40-Node Raspberry Pi Cluster: Introduction

This is an introduction to my custom-designed computer cluster based on the Raspberry Pi. More information can be found at likemagicappears.com/projects/....

[@simonw1169]

This dude is why pi were out of stock so long!!

[@eziohunt2751]

@No No but the parallelism is worth it

[@cedric1731]

@No No but the cores of that GPU don't have the compute power of one of the pis cores. This means that if you want to do like 100 computepower intensive jobs or so in parallel you are better off with something like this than with a GPU. You see for some jobs in IT these clusters make more sense than any GPU power house. You need to know why you buy something like that of course what he does.

[@cedric1731]

@No No I am sorry to correct you yes I have tried GPU programming and yes I know how fast it can be (I own a 1080 myself :)) but believe me CPU Cores have better performance per clock so if your workload is limited on 40 parallel calculations which need very much compute power each this cluster will be a whole lot faster than any of the somehow affordable GPUs...

[@cedric1731]

@No No man I respect your bitcoin faible I like it as well, I even used to mine with GPU and ASIC BUT and that's the important point I want to clarify here: there are usecases where you can't scale your problems on an infinite amount of cores as the problem might not be parallalizable. E.g. if you want to count to 100 it is very hard to synch 100 cores and not very effective, either. Instead you let 1 core count to 100. Your GH/s are how many hash calculations can be done in a second. Depending on the algorithm these problems benefit greatly from more cores. This means your GPU will score a whole lot faster than your CPU because the process of loading the next command into the Calculation Unit of each device will take some time and also your GPU is theoretically faster. The important part is now that although your theoretical performance of a GPU is higher the performance per clock on CPU beats the GPU by a fair amount. That means while the GPU still counts to 1 in the extreme case in one core (the other cores are on idle as explained before they cannot be used for this problem) your CPU already is at 30. You see if the GPU *could* use all cores it would be faster. But it cannot. To get back to the cluster: The cluster is made for a very limited amount of single core processing demanding tasks. This means it is *ultra* fast for e.g. calculating certain simulations where you have to take for example 40 variables into account. Each of the variables can be calculated separately but the problem which must be solved for each variable of the sim cannot be scaled on multiple cores. Now you can let your GPU solve this problem, too, but it would take even longer than your old Pentium 4 processor probably ;) And don't tell me I am wrong before not doing an other research tour... I know usually GPUs are faster but certain usecases require these very high amount of CPU instead of GPU cores. Note: the counting example isn't the best one but I think you get the point.

[@frankborik2816]

Nice design. Very well thought out and meticulous construction. Very impressive!!

[@nukeman239]

Thanks!

[@chiquiloso]

Sr. you have indeed built something really awesome, congratulations!

[@0xf7c8]

One of the coolest things i have ever seen on youtube as I am a computers engineering student! Good job men, loved it.

[@BrianB2579]

What a pleasure to see a person with such good talent........thanks

[@nukeman239]

Thanks, I'm glad you like it!

[@RMarkNolan]

Very nice! The custom case and the cable management are superb.

[@Random_name_42]

Very cool, thanks for sharing! I'm interested in building my own pie cluster for training neural networks so I'm very appreciative of demonstrations like this. And your website looks great too!

[@ljppp7]

Nicely done! Looking forward to the software, performance numbers, and interesting things you can do with a 40-node super computer!

[@rbaleksandar]

Totally cool. And the layout is really neat. Kudos for the work. Well done!

[@vgehts]

WOW! Great job, dude! That system you built looks frickin awesome!

[@TsubasaKato]

Wow!! I really like your cluster, especially it having 5TB of hard disk. I also like your LED lights on the case and fans.

[@TaylorLasch]

Please show a benchmark video of what these Cluster computers can do. This is the 3rd video of one of these on KZbin and no one shows what it can do. It looks sexy Im not going to deny that, but does it function?

[@stedunn563]

Exactly, I wanna know how they scale, compare gflops and all the goodies, it's like making a new car and having no idea how it drives! pointless.

[@sirgalahad777]

It is a waste of electricity. In reality it is useless unless you have a software which can take the most out of the many cores. Something like a render farm.

[@JediOfTheRepublic]

How is it useless? It's an educational tool. It isn't meant to replace a high end super computer. That isn't the point of this.

[@sync0x]

For those who have not been following the most significant years of computing, 1977-1986, 1982-2012 etc, you do not deserve to know what it can do.

[@newton3010]

I sort of agree with the consensus. It's nice n all that but what practical use does it have that can't be replicated with another device? Good for research in parallel and distributed computing maybe but i don't see much. like everyone else I've seen the cluster pi vids abut have yet to see anyone actually use it for anything.

