Cluster Your Pi Zeros In Style With 3D Printed Cray-1

From a performance standpoint we know building a homebrew Raspberry Pi cluster doesn’t make a lot of sense, as even a fairly run of the mill desktop x86 machine is sure to run circles around it. That said, there’s an argument to be made that rigging up a dozen little Linux boards gives you a compact and affordable playground to experiment with things like parallel computing and load balancing. Is it a perfect argument? Not really. But if you’re anything like us, the whole thing starts making a lot more sense when you realize your cluster of Pi Zeros can be built to look like the iconic Cray-1 supercomputer.

This clever 3D printed enclosure comes from [Kevin McAleer], who says he was looking to learn more about deploying software using Ansible, Docker, Flask, and other modern frameworks with fancy sounding names. After somehow managing to purchase a dozen Raspberry Pi Zero 2s, he needed a way to keep them all in a tidy package. Beyond looking fantastically cool, the symmetrical design of the Cray-1 allowed him to design his miniature version in such a way that each individual wedge is made up of the same identical  set of 3D printed parts.

In the video after the break, [Kevin] explains some of the variations the design went through. We appreciate his initial goal of making it so you didn’t need any additional hardware to assemble the thing, but in the end you’ll need to pick up some M2.5 standoffs and matching screws if you want to build one yourself. We particularly like how you can hide all the USB power cables inside the lower “cushion” area with the help of some 90-degree cables, leaving the center core open.

This isn’t the first time we’ve seen somebody build their own tiny Cray-1. A particularly dedicated hacker built his own 1/10th scale replica of the 1970s supercomputer powered by an FPGA back in 2010, and eventually got to the point of trying to boot original software on it.

Thanks to [Xark] on the Hackaday Discord server for the tip.

36 thoughts on “Cluster Your Pi Zeros In Style With 3D Printed Cray-1

    1. A cluster of just about anything can be made to look like a Cray with a 3D Printer. A cluster of graphic cards for example, as used for video rendering or crypto mining.

      What is curious is how back when the Cray’s were top of the heap, the various people simulating climate and weather and many other things always claimed that if you bought them the next supercomputer they could finally solve these difficult problems. Well, this bundle of Pi Zero is probably at least 100 (100?) times better than the Cray and with massively more memory and is weak compared to what one can put on a desktop. Yet the same problems persist and the grad students now need to absolutely newest cutting edge supercomputers. Yes, they can use Amazon and Google server farms and vast networks of PC’s. Not enough! Which is fascinating and understandable.

      The key is the amount of money you can save if you distract them a couple years until the next iPhone is as powerful as that supercomputer they wanted in 2022.

    2. I visited Monsanto research headquarters in St. Louis in the late 1980s. They had a Cray-1 in their football-field-sized computer room (along with a dozen VAXen and an IBM 30XX series. The guy showing me around pointed at the Cray-1 and said “That is NOT a seat.” Apparently people were always trying to sit on it. It’s was Cray’s biggest design flaw. =-D

      1. MFLOPS/MHz is fun and just highlights how mind blowing the Cray-1 was when created.
        Cray-1 (1975, 46 years ago) 2 MFLOPS/MHz
        RPi Zero (2015, 7 years ago) 0.319 MFLOPS/MHz
        RPi Zero 2 W (2021, 1 tear ago) 5.1 MFLOPS/MHz

        And the RPi Zero 2 W is a quad core Cortex-A53

        1. You might if you create a cluster patch for box64 which can run crysis on arm64
          You just need to git clone ( https://github.com/ptitSeb/box64 ), fully redesign it to support running a on a cluster and then bob is your uncle!

          e.g.
          https://youtu.be/EaVGg8G1Ykk?t=223 (3 minutes and 45 seconds is crysis running)
          The example of crysis running on ARM64, was done on a Phytium D2000 Mini-PC
          CPU: Phytium D2000 ARMv8 octa-core Armv8 (FTC663) desktop processor @ 2.3 – 2.6 GHz with 8MiB L2 Cache (2MiB per dual-core clusters) and 4MB L3 Cache; 14nm process; 25W TDP; 1144-pin FCBGA package (35×35 mm)
          GPU: AMD Radeon RX 550 MXM 3.0a graphics card with 2GiB VRAM (PCIe card which could be upgraded)
          RAM: 16 GiB

          In the video it was only getting an acceptable frame rate ~ 17 fps (because the GPU was overloaded).

  1. humf, not a word about a Beowulf Cluster of those.

