An Elegant Modular Enclosure System For The Raspberry Pi 4

[NODE] has been experimenting with Raspberry Pi servers and mini computers for a long time, and knows all too well how the wiring can quickly turn into a rat’s nest. His latest creation is  the Mini Server version 3, a modular enclosure system for the Raspberry Pi 4, is designed to turn it into practical computing box.

The basic enclosure is a 92 mm x 92 mm x 26 mm 3D printed frame with a custom PCB top cover. One of the main goals was to collect all the major connectors on one side and make the micro SD slot easily accessible. To do this [NODE] created a set of custom PCB adaptors to route the USB-C and an HDMI port to the same side as the other USB ports, and move the micro SD slot to the bottom of the enclosure. A low profile adaptor was also designed to connect a mSATA SSD to one of the USB 3 ports, and there is space inside the enclosure for one or two cooling fans. Unlike previous version of the mini server, no hardware modifications are required on the Pi itself.

The only downside that we can see is that it doesn’t allow external access to the GPIO ports, but the entire project is open source specifically to allow people to make their own modifications.

[NODE] is a big fan of turning Raspberry Pis into custom computing devices, ranging from small terminal devices and pocket servers, to complete laptops.

45 thoughts on “An Elegant Modular Enclosure System For The Raspberry Pi 4

  1. The USB-C and HDMI rerouting can be seen in some commercial Pi 4 cases. The Pi can be a bit of a cable explosion, which reminds me to buy a proper Bluetooth mouse for mine.
    Obviously for small-scale server use-cases, you care less about the USB (except for external USB storage – a Pi with USB headers (instead of ports) and a case with a 2.5″ SSD bay and wiring from the header to the drive would be good).

    1. Right? The Argon One case I have reroutes HDMI and USB-C to the same side as the USB and Ethernet ports on the Pi 4. They just came out with a new case that does all that, but converts the HDMI ports to full size and adds in a USB3 to M.2 slot so you can have an SSD inside the case itself!

    2. I would be happy with one that just doesn’t have the big connectors (Ethernet, USB, HDMI, and GPIO) unpopulated, giving me much more freedom in laying out a low-profile enclosure while still having all of the features and performance of the 3B+ or 4B.

  2. Nice box, but i still leaves back the question – why wasn’t the RPi designed correctly from the start?

    Make the PCB a little bit bigger, and have all connectors on one side would make it much more versatile, so it is not necessary to make work-around-boxes to compensate for the bad design.

    And adding on board PSU circuitry (would maybe add 1 Euro to the price) accepting 6-15V input through a standard barrel connector would really enhance the usability of the RPi.

    1. Yeah, the RPi port layout is strange. I always assumed they wanted to make it as compact as possible to keep the total PCB area down for costs, but I have never seen an actual explanation for the weird arrangement.

      1. I would have to guess that you are correct, as the price point was so aggressively low on the RPi 1. Then I think they kind of got locked in with cases and other accessory support that it would hard for them to change the layout now. Although they do offer different form factors already…

        1. It has been mentioned by them that every decision at the start was to keep the price point down. So making it that size and aspect ratio is almost certainly because that is the most efficient tile size for the PCB fab they were using that can fit all the components they required.

          And as you say even if they would change it, it would just upset everyone over again – same way folks were anoyed when the form factor changed for the 40 pin GPIO even though its better. Just making the Pi larger to fit all the connectors down one edge doesn’t really make it better, infact I’d say it makes it much worse – how many pi deployments actually use most of the connectors – the industry ones probably only use the gpio and LAN. If connectors where you want them is what you really want grab the compute module – its basically the whole system on a DIMM just route the connections you personally care about.. And it sounds like the compute version of the 4 is supposed to be here any time now.

    2. One you go from qty 1-10, to qty 1M-10M, your component choices start to matter. Saving pennies starts to matter.

      Once you go from 1-20Mhz , to 1-2Ghz, your trace lengths start to matter, considerably. Savings mms start to matter.

      They have a very constrained budget target, they have a very constrained board design with the frequencies they are dealing with, with the qtys they are dealing with.

      EMI Certification starts to matter.

