@@univera1111 Well it's mostly useful for companies, in the late 90's I was following a guide to set up a webshop, but setting up a webshop doesn't make you a company. Like cooking food don't make you a restaurant, there's much more to it. A more modern version of the LAMP would be L for Linux (unchanged) you can run basically any application on it no license required, A for Azure or AWS (cloud) instead of Apache, I know those aren't the same. M for MongoDB instead of MySQL, and P for Python instead of PHP.
Lots of experience with clustering went into this well thought out video. Thanks. 'Pi Dramble' great for learning clustering and testing. Sandia National Labs uses them as a low cost cluster theory tester.
I had been eyeing the Pine64 Clusterboard and Sopine A64 Compute Modules. I hope you mention why you chose the RaspberryPi and compare it to alternative solutions in the next video.
thank you so very much for this, been arguing the concept with my friend for a while now now i have the proof to shut him up and get on with our super computer implementation.
Just started playing with my own Pi cluster. I agree that the main thing is to have a testbed in actual hardware. It won't ever be super fast, but it's a great learning experience. Looking forward to your series.
I am not a software engineer and dnt no about coding n oll but as a power user, Cluster computing seems very cool, following many cluster projects it seems every project ends up with learning experience only.Really want that it gains some practicality like Powerful password crack-machine,Hosting website,Bulk photo encoder to auto-resize big DSLR images, and many more day to day work even converting(downsizing) collection of 1000s of Video clips and auto uploading to cloud to save a lot of space on cloud. If some how these above said work are possible on Cluster it will become the best time and pain saver of the world. Considering a 15 node cluster these tasks are not impossible but the problem is lack of interest of coders and developers in doing so. I really hope you will develop a cluster which will make life easy&fast. waiting for your upcoming videos.
I build a cluster many years ago using 8 retired 486 machines plus a master node. The buzzword in those days was "Beowulf Cluster". The control software i used was PVM - Parallel Virtual Machine with Linux installed on each machine. The network was an Ethernet bus using RG-58 coax cable - 10Base2. It worked but I did not really have a need for it so it went away. It was a fun learning experience. It is interesting to see what can be done with current hardware. Certainly a lot smaller package. My cluster was about 7-8 cubic feet and of course a lot slower that today's processors.
Apart from the fact that a bramble is an entire plant (roots and canes), and not the berries, this was a very interesting random TH-cam recommendation, that I will need to watch all the parts too lol.
Interesting video, there definitely exists some need for this board. People buying this hardware are most often hobbyists and while it is true that threadripper is likely too expensive for ”homelab”, but used server hardware is not. For example I paid ~600€ for two 8core server cpus (but my goal was to get faster cores, not as many as possible). My current home server has dual socket motherboard with a bit older cpus and ”only” ddr3 ram. Power consumption is definitely a bit higher, but for most people used hardware is probably a better deal than a raspberry pi cluster. Raspberry pi has probably less hardware compatibility issues so it is definitely easier to set up and there’s always need for this kind of bit more experimental setups, otherwise there wouldn’t be any new ideas and innovations, so keep up your truly interesting content!
Another cheap option for home-based computing clusters if you have the space is buying used 2U Dell or HP servers from hardware reclamation companies off eBay. For example, for around $150~$200 you can buy a Dell R710 with 2x Xeon E5xxx/X5xxx or similarly-spec'd CPUs (12 to 24 total threads), 24 to 48GB of RAM, tons of SATA3 hotswap bays and PCI-e 8x expansion options, 4x 1GbE, plus remote IPMI management. Granted, these machines are huge and heavy (~70lbs each), and ideally you would want to rackmount them (the rackmount rails are ~$70 by themselves, which is a bit annoying). Great option if you want to build a Ceph or other storage cluster, transcode video, render 3D stuff, etc. I use a TP-Link T2600G-28TS as my rack switch, which is ~$150 managed L2-ish switch that supports bonding/LACP, VLANs, and other stuff. Not as fun as getting a cluster of RPis working, though :)
id104335409 oh it goes deeper: a drupelet is "any of the small individual drupes forming a fleshy aggregate fruit such as a blackberry or raspberry"... and the Dramble runs Drupal! So each of the Pis is like it's own little 'Drupal'et... 😉
When I see a picture of someone with their mouth open on a show's icon "dumb" is the only thing I think. Congratulations, you are getting a comment, that will help your score. I even let the ads play while I type this. 👍🏻👍🏼 The sound is off and I intend to close this tab as soon as I'm done with this comment. Raspberry Pi clustering sounds interesting. I think I'll do a search and see if anything catches my imagination.