    He had the right idea with doing a rPiZ2 holder which was not solid and since he made the solid slices thicker why not just make the hollowed out slices thicker and bring in 4 screw tabs from the sides. Print way faster and more room for cooling via the back side of the rPiZ2s

    I think I would have opted to connect the power via the 40 pin connector. The rPi4 can’t be powered that way but sherly the rPiZ2 can.

    Fun project but now we know where a bunch of the rPiZ2s went. There’s been no stock for weeks in the US! https://rpilocator.com/?country=US

    1. You can power a Pi4 from the GPIO header if you want to (This Pi4 I’m using right now to type this one is powered that way), just make sure your supply is up to the task and know you have bypassed some protection circuitry attached to the USB power input.

  2. If they want to drastically increase computing power, simply putting some correctly oriented heatsinks on there would allow a nice overclock. Looks like the geometry would be acceptable to get some vertical airflow going through convection.

  3. i guess the first paragraph of the story says exactly what i have to say: why??

    this is usually a personal question for me…i look at some cool new device, or at the pile of cool old devices i already have, and i just can’t think of a reason for it. so i don’t buy it, or i decide to throw it out or at least hide it in a box in the basement.

    if for some reason you want to use an exceptionally slow computer to play with clustering, the clustering software — or at least the physical interlink — would be by far the bigger news than the plastic gimmick. and he even had the goal of going screwless and gave up on it?? come on, it’s not hard — i successfully went screwless for my 3d printed pi mounting.

    i’m on team hacks-don’t-belong-on-youtube, so please forgive me if i missed the meat of this project but it seems to me like it doesn’t have anything going on, he doesn’t even show us the mess of separate usb wall warts he’s got hidden off camera somewhere??

    there are two reasons i think it’s worth commenting on. first is that it represents the perfect convergence of two awful trends on hackaday: youtube and a new plastic box for a pi. second is that i fear it also represents pi itself. i fear the overwhelming majority of pis that are shipped wind up used for nothing more than an excuse to take a photo of a plastic box for social media. i don’t think people use them. pi has become an icon symbolizing a hack instead of the hack itself.

    i’m of mixed but generally positive feelings about the value of a linux machine that is so cheap that you will use it for a simple function that doesn’t really need that much power — in place of a 555, say — but i fear that the real story of pi is that most of them aren’t used at all.

    (yes, i know some are used)

    1. Click on the very first link in the above HaD article and all of your dreams will come true. Kevin did a great job writing articles to support the creation and use of his Pi cluster.

      1. awesome – thanks! it is indeed a much better presentation, making it much easier to verify that my concerns were unaddressed in this project. the hardware interlink documentation is only for the pi4 version of his cluster, for example, making it seem more likely that the zero version has a thoughtless interlink. and the presence of documentation on the software shows exactly how empty that effort is also.

        he suggests he needs a cluster just to install python under docker, even though whatever PC he has obviously can run as many docker images as he cares to. *shrug*

        i mean, i get it. it’s a toy. this content just makes me more confident that after he got done taking photos of his toy, he lost interest in it and stopped playing with it. it’s just sad. it hits me in a much deeper part of my soul than just the superficial fact that his primary goal is clicks. it tickles the part of me that is sad about my own lack of desire to actually play with most things, because they just aren’t fun…there’s no magic anywhere. if i’m trying to get something done, then the tool is a delight…but if i don’t have any use for it then all of the neato niftiness in the world just bums me out. and kevin clearly has no use for his series of clusters.

        emotionally for me this is a lot closer to reminiscing over christmas gifts from my childhood which lost their interest to me when i became only a little bit older.

        1. Real world practical usecases for most builds featured on this site are few, as most things on this site are as much learning exercises in how to make, and if you should make at all the expensive, high performance, practical and deployable version as they are about the built things…

          And that is a great thing, it creates ideas, builds skills, and can actually be useful, and with many builds you can’t be sure how useful until you have built it – I know one person ages ago set up a little cluster of Pi’s to do his renderings overnight because it was much more power efficent than the giant tower, just too slow to do so while you wait on it…

  4. That is neat. The neatest magic trick though is how he got a hold of enough Zero 2s….. I have one, but that was right after they came out. After that they boards have just vanished into the ether as far as I am concerned :) .

    I like!

  5. Sooooooo close.
    But hit by the oh-so-prevalent off-by-one error.

    The fully-populated Cray 1A/1B/1S/X-MP and Cray-2 had twelve columns.

    But this one goes to eleven, so that’s a win right there!

    Neat job, nice project.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.