MPlate

The mPlate is Pulsed Media's custom-designed server mounting platform — a laser-cut 3mm aluminium tray that holds compute nodes, storage, power distribution, cooling, and monitoring in a standard rack form factor. It is the physical foundation of the MD (MiniDedicated) dedicated server product line.

The platform accepts multiple hardware form factors: mini-ITX motherboards, VESA 100×100 mini PCs, and STX barebones. In its 1RU configuration, an mPlate holds 2 nodes (N100 NAS with 6× HDDs each) or 8 nodes (HP ProDesk G3/G4 Mini compact nodes).

The mPlate is designed in-house by Pulsed Media. Metal fabrication (laser cutting) is outsourced to an Estonian workshop partnership; 3D-printed components, wiring, assembly, and QA happen at Pulsed Media's facilities in Finland. The entire Kerava datacenter was built with this platform in mind.

Design philosophy

The mPlate exists because no commercial server chassis does what Pulsed Media needs: pack many small, independent compute nodes — each a full dedicated server for a customer — into standard rack space with individual power monitoring, remote management, and the ability to swap hardware generations without redesigning the platform.

Commercial 1U servers give you one powerful machine. Blade servers give you vendor lock-in and enterprise pricing. The mPlate gives you up to 8 independent dedicated servers in 1RU, each with its own power feed, cooling, and management SBC — at a hardware cost that makes sub-€10/month dedicated servers economically viable.

The guiding principle: the platform accepts whatever hardware makes sense today and adapts to whatever makes sense tomorrow. When a better N100 board appears, or a new NUC form factor emerges, the mPlate doesn't change — the node does.

> "The beauty of our platform and design is that we can just 'slap in' whatever, and it scales. There are some quite interesting ideas! We can easily integrate any 1L MiniPC (Vesa 100x100 mount), mITX (mITX mounting pattern), or make adapters for say 4x4 NUC." — PulsedMedia on LET

How it came to be

The mITX experiments (2012–2015)

Pulsed Media first explored compact server hardware in 2012–2013, building stacks of mini-ITX servers on threaded rods with M6 extension nuts as spacers. These were not mPlates — they were improvised, janky, and used network storage instead of local drives — but they proved the concept: small, efficient nodes could deliver real hosting services.

> "We built our first 'stacks' of mITX servers back in 2012-2013. Damn those were janky, but rock solid performers too." — PulsedMedia on LET

By February 2015, the platform had evolved: custom 12VDC power distribution (860W and 1200W supplies at 90%+ efficiency), picoPSU DC-DC converters per node, welded HDD cages, and custom wiring. A single power supply could feed 64 nodes across 4 stacks. The bones of the mPlate concept — shared power, individual nodes, custom mounting — were forming.

The mPlate takes shape

The mPlate as a designed platform began development around 2023. The core design was established that year:

Laser-cut 3mm aluminium base plate — rigid, thermally conductive, manufacturable at scale through the Estonian workshop partnership
Standardized mounting patterns — VESA 100×100 and mini-ITX, with adapter designs for NUC and other form factors
Industrial MeanWell power supplies — currently MeanWell UHP-350 and UHP-500 (fanless, up to 94% efficient), replacing the earlier mixed PSU approach
Per-node power conversion — mITX nodes use picoPSU DC-DC converters from the shared 12V rail; 1L mini PCs (HP ProDesk G3/G4 Mini) use their own DC-IN power adapters
3 exhaust fans per mPlate — standardized cooling module
Per-group power monitoring — power meters on node groups tracking power quality, current, and total kWh

By the end of 2023, 18 different mPlate models existed. Assembly jigs, custom Kanban bins, and a growing 3D printing operation supported manufacturing.