I would say to make low power servers, although I think that a high performance server with several virtual servers are probably more power efficient on a per server basis. In the end, I would say that this is more like a proof of concept that you can make at home for little money and learn in the process.
Cool think I've watched it with great interest because my uni thesis is exactly about clusters using SBC computers - I'm sure some of what you're doing will prove helpful as well as my research for you once it's done.
Ahhh, thats why I recognize you, I seen the bramble video lol. I literally bought my first pi because of your video bro lol. Dude the movie reference....
one of the cool things I think would be intersting to expand on with the zero boards is running a cluster within a single robotics project that may have a large chassis yet need IO management for motors, etc while running tensorflow over a single robotics system... a pi zero cluster would be easy to fit, be cheap and light as well... I found the reason for your rack finally XD.
Man thats a great Project, i am fixed, looks great, powerful, and that tower is quiet outstanding futuristic looking technology, that i can build alone...cool
@@JeffGeerling you're totally right I misread your video publication date! Maybe I'll edit it to "Update:" rather than correction. Sorry to sound critical
In every one of your videos, you do one small thing that reminds me to Like the video. This is also what got me to subscribe. I like your content! It really is the little things, like your movie homage and Red Shirt Jeff, that add a great touch!
I do the exact same thing. I build software on a raspberry pi and then deploy to the cloud. If it's fast on a raspberry pi then it's gonna be fast everywhere.
@@emeraldbonsai Not quite true. In terms of source code compilers do a lot of the work plus some VPS specifically use arm for them due to risc processing advantages over traditional x64 bit etc.
This is a great series, thanks for doing this! I wonder if they'll make one of these when the CM4's come out (seems like the pinout wouldn't be the same)
I have been told that they are anxiously anticipating the CM4 and plan on making a new model for it. If the preliminary rumors I’ve heard are true, they might even be able to reduce the cost a bit!
@@JeffGeerling Wow that sounds awesome. I've been using a RPi4 as a primary desktop for work (VDI helps a little for its shortcomings ofc) and it has been surprisingly capable. Very much looking forward to the CM4 version of the Turing, the cost/performance output potential is exciting :)
I am watching Apple's announced change to Apple Silicon (ASi) using the ARM instruction set and hoping to see Apple come up with a blade server or cluster capability. Now that I know the Raspberry Pi Compute Module exists, I see no reason why Apple couldn't build their own ASi-based compute module for those users who want a configurable workstation-level system (replace the Mac Pro) they can easily upgrade when they need to. Seeing the Turning Pi cluster board shows me Apple could do the same thing and create a very powerful Mac desktop/server/cluster without taking up a lot of space (6.7" square, ~1.5" high). The Turning Pi has a maximum CPU capacity of 28 cores (limited by compute modules? limited by number of slots--hardware or software restriction?), which is still a lot. Yes, Apple's A-series has more CPU cores than the Broadcom-based compute module but I'm sure Apple could easily duplicate this setup, providing a very powerful macOS-based system.
CM3 vs ZeroW is a no brainer for clustering for two reasons. First is 1U spacing ... the turing config maybe you can cram 28 in 1 x 1U server ... computationally the cost benefit does not make sense. ZeroW you can easily deploy 100 in a 1U space if u desolder the HDMI connector. Second is flexibility of deploment. CM3 requires specific socket placement, without ubiquitous support from the community and after sale innovators while ZeroW has a lot of opportunities for how to deploy in the field. ZeroW is truly a clustering tool for innovators looking to go industrial.
