Someone in a small-form-factor PC community took a compact chassis to retail: in-house CAD, an overseas machine shop for tolerance and inspection, packaging, an assembly guide. What stopped it was not the engineering. A move was coming, the inventory had nowhere to go, and a second production run was no longer feasible.
The hard part was never the drawing. Everything downstream had to be redone: tolerances, inspection, materials, maintenance. The same logic applies to shrinking an industrial PC. Every centimetre off the case has to be paid for somewhere. A more practical question: what was that part of the case doing?
A fanless industrial pc is where these trade-offs concentrate. There is no fan to move air, so heat leaves through the chassis. The housing is also the heatsink.
A smaller case has less surface area, so at the same power draw the surface runs hotter. The deciding metric is not the CPU clock but whether the dissipation budget matches the shell area.
USR-EG228 is built around the RK3506J, a triple-core Cortex-A7 at 1.2 GHz with 512 MB of memory plus 8 GB of storage. The housing is aluminium, the rated range is -25 to 75 degrees Celsius, the watchdog runs in hardware and software, and the storage sits on a removable TF card slot. It can be this small because the platform is low-power and the software only handles acquisition, conversion, and upload.
Once the workload goes up, the chassis has to grow back. USR-EG928A pairs an RK3588J octa-core with a 6 TOPS NPU, 8 GB of memory, and 128 GB of eMMC. Same aluminium housing and -25 to 75 degrees Celsius range, but the spec sheet also lists two expansion slots and the ability to ingest up to 32 network cameras. To keep passive cooling viable, the shell and expansion area must stay generous.
The check is measured. Take the chassis surface temperature during the hottest stretch of the year. Anything less than 20 degrees Celsius below the rated ceiling means it was pushed too far.
The first things to get dropped from a small machine are the DB9 connectors and terminal blocks. USR-EG228 carries 2x RS-485, 2x CAN FD, 1x USB, and 2x Ethernet, with no RS-232. If a site still has one device speaking only RS-232, the gap has to be bridged.
Smaller does not mean fewer ports, but the cost shows up elsewhere. USR-EG628 measures 118 x 87 x 60 mm and still fits 2x RS-485, 1x RS-232, 1x CAN, 2x Ethernet, and 2x USB, with wall and DIN-rail mounting. A small machine with that I/O density carries a more demanding thermal design and a higher unit cost.
The practical cost of dropping one serial port is rarely that the job cannot be done. It is that another converter ends up in the cabinet: another box, another cable, another power feed, and another point to reverify on acceptance. The volume saved on the PC is eaten by the sidecar, with an extra failure mode added.
The check is straightforward. Take the on-site equipment list, copy the I/O from it, and add one spare. On USR-EC300, two of the six serial ports can be switched between RS-232, RS-485, and RS-422. While the list is still being finalised, switchable ports are the safer bet.
Past a certain point of compaction, storage stops being replaceable. USR-EG228 keeps a removable TF card slot, with SD cards up to 128 GB. USR-EC300 uses a 128 GB wide-temperature SSD and a 255-level programmable watchdog. Both serve the same end. When something goes wrong, the fix is a card swap rather than a service visit or a teardown.
The check here is simple. If swapping a card requires opening the case, the box was pushed too far. Two things have to be done alongside the hardware choice. First, keep an archive of images and configuration exports, with filenames that include the site code and date. Second, flash a standby unit with the same image on the day of acceptance, and do a real swap at least once. An image that has never been deployed is not a backup.
The reason that project stopped is also the most common place where a custom-compact machine breaks. The housing and board are made for that one batch, and a different lot of components leaves the part numbers out of alignment.