Building and iterating (2024–2025)

Development continued through 2024 with hardware iteration and remote management work. Mass production is not yet achieved — the platform is still in the building phase, though 2025 saw significant production volume:

19 hardware models completed across the mPlate family
Remote reboot capability reached 50%+ of nodes
New power cabling reduced per-node assembly from 20 minutes to several minutes
mControl SBC management boards deployed — Milk-V Duo S computers providing per-node relay control, power monitoring, fan control, and sensor data via PoE
3D printing infrastructure expanded: a fleet of FDM printers (Sovol SV08, Elegoo Centauri Carbon, Creality K1 Max) producing HDD mounts, jigs, enclosures, cable management parts, and even functional M2-threaded fasteners

The N100 NAS variant — 6×3.5" HDD bays per node on an Intel N100 mITX board — entered prototyping in 2025, opening the path to high-density storage nodes on the mPlate platform.

Current state (2026)

mPlate V3 arrived from fabrication in January 2026 — simplified, production-ready
44 custom rail kits produced and in testing (65–90cm rack depth flexibility)
Custom PCB (mControl) in specification/design phase — integrates power distribution, relay control, and sensor monitoring into a single board, eliminating 2+ hours of manual wiring per unit
Petabytes of HDDs and hundreds of nodes in inventory awaiting deployment
Target: majority of inventory online by Q4 2026

Platform specifications

Specification	Detail
Form factor	1RU standard (scales to multi-RU for larger boards)
Nodes per 1RU	2 (N100 NAS with 6× HDDs each) or 8 (HP ProDesk G3/G4 Mini)
Nodes per 42U rack	~252 in Kerava (IP-limited: /24 subnet minus network + management IPs, 1 IP per node)
Base plate	Laser-cut aluminium, 3mm thickness
Power supply	MeanWell UHP-350 (350W) or UHP-500 (500W) — fanless, up to 94% efficient
Per-node power	mITX nodes: picoPSU DC-DC from shared 12V rail. 1L mini PCs (HP ProDesk): DC-IN adapter.
Cooling	3× exhaust fans per mPlate (standardized module)
Mounting patterns	VESA 100×100, mini-ITX, STX (with adapter)
Remote management	mControl: Milk-V Duo S SBC per 8 nodes — relay control, power monitoring, fan control, sensor data
Networking	Per-node 1GbE or 2.5GbE (board-dependent). 10G uplink per 8-node group. TOR: 48×10G, 2×100G uplinks
Rail kit	Custom design, 65–90cm rack depth adjustable
Assembly time	~8 hours per mPlate (4h mechanical, 4h QA/racking/cabling/documentation)

Supported hardware

Hardware	Form Factor	Mount Type	Use Case
Intel N100 mITX NAS (SZBOX, CWWK, Topton)	mITX	mITX pattern	Storage nodes — 6×SATA onboard, 2.5GbE, ~140W idle per mPlate
HP ProDesk G3/G4 Mini (7th/8th gen i5)	1L Mini PC	VESA 100×100	Current highest production volume. NVMe-only (no spinning disks).
ASRock DeskMini A300 / X300	STX	Mounting adapter + ultra low profile heatsink required	Ryzen APU nodes — migrating from ZEN MiniDedi
Experimental: 12×HDD NUC config	NUC	Adapter plate	2× Intel N100 + 2× i3-N305 NUC nodes with 12× HDDs. Experimental, not in production.

mControl — remote management

The mControl (working name) is a custom PCB being developed for per-node power control, relay switching, sensor monitoring, and fan management. It mounts a Milk-V Duo S RISC-V SBC and connects via PoE.

The mControl replaces the earlier manual wiring and standalone relay boards ("Ghosts in the Machine" — the first-generation SBC management boards). The integrated PCB design eliminates approximately 2 hours of manual wiring per mPlate.

Per-node management BOM (current SBC sidegrade):

Component	Cost
Milk-V Duo S	~€15
PoE power splitter	~€4
USB-C cable	~€1
PoE per-port cost	~€5
3D printed enclosure + hardware	~€10
USB-A cable	~€0.50
Total per 8-node group	~€35 (~€4.40/node)

Software architecture: identical SD card images across all units, DHCP with MAC as UUID, all configuration fetched remotely, sensor data dumped over network to conserve SD card writes.