Horizontal scaling has limitations, too. Kubernetes is very cool for resilience, but adds huge complexity of its own--especially Helm and Tiller. It's great when someone else automates all the work for you and manages the cluster, but drudgery when you are responsible for all of it.
My first beowulf cluster was 32 dual cpu Celeron 300A's on Abit BP6 motherboards in 1998. You don't even want to know what that cost to build (64 processors, 32 systems each with 2 GB harddrives, 256mb ram, and all the other stuff you need to build a small server, the 100mb switches, ethernet cards, and so on). Funny thing is that for a fraction of the thousands of dollars it cost to build that cluster, a Threadripper 2990wx and ASRock X399 TaiChi was 32 cores/64 threads for just a few bucks over $1040 this past month (and much faster). Unreal.
These would be great for building a Skycoin Skyminer/node, Also great video explaining how to set the Turing Pi, just subscribed to your channel after watching the video. Keep up the great work :)
Hello Jeff, Can you consider making a RPi5 cluster video? In 2024 many of the RPi cluster videos are somewhat obsolete. Would using K3S with Rancher still be a good option with RPi5? Would Docker Swarm be a better option? what other options are there? Thanks!
Jeff, you should have mentioned about architecture being a huge drawback of using Pis. Not all software is built for ARM/ARM64. This severely limits what applications you are allowed to deploy to a Pi. I have a hard time recommending Pi to people who want to learn k8s unless they are familiar with building Docker images and compiling software for ARM.
We'll get to that, don't worry-episode 1 is intro, episode 2 is hardware, episode 3 is K8s setup, and either in that episode or ep5, I'll discuss the availability of images and other software for the 32-bit Pi OSes and ARM arch.
@@devinbuhl True; that solves the 64-bit part of the equation, though there are still plenty of applications and container images out there that are only built for amd64!
For 32 cores, I'd go with a dual g34 board with a pair of opteron, you'll get a lot better performance at the same price, even optimizing for core count. Though the other option is to go with an orange or lichee pi, to halve the price.
So I've got a cluster of 20 workstations. They are not band new, but at least have an i7 each. My total power consumption *when idling* is more than a thousand watts! Start some calculations and you're at 4 kW :D I think that's the biggest advantage of raspberry pis.
After you make these clusters, how do you utilize them? I'm interested in doing these sorts of projects but not sure what to do with them once I make them.
Definitely! There are also some other valid uses where it *can* compare favorably to other solutions for a production cluster, but it's definitely not going to rank up in the 'supercomputer' realm. Not for any reasonable price, at least :D
Hey Jeff, loving your videos! Can you add the date to your video descriptions? I find it really hard to watch a video and not knowing when it was released.
I just have a 2200G apu in a 1U case with flex-atx. Second hand set of IB 40Gbit cards, and then a 1920X attached to it in another 1U case (boots from a tiny xpoint drive to make sure it is up as fast as possible when needed), both tucked neatly under a minimalist desk. 1920X stays off, can be waken up by lan, and then used whenever needed. Software is a bit iffy at times, but the 1920X can be used as a rendernode, compute node and with a bit of messing around can be used to run like I run software locally on a lot more cores. As a bonus, I made it reachable from outside and can use it from wherever I am on the globe, but network speed than becomes an issue. Still fun though. Not sure if I should sell the 1920X, then add another 2200G and make tiny stackable cases and use a 3300X as the main one. Or maybe 2×3300X and keep the 2200G as the main. Clusters are fun.
Syfy had a self promo on there channel where the spaceship from the movie you reference was scratching vinyl against a dj which is what I think of when I think of the movie. That and mash potato replicas of rock formations.
I'm currently playing with Kubernetes on a cloud platform like digital ocean/AWS for video and audio processing applications. Multiple rpi 4 compute looks like a potential answer as currently transfer to/from cloud is the main bottleneck for large files. Can you recommend any toolsets that take the hard work out of determining whether a solution should be cloud, local or hybrid (without actually building both and running multiple tests)?
