the same components. The OG of the 8x Pi's video needs do a temp comparison of each of the Pi's CPU's and the coolant temp at the in/out of each block and the reservoir to make this a valid test and argument.
@@robertshort9487 also that link you sent, he is making a lot of assumptions. "If you were to...." - "you would probably...." etc etc. No actual proof - just his assumption. Fact is, if the radiator/Heat Exchanger is efficient, then the placement of the blocks in series vs parallel will matter. First proc is getting 20C coolant, 8th could be getting as much as 40C coolant. Core temp of the first will be much cooler than the 4th. It just will be.
fantastic work really nice 👍👍
Great idea. I was exploring some similar options too. This gave me good ideas. Thank you!
Awesome, let us know what you land up building!
Looks great. But for what could you use it?
I think most people (myself included) build them to learn about cluster computing, they're a great learning tool. Long-term I'll hopefully be able to get it running some simulation software packages.
@@MichaelKlements Thank u. I like this projekt. I like your use of the k40 4 cuttig plexiglas. Great work.
see, Luke Smith, *this* is how you become a digital landchad. forget renting a VPS
Everybody gangsta until your internet or power cuts off
All wires to inside and water tubes to outside would have been easier on us ocd people. Beautiful build either way.
Exponentially more powerful compared to my potato pc
DOUBT IT HONESTLY
Super nice..definitely do a follow up on the use cases.
Did I miss the part where he somehow plugged 8 Raspberry Pi's into a 6 port, 12amp charger?
I was wondering exactly how that worked too. Maybe he used a 2 port usb spiltter?
@@TmanT321 You can see he uses 2 splitters on the back side (in black). I found that curious because they didn't make the parts list. Normally I would overlook that, too, except he is EXCEPTIONALLY meticulous about naming all the parts down to the screws and even the machines he used to create them. Also, the budgeted wattage seems a bit low for that many pi's. I'm sure they will power up, but I wonder how they will fare under load.
Very nice project. Thanks for sharing.
Although it does hurt me a little bit to see paint applied to unprimed MDF. :')
But, can it run crysis?
I remember 10 years ago there was this trend on TH-cam where everyone was buying inflatable toy boats to mount floor, electric engine and expensive furniture on it and they sailed with entire familys while filming TH-cam videos. These were absolutely overkill mods for toy boats, and extremely dangerous rides. 10 years later and I see similar trend with Raspberry Pi's and small microcomputers. It looks extremely cool and its fun to watch. Just like these inflatable made in china Intex toy boats 10 years ago.
You made a statement which has been debunked many times when it comes to water cooling. The statement saying that having the loop in series will make the pi's have hotter water the further you go in the loop. This has been proven by top water cooling pros to be a total non-issue. The water in the loop moves much too quickly for it it become noticeably warmer from when it gets from component to component or in this case from pi to pi.
Fantastic design, haven't seen one like it yet, good job man
i just bought your plastic mini pc case for the pi4! from canada. and i got lucky to get one of the last! cant wait to look at it :)
That's great! Thanks for the support!
@@MichaelKlements keep up the good work bud! :D realy shows in your vids that you have passion for the product (sorry for my english ahah)
Overclock it and some gaming of course
hello, sorry for asking. why you need that many raspberry anyway? what is it for? thanks for answering
"i realized that if i ran them in series the last one would be hottest so i'll run them parallel"
fast forward video: they are ran in series
He said, that if he connected them in parallel they wouldn't get any water because of the lack of flow.
It looks super cool. Look forward to seeing what it can do.
its only 8 pis so I doubt it can do anything useful. Its a cool overkill project for learning and putting on youtube for ad revenue but its probably not very practice for anything in real life. Id love to be wrong though so if there is actually something useful that can be done with it that would be cool
Looks cool! In the last shot I missed some (led) light on the pi’s themselves. Maybe some rgb gpio headers?
That's a great idea, it would be a nice touch to have some LEDs indicating the load on each Pi.
Any chance you will start selling those cooling block atachments
I can definitely look into that. I was planning on packaging them together with the cooling block. International shipping becomes expensive (or takes forever) for smaller items, but if you can package a few together then it might be worthwhile.
As i remember the Bitscope have modules, cases for interesting clustering solutions.
If I might ask what would Some of the practical application to use a pi cluster for.
Im asking myself the sane thing. Im pretty sure its just for the novelty of it.
For that money you could buy a better "pc" than this
So what?
Woosh…
that is fresh as hell
Looks awesome! Waiting for the follow up video.
Coming soon!
Excellent video
Looks awesome! Super curious as to what you are using this for. A small suggestion, get some 90 degree connectors for your monitor so its not sticking out as much.
Thank you! I built it primarily to learn about cluster computing, but I'll hopefully be able to get it to run some simulation and rendering tasks eventually. Yeah they would make the cabling look a bit better. I actually debated soldering the cables to the underside of the Pi and the monitor so that they're completely hidden, this was just quite a lot of work!
Shut up and take my money!! Nice work, I'd really love to see one of this mounted on my wall, it looks really amazing.
Please tell me you used thermally conductive adhesive between the spacer and the blocks.
Yes, I used thermal paste
@@MichaelKlements Phew! Good to hear. I thought I was seeing hardware gore.
I want to build a pi cluster now! XD
Nice video.
You should add air assist to that K40
Currently working on watercooling my pi just for fun. I did find a way to mount a small waterblock without needing a spacer. I just used a water cooling block that is typically used for the north bridge I found on ali express and 3d printed the bracket to attach it. wondering if water cooling will provide any better performance for overclocking!
Thanks for sharing, I'll have a look at the north bridge cooling blocks available as well.
It definitely keeps the Pi cooler than any other solution I've tried, but for the price and complexity you're probably better off with something like an Ice Tower. You can still overclock the Pi to it's maximum without overheating and it costs around a quarter of the price. I did a comparison video a few weeks ago - th-cam.com/video/O6O_hcckXrg/w-d-xo.html
@@MichaelKlements I had already used the Ice tower and got to 2.4ghz stable. Just trying to see if I can get to my goal of 2.5. I also 3d printed/designed a simple mount.
If you go to 2.5Ghz, does the Pi overheat with the Ice Tower? The water cooling system would definitely keep it cool at 2.5Ghz, but you're likely to start running into power related issues rather than heat.
How did you realise the power supply for the Pis? You'll need at least 5V@20A, I suppose?