[@atomicorang]

You have an awesome computer exactly what I'm interested in.. Great job and keep it up!

[@siesonekmk]

Very cool. I have 4 RBpi right now and am in the process of clustering them. I'm currently figuring out how to cram the boards, switch, usb hub and storage into a old 1U sun server case.

[@e1woqf]

Very impressive. Looks like a lot of work.

[@igornakayama4206]

Really nice work. Thanks for sharing with us.

[@steveblanchard3581]

Excellent job David.

[@NoahRevoy]

This is the best looking Pi cluster on KZbin.

[@tinotrivino]

woaow... this is literally AWESOME!!!

[@xornot2720]

Beautiful build!

[@jacecockayne2054]

I don't know what a raspberry cluster is or what it does but that looked really cool and I can tell you took some time .... Good job 😁

[@kainbrandt1820]

I'm with jace on this one what is it ???!!!???

[@rtesimpson]

very neat design!

[@JoeJohnston-taskboy]

Fantastic rig. Thanks for sharing.

[@Nesmothdesign]

amazing work!

[@Retrostasis]

Very cool, nice work.

[@STORM6490MT]

Kick ass build. Nice work.

[@boredgrass]

Steve Jobs looks down from his cloud full of envy! It's not only the beautifull look, but someone without understanding of supercomputers (or computers at all, for that) , can see, that it has to do with clusters of small interconnected parts, that work together in some organized mannor. Engineering meets art! Keep up the good work!!! :-)

[@zebratangozebra]

Fantastic build ! WOW

[@lcividin]

wow! this is really great!

[@jameskdobbins]

Do you have blueprints for acrylic and schematics for your setup? Looks very nice! Also have you considered Beagle Bone Black?

[@TediChannel23Ja]

gorgeous project

[@henrymonroy9533]

Awesome dude!

[@stephanevermette145]

Sir, allow me to extend a virtual high five to you. This is both an amazing feat of technology as well as a very attractive design!

[@godzillaganja]

Really love the casing and the whole setup in general! Is it fully custom and you built it with acrylic or did you use another case that you modified? The Cable management and disposition of the hardware is superb! Can't wait to see what you will do with this :)

[@juanmshaw5617]

Pretty good idea and execution. Two personal thoughts/questions on this: 1) I can see the whole enclosure and how neat it is but: What if you need to replace any of the modules? Whay would it imply? Would you need to turn off the entire thing o dismantle it? I'm thinking that if taking this to an enterprise level, "hot-swapping" capabilities of each modules (in a similar way an array of disks work) could be a plus. 2) There are new brands now on the market that feature double or triple capacity than the raspberry pi, such as banana pi for the same or even less money. It could be very cost-effective (volume, power consumption and efficiency) to start looking at those pieces of hardware. Cheers!

[@BGroothedde]

Nice rig man! You should do a tutorial on Raspberry Pi clustering and interesting applications!

[@clusterbit7301]

Awesome work

[@sedwardstx]

Beautiful work. i'll be interested to see it with software running.

[@nukeman239]

Thanks! I'll be posting video here of software running on it once I have it installed.

[@brunofunnie]

And we'll never see that software runing :(

[@SuperLbptutorials]

Could you theoretically use a setup like this for mining? I know that one pi unit doesn't have a powerful graphics processor, but if there was 40, would that be possible?

[@dat92kid]

subbed,that was frickin awesome :D

[@nukeman239]

Thanks!

[@vicktory4712]

this is great dave g. what's better than your 40 node cluster? your death stare before the lights go out. will subscribe.

[@StudioAnnLe]

That's a lot of Pi. Looks delicious =)

[@JaySmith-kq2jq]

well done and could be classed as art

[@BenByford]

What would you run on this ideally? could you use Docker for example?

[@Swelgje81]

When do we see it running ?

[@gregliako7995]

Hello Dave, would you like to show as what can this cluster so when it is connected on a deskop? It would be intersting to see the difference such with a render or a multicore use software. Excellent work

[@JardindeGuillaumeTV]

Beautiful machine !

[@wronski11]

I am curious what do you use it for. It is quite interesting as a project, but what is the real application of that, besides playing with mpi.

[@toxiiciity2355]

wronski11 he says in the beginning of the video

[@anupambatra]

folding@home

[@marcmuniz-hu9zj]

Not bad not bad cool idea I'm sure as time goes on this will only get better

[@vizionthing]

Beautiful. and artist as well as a scholar.

[@aabood007]

This is Amazing!!!

[@robertola3]

good job dude

[@johngrear6506]

Awesome job 😀

[@ezzekiel9201]

Very well done.