I think its funny how both the intel nuc and this turing board (to some extent) are actually going back to a really old computer form factor concept, with basically a "dumb" motherboard with all the compute capacity embedded in the daughter boards. The turing board in particular looks eerily similar to the earliest desktop pcs with the numerous perpendicular logic boards, albeit with far smaller components.
![ร้อยพ่อพันแม่ - WanMai [Official MV]](http://i.ytimg.com/vi/XY2pNYASEbg/mqdefault.jpg)
"Since my bramble ran droople, i came up with the portmanteau, dramble" - The Sims 4
I literally came down here to post the exact comment. Thanks for covering my nines LOL
Drupal
Great explanation. I appreciate that you explained simple things like what a LAMP stack is, instead of assuming the audience would know.
I appreciate that sentiment. I did not know what that meant until I watched this, and then researched it.
Is the LAMP stack still around? One of my first projects as a kid was setting up the LAMP stack, when I was done I was like: Now what do I do with it?
@@truedreams1 have you gotten what to do with it.
@@univera1111 Well it's mostly useful for companies, in the late 90's I was following a guide to set up a webshop, but setting up a webshop doesn't make you a company. Like cooking food don't make you a restaurant, there's much more to it. A more modern version of the LAMP would be L for Linux (unchanged) you can run basically any application on it no license required, A for Azure or AWS (cloud) instead of Apache, I know those aren't the same. M for MongoDB instead of MySQL, and P for Python instead of PHP.
@@truedreams1 so in a nutshell it's for web hosting of any services I can provide?
Jeff, you are an amazing teacher and have been very instrumental in expanding the Ansible community. Thank you so much for these videos.
Since my bramble ran droople I called it dramble
Lol sounds like intergalactic cable on rick and morty
It's Drupal, not droople. ;)
Lol, this is so accurate. I had to see if I was in a real fake doors commercial.
@@EvenTheDogAgrees idk Im pretty sure it is DrooOple?
Sounds like someone dropped acid.
I dont use a dramble, I use a grapple...
‘What is a cluster?”
Low hanging fruit....
Rather, low hanging raspberries.
Multiple computers connected on network to process ,
That Turing board should be called a Custard board; "A game changing Custard Pi."
A Cluster Compute Module... Or should I say, "Bramble Custard Compute Module".
I give this comment a golden sticker
Damn I said this but you beat me to it lol
Custard cluster
And powering all of this is schleem.
Lots of experience with clustering went into this well thought out video. Thanks. 'Pi Dramble' great for learning clustering and testing. Sandia National Labs uses them as a low cost cluster theory tester.
I'm a little jealous you've got your hands on one of those boards. I pre-ordered one just a week or two ago.
I don't get to keep it long... but I'm pretty sure I'll join you on that pre-order list soon!
Where did you pre-order it from? I can’t find the place for it
@@IhabSolimanM turingpi.com -- it looks like the first production run is sold out, though.
Yea, it’s been out of stock for a while now
3:25 A touch of brass for that steampunk feel. ;)
i didn't understand anything.
But it was amazing to watch!
I just ordered my first two Pi's...I was lost quickly.
I've been doing devops with ansible since 2018 and I learned a lot of it from roles that you wrote. Thanks for that! Glad I found your channel.
This is everything I've wanted. Bought a ClusterHat a year ago and had no idea what do with it. Informative and encouraging, I don't feel so lost.
I had been eyeing the Pine64 Clusterboard and Sopine A64 Compute Modules. I hope you mention why you chose the RaspberryPi and compare it to alternative solutions in the next video.
thank you so very much for this, been arguing the concept with my friend for a while now now i have the proof to shut him up and get on with our super computer implementation.
Just started playing with my own Pi cluster. I agree that the main thing is to have a testbed in actual hardware. It won't ever be super fast, but it's a great learning experience. Looking forward to your series.
I made a pi zero 2 w with 300 of them and it was crazy fast
I like that you named it after Alan Turing, the person who created the first supercomputer, the Turing supercomputer.
He didn’t name the board, the people who made it did
First time hearing you, instant subscribe after hearing your few point of views on systems design and architecture principles!