The mControl board is designed for dual use — mPlate infrastructure management and standalone IoT applications. It will be available as a separate product.

Manufacturing

3D printing

3D printing is central to mPlate manufacturing. Pulsed Media operates a fleet of FDM printers producing:

HDD mount bays — 3D printed in ABS and PLA, glued with pressure jigs for alignment
Assembly jigs — ensuring consistent part placement during build
Cable management — custom clips, guides, and enclosures
Electronics enclosures — for mControl SBCs and relay boards
HVAC ducting adapters — custom-printed at €20 vs €500 market price for commercial equivalents
M2 threaded inserts — functional 3D-printed threads (printed vertically with tight tolerances)

Materials used: ABS (primary structural), PLA (with surface annealing for heat resistance), ePA-CF (carbon fiber — discontinued due to health concerns). Black or white filament used in production.

Assembly process

Each mPlate takes approximately 8 hours from production start to online and inventoried:

~4 hours mechanical assembly: mounting nodes to plate, drive installation, power wiring, fan module, cable management
~4 hours QA and deployment: testing, racking, cabling to switches, documentation, inventory

The custom mControl PCB is expected to cut the wiring portion significantly — currently the "total PITA" of the build process.

Fire safety

Fire safety is a primary design constraint:

MeanWell UHP PSUs have over-temperature protection with automatic cutoff
All 3D-printed parts use fire-resistant filament (black or white)
Staggered safeguards throughout the power chain

> "Fire safety was a huge concern, and we took it seriously. We've got staggered safeguards throughout." — PulsedMedia on LET

The Kerava datacenter

The Kerava datacenter's power distribution, cooling, rack layout, and network topology support dense mPlate deployments — hundreds of nodes per rack, with per-group power monitoring and PoE-based management networking.

Power efficiency

The mPlate achieves competitive power efficiency through the combination of low-power nodes and efficient shared power infrastructure:

Configuration	Idle	Spinup	Typical Load
N100 NAS (2 nodes, 12× 3.5" HDD total)	~140W	~308W	~167W
HP ProDesk G3/G4 Mini (8 nodes, NVMe only)	—	—	~170W

For comparison: an old Dell DCS6100 2U server with 3× Opteron 2419EE consumed 180–185W. A single mPlate with 8× HP ProDesk nodes delivers vastly more compute at similar power draw — and fits in 1RU instead of 2U.

Full rack target: ~7 kW for ~250 nodes in production. In a purpose-built datacenter like Kerava, power can grow to ~14 kW per rack.

> "N100 utterly obliterates the 2419EE in every meaningful metric. It's not even a friggin' contest. It's a gladiator vs. skeleton buried half a decade ago type of contest." — PulsedMedia on LET

Software

Several open-source tools were developed for mPlate node management:

mcxTemplate — templating engine for any Linux distribution
mcxRescue — live rescue image (forked from Finnix)
nodeAgentCollector — API endpoint for node telemetry
nodeAgent — barebones health metrics agent (opt-in)

Design goals

The mPlate is built around several non-negotiable design goals:

Hardware independence: The platform must accept new hardware without platform redesign. When the next generation of efficient SoCs appears, the mPlate adapts — new node, same plate.
Sovereignty: Own the hardware platform. No vendor lock-in to blade chassis, no proprietary backplanes, no licensing per node. Finnish stubbornness as competitive advantage.
Data-driven iteration: Deploy multiple hardware models, measure which performs best for customers, retire the rest. Data filters out the bullshit.
Boring at scale: The goal is mass production. Prototype complexity is acceptable; production complexity is not. When it's boring, it means it's working.

> "Goal is to make it boring tho... when it's boring it means we are mass producing them." — PulsedMedia on LET

> "This is HW Development at it's finest; 1 little piece missing causes 1 month of delay. No one ever sees that part, end users just click 'buy' and it just works — not knowing the pain and blood behind the other side of screen it took for you to be able to click that 'order now' button." — PulsedMedia on LET