[@OVERLOARD949494]

This is MUCH cheaper then a PC with the same power, but you haven't got any support, so you have to do anything on your own. But this is some really good work.

[@JanNorten]

not at all its realy expensive i mean you need: 40Rpies @~30€ =1200€ plus an 48ports network switch ~400-600€(lets say 500) tons of cabels ~100€ fans plus Power supply ~100€ plus a LOT of other stuff I didnt even count in. In the end its about 2000€ and for the power i would say that a pc in that price category can outrun that cluster. EDIT: On his site he says that his system costs around 3000$

[@landwolf00]

I was wondering about this. Can 40 Rpi-3's outperform a 6900k processor with a 1080? Almost certainly not, but unclear.

[@germanmosca]

landwolf00, depends on what you wanna do. Most people don't build RasPi clusters to be faster than a home computer though. But for software development for big Clusters. Cheapest environment for such use cases.

[@aceofspades02]

Can I use for making super fast CGI renders with softwares like Blender, 3DSMax, Maya, etc... ?

[@EmanueleSantoro]

Very nice, congrats. How much power would such a thing draw? (avg/peak)

[@epsyuma]

A very nice looking build, but what can it do? Also, why is it important for the cluster to look good with the room lights off?

[@streamx2]

i have 3 raspberry pi's . The first two say PI R1 R2, while the third says raspberry pi 2 model b v1,1 2014. I would really like to set up a home network can i do this? Do i need to have the same models?

[@bigoxtailbone2]

can different development boards like raspberry pi, beaglebone black, pcduino be combined to make similar clusters? do you know of programs that can combine these boards to do this.

[@telefonkirtys]

I'm interested. Can this thing run windows or try linux v-ray stand alone for distributed rendering?

[@cosimoto1]

Oh good job! Just the way I would want to do it!

[@rajeshg8971]

good to see.... can we use it for 3d max .vray rendering??

[@charltonh2012]

It seems you can do just as well if not better using any processor technology fabricated at the latest/smallest (22nm) form factor running in pretty much any parallel system. It does not have to be RPi, but as far as I know the RPi2 does not use 22nm chips. So watt-for-watt performance will be better on a smaller fab processor. (thinking in watt costs more-so than equipment costs).

[@oldboy7856]

Thats tight af my guy a mini super computer cluster.. Let me know how works I would love to test it from my end

[@corporategamingtechnologie748]

Beautifully done , what the cost of the unit, spec sheet next try to make a tablet LFJ II Communicative Group salutes you, Sir

[@rushnerd]

This is very attractive and impressive! But what exactly can you do with it? Can you use their processing power in unison or what?

[@apofis231]

Number crunching, complex data analysis, raw computing, hashing. Do you think that computers are only used for entertainment? rofl

[@Killerwalrus234]

***** What does "Raw computing" even mean?

[@rushnerd]

***** Yes I am a stupid gamer and computers can only do entertainment, that is why I asked, thanks.

[@Killerwalrus234]

Ｒｕｓｈｎｅｒｄ If you're still curious it's most likely to develop network software, supercomputer architecture, or serve as a cloud storage/computing analog.

[@rushnerd]

Killerwalrus234 That makes logical sense, thank you.

[@302RELL]

looks cool!

[@kd1s]

That's pretty cool! Now just to find software to run across. I wonder, does mpiexec work that well across 40 nodes?

[@samsa9019]

hi what's your pi cluster benchmark score can you compare

[@bestaeroplane]

So far, there is many people already created, raspberry pi cluster. Bravo and kudo to you guys, woo-hoo!!!!! OK now, the next phase, to figure what we can do, with this cluster. Actually I am going to build my own cluster and, and, and, and later figure out, why did I create this for? Oh yeah, To test new softwares! I bet, not many computers today, can test new software.

[@uriskotec4040]

I am interested in building this do you have diagrams and materials list. I would also like to know if you have considered building a 30 cluster udoo x86 system . which system will have better performance .

[@brucecox5884]

I remember years ago people making clusters out of old Pentium systems at various companies, they even hired a tech or two to make the clusters, they mostly took out things like sound cards and such out of the computers and created a cluster machine for certain types of engineering analysis.The computers were all "scrape" from the companies due to desktop upgrades. so almost no cost. that being said I remember a video where some person took a high end Pentium 1 system that he used for rendering of graphic videos , did it on the high end machine then compared it to doing it on a 4x4 node of 486 machines,and got the same video done in a lot less time. I could see this being used for the same reason, Pixar started as a group of people,using the idle time of the SGI computers after hours to create the graphics. The PI is don't what is was made to do , give various hobby people a platform to experiment with at a cheap cost.I almost wish they came out with a version of the PI2 that was a bit stripped down, and made for making clusters All in all a vary good video, true hackers delight, reminds me of the good old days with the KIM 1 computer or the Motorola D2 kits.