"including.making an homage to an old movie I enjoyed." - Immediately an ad for an Age of Empires mobile knockoff plays. 🤣
Love the Close Encounter reference.
I am not a software engineer and dnt no about coding n oll but as a power user, Cluster computing seems very cool, following many cluster projects it seems every project ends up with learning experience only.Really want that it gains some practicality like Powerful password crack-machine,Hosting website,Bulk photo encoder to auto-resize big DSLR images, and many more day to day work even converting(downsizing) collection of 1000s of Video clips and auto uploading to cloud to save a lot of space on cloud.
If some how these above said work are possible on Cluster it will become the best time and pain saver of the world.
Considering a 15 node cluster these tasks are not impossible but the problem is lack of interest of coders and developers in doing so.
I really hope you will develop a cluster which will make life easy&fast.
waiting for your upcoming videos.
"Close Encounters of the Third Kind"...
cool !
This guy is totally full of.......... interesting and fun ideas for computing!
I build a cluster many years ago using 8 retired 486 machines plus a master node. The buzzword in those days was "Beowulf Cluster". The control software i used was PVM - Parallel Virtual Machine with Linux installed on each machine. The network was an Ethernet bus using RG-58 coax cable - 10Base2. It worked but I did not really have a need for it so it went away. It was a fun learning experience. It is interesting to see what can be done with current hardware. Certainly a lot smaller package. My cluster was about 7-8 cubic feet and of course a lot slower that today's processors.
Apart from the fact that a bramble is an entire plant (roots and canes), and not the berries, this was a very interesting random TH-cam recommendation, that I will need to watch all the parts too lol.
Interesting video, there definitely exists some need for this board. People buying this hardware are most often hobbyists and while it is true that threadripper is likely too expensive for ”homelab”, but used server hardware is not. For example I paid ~600€ for two 8core server cpus (but my goal was to get faster cores, not as many as possible). My current home server has dual socket motherboard with a bit older cpus and ”only” ddr3 ram. Power consumption is definitely a bit higher, but for most people used hardware is probably a better deal than a raspberry pi cluster. Raspberry pi has probably less hardware compatibility issues so it is definitely easier to set up and there’s always need for this kind of bit more experimental setups, otherwise there wouldn’t be any new ideas and innovations, so keep up your truly interesting content!
I've thought it would be a great idea to put a bunch of Compute Modules on some sort of bus board for years. Nice to see I'm not the only one.
Another cheap option for home-based computing clusters if you have the space is buying used 2U Dell or HP servers from hardware reclamation companies off eBay. For example, for around $150~$200 you can buy a Dell R710 with 2x Xeon E5xxx/X5xxx or similarly-spec'd CPUs (12 to 24 total threads), 24 to 48GB of RAM, tons of SATA3 hotswap bays and PCI-e 8x expansion options, 4x 1GbE, plus remote IPMI management. Granted, these machines are huge and heavy (~70lbs each), and ideally you would want to rackmount them (the rackmount rails are ~$70 by themselves, which is a bit annoying). Great option if you want to build a Ceph or other storage cluster, transcode video, render 3D stuff, etc. I use a TP-Link T2600G-28TS as my rack switch, which is ~$150 managed L2-ish switch that supports bonding/LACP, VLANs, and other stuff. Not as fun as getting a cluster of RPis working, though :)
You nerd! 😄
I see you smiling as you explain how you came up with the name Dramble!
id104335409 oh it goes deeper: a drupelet is "any of the small individual drupes forming a fleshy aggregate fruit such as a blackberry or raspberry"... and the Dramble runs Drupal! So each of the Pis is like it's own little 'Drupal'et... 😉
@@JeffGeerling ha ha, naturally.
Awesome. Crystal Clear explanation. Support from India. Keep posting.
When I see a picture of someone with their mouth open on a show's icon "dumb" is the only thing I think.
Congratulations, you are getting a comment, that will help your score. I even let the ads play while I type this. 👍🏻👍🏼
The sound is off and I intend to close this tab as soon as I'm done with this comment.