      This is not an arduino, or some other low speed hobby board, this is a computer.

      This is not an etsy hobby board, this is a commercial product.

      The armchair engineers have spoken it seems, again and again, without providing their schematics, or any evidence they have designed boards larger or more complex than simple 2 layer boards, with a few Mhz of clock speed.

      1. Thank you for pointing out what is obvious to anyone who has actually designed or built something: all designs involve trade-offs. Only Tony Stark and the designers at Marvel can do everything with just the power of ego, fanciful thinking, and thinking outside the box of physics and other constraints of the real world.

    3. Designed correctly for what purpose? I believe it is designed correctly for the purpose the designers had in mind, which may not be exactly what you had in mind, nor what somebody else had in mind…

    4. If you have all the connectors on one side of the board, you are effectively trying to fold the board in half. With a lot of effort this would not double the number of layers required, but it will definitely add 2 or 4 more layers to the PCB. So instead of a 6 layers PCB (RPi 4) at a baseline price, it would be in reality a 10 layer PCB which would about double that price of the baseline PCB. Or if you are very lucky an 8 layer PCB at only ~40% bump in price over the baseline. The reason I say 10, is for FCC compliance 10 would be just be simpler than 8 because of everything that is on a RPi board. If someone banged their head against a wall a lot and went slightly insane optimising with “via’s”, maybe 8 might be possible and still be able to pass FCC testing.

      The temptation would also be to reduce the dimensions of the board as well, but then you would have less surface area for air convection to cool the board. So if you reduced the board size then because of the same amount of heat being created in a much smaller area (higher heat per unit area) you would need to add active cooling which would also add additional costs, and another point of failure.

      I suspect the reason is to minimise the cost of every individual item on the BOM (bill of materials) is why all the connections are around the board as opposed to one side. Could it be done, of course it can but at a higher price.

    1. i love the “Argon one” case for that. it moves all connectors to the back. also adds cooling and power management with real shutdown and a start button, even though the “remote pi” board still does power management a little bit better.

  3. I had issues with my Pi 4 going into thermal throttling when running GNU Radio with several enclosures I tried. I solved the problem with a Flirc enclosure that has the entire mounting well made of cast aluminum with pedestals to contact the heat generating components. The case gets warm, but there is no annoying fan noise and no throttling. The down side is there is no room for HAT boards, but there is space to run a header cable out if required. The USB and HDMI ports are on different sides, but I am OK with that. The Flirc enclose is also relatively inexpensive.

    1. I had some copper memory heatsinks laying around from an old GPU mod, and they are the perfect size to fit the SoC on the Pi if you want to use a regular case. Here’s almost exactly the same part:

      I’ve tried to load my Pi down, and it never gets above 60°C. Usually it’s just running a PiHole DNS server, which doesn’t load it hardly at all and it stays at ambient.

      1. Can’t go wrong with using an before all words that start with a vowel, and a before words that start with a consonant that is pronounced at all, even if it may sound a bit vowel-ish.

        The H in Heroic and Historic should be pronounced with a very definite exhale emphasis so the listener hears it, not making the H silent or quiet. It’s Hĭ-stor-ic, not ĭh-stor-ic or ĕh-stor-ic. It’s Hē-ro-ic, not ē-ro-ic or ĕh-ro-ic.

        The word Hour, in contrast, has a totally silent H so should be preceded by an.

        These and various other reasons are why a general Pacific Northwest “accent free” pronunciation is what is most sought after by national news media. It’s easy to understand and we by damn learned the rules of grammar are the RULES of grammar. The general rules that almost always apply, then all the special ones for specific words and cases like totally un-voiced consonants, plus how certain words don’t follow the general rules for pluralization.

        In the first couple decades of television the sought after accent was “Mid Atlantic”, almost but not quite Bostonian. (Remember Dan Rather and Roger Mudd?) That sort of timbre but able to pronounce long and short vowels properly, along with most of the rest, but irritatingly lax on proper use of an and a few other things. Definitely NOT a “Jersey” accent that has no idea about long vowel sounds.