[@KenSamaGomenasai]

Amazing sir, simply amazing... If I had the money to just throw around, I would do something like this with the new OODU board... 10x the power of a raspberry pi 3 in a single board.

[@Nilaytv]

what is OODU?

[@KenSamaGomenasai]

Nilaytv Look it up on kickstarter, it's a hackerboard like the raspberry pi only it's as powerful as most desktop computers.

[@TundralTV]

Could it run java programs with the computing power of all of the Pi's?

[@thedude4795]

great but what can you do with it, except spin it around on a desk?

[@cistano]

What power supply would I need to power 4 Raspberry Pi 3s?

[@1timby]

So are you planing on running Beowulf or just to use this as a server (file, application, etc)? Nice

[@dummeswe]

Cool stuff man!

[@playswithlife]

That is an awesome system.

[@RandomTorok]

Have you done any Linpack testing on this to get a flop rating?

[@Wumbledrive]

Great build! Why did you choose hard drives over SD cards though? I'm assuming you have them partitioned at 8 Pi per drive, correct? The read/write speeds would probably take a big hit, especially when distributed across a USB 2.0 hub. Not to mention that HDDs are more susceptible to drive failure, higher temperatures, and more energy consumption. Seems like a bad deal to me, over individual on-board SD cards. Either way, excellent build. I'm sure it looks even better in person. How are your temps on idle/heavy load? Lastly, have you considered a new build using the new $5 Pi? :)

[@Wumbledrive]

+Roger SD cards are also cheaper than the HDD. I just realized you are probably using the HDDs for external storage, not as boot drives. Duh! Otherwise it wouldn't be possible to partition them and boot from a USB hub. That's crazy talk. Anyway, really cool build! Thanks for making the video! :)

[@exaMAB]

I'm curious why you chose to use an ATX power supply which usually has much more emphasis on 12V power rails than 5V. Although I guess if you spend the money on 40RPi, the power supply is pretty much a detail.

[@DudeOfX]

as a commercial web server it can handle some damage too, right?

[@kidgoldenarm]

How does all the hot air exhaust in that thing?

[@ernstwigmann829]

what is used to test the computing power for this?

[@ESStonian]

Can you put ESXI on it and do virtualization stuff?

[@souljahmon16]

I would like to know the tested processing power of this cluster.

[@ansys_cfd]

Excellent video friend, What was the total cost? Do you try to run some CFD software or another? Regards

[@frozen_tortus]

this is really clever.

[@magna59]

Very impressive . Why did you decide to place the Pi s in a horizontal position ? If vertical , or semi vertical they could at least help remove some heat through convection & reduce the requirement of fans ...... but still project is very impressive . What 's the price so far ? $ 2.5K ? Regards from London (UK)

[@djburdine]

That was amazing

[@michaelolz]

What types of software do you plan on testing?

[@leeharvey3550]

Hey there, this was a year ago now, any updates? Thnx

[@herauthon]

Are these Ras-1B or Ras-2 devices so, you run a cluster management OS on there is that Linux or Something else.. (end of questions for now ;)

[@JacobMischka]

This is very cool

[@jamesgasek6983]

so cool!

[@brucecox5884]

had a second and third thought, if building clusters becomes more and more popular , I could see some kickstarter project, of making standard power supplies that are designed for a cluster builder, and a standard office type case. a PS with lots of micro usb power ports only having to supply the power for the cards. and like normal pcs I can see the growth of this card,the PI3 having more power and such,going form quad core to octal or higher

[@herauthon]

Bruce Cox first it needs to be useful - or it needs to be beautiful - which is inherent useful - then it needs to survive all the art critics..

[@BingChiller227]

Very impressive

[@SevenDeMagnus]

Hi. Great God-given talent. Is this also called parallel computing? Is this method the one used for data centers and render farms as well as supercomputers? Any idea or theory how to combine the power of Dreamcasts (lens are no good but the boards are ok) into one powerful Dreamcast scalable cluster that can be made for developing DC games that rival the modern console (just for technical fun) coz' the DC are still being developed for. Also any idea or theory how to make a cluster off Macs, say you have an old iMac and bought a new iMac, you'll double it's graphics power and CPU when it's doing parallel computing as well as have two monitors (scalable so everytime your iMac or Mac or PC becomes obsolete it can be added to the cluster so you can combine its power). Thank you. God bless. Rev. 21:4