Raspberry Pi clustering sounds interesting. I think I'll do a search and see if anything catches my imagination.
Incredibly straightforward and insightful vid. Thank you
Heck ya my dude. Thanks for all your awesome contributions. This was super cool.
thanks for being thorough in ur explanation as someone who didnt know the basics of servers xd!
Great teaching! You have a gift of explaining...
Thank you for making this video. I often wondered what advantages a making a clustered PI and what software would it run.
No advantage except fun. Every other computer is superior in every way to cluster.
I would say to make low power servers, although I think that a high performance server with several virtual servers are probably more power efficient on a per server basis.
In the end, I would say that this is more like a proof of concept that you can make at home for little money and learn in the process.
Cool think
I've watched it with great interest because my uni thesis is exactly about clusters using SBC computers - I'm sure some of what you're doing will prove helpful as well as my research for you once it's done.
Damn I have been using your ansible roles, didn't know you had a YT too
i was just about to say the same thing. those docker roles are great.
ive been interested in this idea for years, why did it just blow up now ive seen a few youtubers doing it
Excellent explanation. The diagrams are great especially comparing older ways of doing things versus newer (Scale up vs Scale out). Thank you.
Damn! Look at that stack of compute modules!
Wow. Super tempted to get that turingPI to play with K8
Looks like I have to watch this "series". My job wants to look at a Ceph clustering infrastructure. LOL
I hope someone besides me appreciated the Close Encounters themed tones 😋
4 Laptops on a table (in a square)..... this looks nice and organized :)
Looks the algorithms brought me another great channel!
Clustering multiple touring boards sounds great.
Ahhh, thats why I recognize you, I seen the bramble video lol. I literally bought my first pi because of your video bro lol.
Dude the movie reference....
one of the cool things I think would be intersting to expand on with the zero boards is running a cluster within a single robotics project that may have a large chassis yet need IO management for motors, etc while running tensorflow over a single robotics system... a pi zero cluster would be easy to fit, be cheap and light as well... I found the reason for your rack finally XD.
Man thats a great Project, i am fixed, looks great, powerful, and that tower is quiet outstanding futuristic looking technology, that i can build alone...cool
Very good! I am user pi3, i work in for develop my web with docker, is incredible solution low cost!
Thank you from Argentine!!🇦🇷🇦🇷🇦🇷
Update: Latest RPI 4 B has 8GB of RAM!
Note: Keep this series going!
Wouldn't you know they decided to release the new Pi just a couple days *after* I made this video :D
@@JeffGeerling you're totally right I misread your video publication date! Maybe I'll edit it to "Update:" rather than correction. Sorry to sound critical
Close encounters of the 3rd kind!
I only saw that movie for the first time 2 weeks ago.
Damn your good ! Loved the setup and how simple you explain things. Subbed!
The board should be call "clustard Pi" like Custerd pie
In every one of your videos, you do one small thing that reminds me to Like the video. This is also what got me to subscribe. I like your content! It really is the little things, like your movie homage and Red Shirt Jeff, that add a great touch!
Yay, Drupal! Can't wait to build a Dramble!
The Close Encounters shout-out was epic.
Finally I learned about it. Great explanation.
Thank you for covering this,really helped me alot.
Super informative video! Can't wait to see you explore the Turing motherboard
I do the exact same thing. I build software on a raspberry pi and then deploy to the cloud. If it's fast on a raspberry pi then it's gonna be fast everywhere.
@@emeraldbonsai Not quite true. In terms of source code compilers do a lot of the work plus some VPS specifically use arm for them due to risc processing advantages over traditional x64 bit etc.
Good stuff.
For
QBit[mm].Coil[1]=1:
QBit[mm].Coil[2]=0;
QBit[mm].Coil[3]=-1;
...
switch(priorState)...
Jeff dude, you are just amazing
This is a great series, thanks for doing this! I wonder if they'll make one of these when the CM4's come out (seems like the pinout wouldn't be the same)
I have been told that they are anxiously anticipating the CM4 and plan on making a new model for it. If the preliminary rumors I’ve heard are true, they might even be able to reduce the cost a bit!