        We also learned that -1 and 1 are the only singular numbers. 0 is also treated as plural. So there are zero pages or there is 1 page. Yet all over the place online I see “1 results”, “1 credits left” etc. When did setting the correct terms to pair with -1 or 1 stop being a basic beginning programming exercise? It sets the learning stage for writing ‘traps’ with variables to trigger other things in a program.

        1. You are kidding, right? EVERY generation of parents tries to teach their children the rules of grammar and pronunciation, and EVERY generation of children says, “we will talk however we want. Stuff it.” And you can argue until your face is red, because you’re RIGHT, to no effect.

          What seems like a lax attitude toward the use of singular in numbers in software has just given in to general laziness and/or orneriness on the part of programmers, who as a rule (so to speak) are offended by rules that make their job harder and don’t make sense. So in another generation, the singular ‘1’ will disappear. Also what will disappear are the abbreviations for ordinal numbers (1st, 2nd, …), which have always been a chore for programmers. We’ll probably see these all become “th”, as in 1th, 2th, 3th, 4th, and so on. And the FOLLOWING generation will dispense with the original words completely, and will just say things like “we took oneth place!” “First” will become a variation “usually considered obsolete”.

          In my short lifetime, I’ve already seen a number of pronunciations change to make them conform to their spelling, which in most cases I think is a good thing. No reason for things to be inconsistent just because “they’ve always been that way”.

          But just to be clear, “literally” NEVER means “not really”, and “I could care less” is NEVER correct. That’s where I draw the line. Oh. And there’s no such thing as “five times less”, but that’s a fight I will never win.

        2. As far as the American TV accent, I don’t think it’s really Pacific Northwest – it’s really just west-coast. There has been some tendency to exclude California as a source for this, mainly because of the wildly exaggerated influence of the surfer dialect, which became valley-speak, and whatever that evolved into. Having spent over thirty years in different parts of California (and another thirty in Oregon and Washington), my observation is that the dominant California accent is not significantly different from those of Washington or Oregon. In fact, the most likely origin of the accent is Hollywood, since TV evolved on both coasts, but cinema was strictly a west coast thing.

    1. An excellent point. My only real complaint of the Argon 1 enclosure is the the led’s are not easily visible (they are visible through a translucent panel). This can be essential when you are trying to work out power issues, etc

    1. The ‘official’ pi display and many stands and cases for it are fine with anything put on top of the pi – just make sure to mount the pi to the back so the GPIO is accessible (which I’ve never done as the one time I used that screen it was all about compact footprint).

      Beyond that suggestion I don’t know – always made my own case if I wanted something that wasn’t stupidly cheap and easy to find..

      1. But there are TWO official pi displays, 5″ and 7″, and the “just stack it” concept doesn’t work so well with 7″. I do think the “Echo-Show” form factor is a good option for thess, but it looks like I’ll be printing of of my own once the bits and pieces I have on order come in.

  4. Very nice, you’ve got my attension with this one, quick question on the fan and vent ports side I was curious if by placing the fans on their sides and directing the heat threw a sort of (for lack of a better term) “think of a verial ramp, somewhat as skate boarders use this way you could intake air from the fan side and direct the air in and threw the little walled barrier via vent slots that take and aim the air downwards while the vertical ramp style inside curved the air right out of the top and if staked then to add to your modular design aim the stack vertical ramp air dump to bottom up which would keep air flowing in its single directional path while still offering some air flow from the first stack to run threw the bottom of second and again it would solidify a single directional air path way? I suppose I wanted to mention if this was considered or if someone with a 3D printer, might be able to try out, because when I think of server racks their usually in their tower cooled casing but the air floor usually depending on preference of brand and areas some tower enclosures intake air from a bottom a/c cooling system where its pushing cold air in and pulling the hot air out at the top, some times recycling the pulled air which runs threw its condenser and recooling the server, But also preventing moisture to not form inside the tower if the air is too cold and heats up to much their is a sort of trap like suctioning ventilation system which really its job is to just keep the airflow going will preventing cycling hotter air from getting the chance to form moisture. But that can be maybe to much maintenance and setting up much less R&D to create but I dont know, I just thought of it on the spot so🤷‍♂️😖I just wanna be helpful.

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.