The workable pattern is to keep two or three interchangeable steps inside the same product family. On the x86 side, USR-EC100 measures 159.6 x 127 x 53.2 mm, weighs 1.2 kg, ships with 2x serial and 2x Ethernet, runs on DC 12 V at roughly 15 W idle, and supports desktop, DIN-rail, and panel mounting. USR-EC300 is 236 x 183 x 72 mm with 6x serial, 2x Gigabit Ethernet, and DC 12 to 36 V input. USR-EC500 reaches 8x serial and 5x Ethernet. On the ARM side, USR-EG228, USR-EG528, and USR-EG628 share the WukongEdge engine and Node-RED, so application logic moves between models. The change between steps is the port count, not the program itself.
The check: write the lead time into the RFQ. Keep at least two interchangeable models in the same family. Burn the standby image using the method above before the unit is needed.
The standard explainer on a fanless industrial PC covers the same ground every time. The fan is the only moving part, the metal chassis conducts heat passively, wash-down and foundry environments are the canonical use cases, and the protection vocabulary runs to IP65 and NEMA 4X. A six-row comparison table also notes a lower upper bound on compute for fanless models, while higher-compute variants rely on external finned heatsinks for surface area.
That half of the question is whether the machine will break. What is not on the page: a bolted-on finned heatsink cancels the volume saving. A cut serial port means an external converter. A sealed storage means the unit goes back to base for service. Three years out, will the part still be available, and is there a model in the same family that can stand in.
The differences sit in four places. The thermal budget is sized to dissipation and case area, not to clock speed. The serial port count comes from the on-site list, with switchable ports left in for those not fixed yet. The TF card slot, the wide-temperature SSD, and the watchdog are written into the spec, not left as optional. The family keeps multiple interchangeable steps, and the lead time is on the table before the order is placed.
A fanless industrial pc is never a question of how small the box can get. The right size is where the cut would start charging the thermal, I/O, service, or supply side of the design.
1. Is a smaller industrial PC always better?
No. Compaction reduces the heat-spreading area, the number of ports the chassis can hold, and the ease of swapping storage in the field. The model should be chosen by the on-site I/O list, the measured cabinet temperature, and the service procedure, not by chasing the smallest available box.
2. How does a fanless industrial PC dissipate heat, and does a smaller case run hotter?
Heat leaves through the metal chassis, which conducts it from the internal components to the surface and from there to the surrounding air. The chassis is the heatsink. A smaller case has less surface area, so at the same power draw the surface runs hotter. The practical check is to measure the surface temperature on site and compare it with the rated ceiling, leaving at least 20 degrees Celsius of margin.
3. Will the serial ports on a small industrial PC be enough?
Compact units often keep only terminal-style interfaces such as RS-485 and CAN, without a DB9 form-factor RS-232. Before selection, list the port types and counts on site, and leave one spare. Models with switchable serial ports are the safer choice while the I/O list is still being finalised.
4. Can an external converter fill in for a missing serial port?
It can, but the cost has to be counted. The cabinet gains an extra box, an extra cable, an extra power feed, and an extra point on the acceptance checklist. It also adds another failure mode. The volume saved on the PC is usually eaten back by the sidecar.
5. Is there a size difference between ARM-based and x86-based industrial PCs?
ARM platforms run at lower power, so the passive cooling pressure is lighter and the chassis can be more compact. x86 platforms typically come with higher configurations and more I/O, and the housing grows accordingly. The choice is driven by what operating system the software expects, not by the dimensions.
6. What happens when the storage on a small industrial PC fails?
Pick a model with a TF card slot or a replaceable wide-temperature SSD, so that a failed card can be swapped and the image reflashed without opening the chassis. Confirm the storage is field-replaceable at selection, and keep an archive of images and configuration exports ready.
7. What are the risks of a custom-compact industrial PC?
When the housing and the board are made for a specific batch, a different lot of components can leave the part numbers out of alignment, and spares, acceptance criteria, and delivery documentation all have to be reworked. The workable practice is to keep two or three interchangeable steps in the same product family, and to put the lead time on the RFQ.
8. Are fanless industrial PCs and embedded PCs the same thing?
No. Fanless describes the cooling method, while embedded describes the installation form, mounted inside a machine or a cabinet rather than sitting on a desk as an independent unit. Em