@@JeffGeerling Wow that sounds awesome. I've been using a RPi4 as a primary desktop for work (VDI helps a little for its shortcomings ofc) and it has been surprisingly capable. Very much looking forward to the CM4 version of the Turing, the cost/performance output potential is exciting :)
This was quite informative.
It is really useful for learning purposes
Might be my next project. After a retro pi!
I am watching Apple's announced change to Apple Silicon (ASi) using the ARM instruction set and hoping to see Apple come up with a blade server or cluster capability. Now that I know the Raspberry Pi Compute Module exists, I see no reason why Apple couldn't build their own ASi-based compute module for those users who want a configurable workstation-level system (replace the Mac Pro) they can easily upgrade when they need to. Seeing the Turning Pi cluster board shows me Apple could do the same thing and create a very powerful Mac desktop/server/cluster without taking up a lot of space (6.7" square, ~1.5" high). The Turning Pi has a maximum CPU capacity of 28 cores (limited by compute modules? limited by number of slots--hardware or software restriction?), which is still a lot. Yes, Apple's A-series has more CPU cores than the Broadcom-based compute module but I'm sure Apple could easily duplicate this setup, providing a very powerful macOS-based system.
CM3 vs ZeroW is a no brainer for clustering for two reasons. First is 1U spacing ... the turing config maybe you can cram 28 in 1 x 1U server ... computationally the cost benefit does not make sense. ZeroW you can easily deploy 100 in a 1U space if u desolder the HDMI connector. Second is flexibility of deploment. CM3 requires specific socket placement, without ubiquitous support from the community and after sale innovators while ZeroW has a lot of opportunities for how to deploy in the field. ZeroW is truly a clustering tool for innovators looking to go industrial.
Horizontal scaling has limitations, too. Kubernetes is very cool for resilience, but adds huge complexity of its own--especially Helm and Tiller. It's great when someone else automates all the work for you and manages the cluster, but drudgery when you are responsible for all of it.
Rasp pi has 8gb of RAM now. Thinking of making my own cluster for some aero CFD. Thanks for this vid
My first beowulf cluster was 32 dual cpu Celeron 300A's on Abit BP6 motherboards in 1998. You don't even want to know what that cost to build (64 processors, 32 systems each with 2 GB harddrives, 256mb ram, and all the other stuff you need to build a small server, the 100mb switches, ethernet cards, and so on). Funny thing is that for a fraction of the thousands of dollars it cost to build that cluster, a Threadripper 2990wx and ASRock X399 TaiChi was 32 cores/64 threads for just a few bucks over $1040 this past month (and much faster). Unreal.
These would be great for building a Skycoin Skyminer/node,
Also great video explaining how to set the Turing Pi, just subscribed to your channel after watching the video. Keep up the great work :)
Hello Jeff,
Can you consider making a RPi5 cluster video? In 2024 many of the RPi cluster videos are somewhat obsolete. Would using K3S with Rancher still be a good option with RPi5? Would Docker Swarm be a better option? what other options are there? Thanks!
Well explained. Thanks
Well done! And to think, I was there at that original DrupalCon BoF! . . . #dramble
One of the OG!
I haven’t even used threading or parallel programming but yet here I am....
I imagine a cluster of 100,000 pi's as a epic gaming pc
Nice Strat!
Jeff, you should have mentioned about architecture being a huge drawback of using Pis. Not all software is built for ARM/ARM64. This severely limits what applications you are allowed to deploy to a Pi. I have a hard time recommending Pi to people who want to learn k8s unless they are familiar with building Docker images and compiling software for ARM.
We'll get to that, don't worry-episode 1 is intro, episode 2 is hardware, episode 3 is K8s setup, and either in that episode or ep5, I'll discuss the availability of images and other software for the 32-bit Pi OSes and ARM arch.
@@JeffGeerling Cool! You could also mention that Ubuntu has Raspberry Pi images that are ARM64, no need for Raspbian's 32bit OS on RPi4s
@@devinbuhl True; that solves the 64-bit part of the equation, though there are still plenty of applications and container images out there that are only built for amd64!
All the Open Source stuff is available. What are you missing?
@@lawrencedoliveiro9104 Huh?
For 32 cores, I'd go with a dual g34 board with a pair of opteron, you'll get a lot better performance at the same price, even optimizing for core count.
Though the other option is to go with an orange or lichee pi, to halve the price.
Opertons don't exist anymore
So I've got a cluster of 20 workstations. They are not band new, but at least have an i7 each. My total power consumption *when idling* is more than a thousand watts! Start some calculations and you're at 4 kW :D
I think that's the biggest advantage of raspberry pis.
After you make these clusters, how do you utilize them? I'm interested in doing these sorts of projects but not sure what to do with them once I make them.
In short, a Raspberry π cluster is a great training ground for learning to drive a full-on supercomputer.
Definitely! There are also some other valid uses where it *can* compare favorably to other solutions for a production cluster, but it's definitely not going to rank up in the 'supercomputer' realm. Not for any reasonable price, at least :D
Love the content. Keep it up bud!
Very nice work Jeff. How many videos will you be creating on this topic?
I'm planning at least three or four. The next video will walk through the basic hardware setup and getting an OS onto the boards.
Sweet! Love these sort of videos ;)
I don't get how useful it is, what for? What do you do with it? I mean, I have my own ideas, what are yours?
Masters of the Universe!
Hey Jeff, loving your videos! Can you add the date to your video descriptions? I find it really hard to watch a video and not knowing when it was released.
TH-cam always displays the upload date (at least on desktop, haven't checked on mobile) right under the title. At least it _should_.
@@JeffGeerling you're right on mobile but I can't see it on desktop. Weird. 😬 Thanks anyway.
I just have a 2200G apu in a 1U case with flex-atx. Second hand set of IB 40Gbit cards, and then a 1920X attached to it in another 1U case (boots from a tiny xpoint drive to make sure it is up as fast as possible when needed), both tucked neatly under a minimalist desk.
1920X stays off, can be waken up by lan, and then used whenever needed. Software is a bit iffy at times, but the 1920X can be used as a rendernode, compute node and with a bit of messing around can be used to run like I run software locally on a lot more cores. As a bonus, I made it reachable from outside and can use it from wherever I am on the globe, but network speed than becomes an issue. Still fun though.
Not sure if I should sell the 1920X, then add another 2200G and make tiny stackable cases and use a 3300X as the main one. Or maybe 2×3300X and keep the 2200G as the main.
Clusters are fun.
Syfy had a self promo on there channel where the spaceship from the movie you reference was scratching vinyl against a dj which is what I think of when I think of the movie. That and mash potato replicas of rock formations.
Sweet! Definitely a new subscriber here. Thanks for the effort!
Sounds like the tune that Gizmo plays in Gremlins.
Exactly minimum requirements to run the k8s :). This is fun and just makes me want to upgrade my dusty Pi
@santosh kumar see 5:10 and following ;)
I got new things here. Hence, I like and subs. Then explore other videos for more knowledge about Pi Clustering. Very interesting.
The traditional cluster looks cooler than the turing version in my opinion.
I'm currently playing with Kubernetes on a cloud platform like digital ocean/AWS for video and audio processing applications. Multiple rpi 4 compute looks like a potential answer as currently transfer to/from cloud is the main bottleneck for large files. Can you recommend any toolsets that take the hard work out of determining whether a solution should be cloud, local or hybrid (without actually building both and running multiple tests)?
I don't need another project but I really want to do this! I really need to learn ansible.
Lucky for you, I’ve been doing an Ansible 101 livestream series; check out all the episodes on my channel!
I think its funny how both the intel nuc and this turing board (to some extent) are actually going back to a really old computer form factor concept, with basically a "dumb" motherboard with all the compute capacity embedded in the daughter boards. The turing board in particular looks eerily similar to the earliest desktop pcs with the numerous perpendicular logic boards, albeit with far smaller components.
Outstanding vid, very informative; thank you! New sub.