Power Supply & Sync (Breakout Boards & PicoPSU)
The power subsystem is the foundation of a DePIN node’s resilience and continuous uptime (SLA). Four flagship RTX 4090/5090-class graphics cards, combined with a dual-processor server platform, can consume between 1.8 and 2.4 kW of power during peak load periods. Our goal is to design the power distribution network so that the motherboard gets off to a smooth start, while the heavy power load from the graphics cards is routed strictly around its circuits.
1. Server units (“skis”) and breakout boards
Powering such a system with standard home ATX power supplies is inefficient and expensive. Our choice is industrial-grade server power supplies (commonly referred to as “skis” from HP, Delta, or Lite-On) with a power rating of 1200W–1600W and Platinum or Titanium certification. They are designed for 24/7 operation year after year.
However, the server unit itself is just a bare metal bar with a blade connector at one end. It’s like “a ski without bindings.” To bring it to life and connect the cables, we need a breakout board (distribution board / “green board”).
Key functions of circuit boards:
- The distribution board snaps securely onto the server unit’s blade connectors and converts a single high-power output circuit into a set of standard 6+2-pin PCIe connectors.
- Important buying tip: Always try to purchase power supplies as a set with their original motherboards. This ensures a perfect fit with no play, prevents contact arcing under heavy loads, and saves you the trouble of searching for hard-to-find adapters.
2. The Role of PicoPSU in Booting the Motherboard
Since the industrial server power supply unit outputs only +12V power lines, it does not have a standard 24-pin home power connector, which requires +3.3V, +5V, and standby power lines to start the motherboard’s logic. To solve this problem and properly boot our Supermicro X11-DAI-N motherboard, we use a miniature, high-efficiency power supply— the PicoPSU.
Engineering diagram of the board circuit connections:
- Direct Installation: The PicoPSU board itself is inserted directly into the motherboard’s main 24-pin connector. It should fit snugly, all the way in until it clicks into place, completely eliminating any play caused by case fan vibration.
- Power Supply (+12V): A power input cable runs from the PicoPSU. This “tail” is connected via a separate, reliable wire directly to the +12V power connector on the breakout board of our First (Master) power supply unit.
Hard rule for the circuit: The main power supply unit, via the “Board–> PicoPSU–> 24-pin” connection, fully powers the startup and logic of the motherboard itself, along with all its built-in peripherals (processors, memory, IPMI, chipset, system SSD). However, it is strictly prohibited to draw additional power for graphics cards directly from the motherboard itself! All other power connections must be supplied directly from the power supply units’ distribution boards.
3. Dual PSU Architecture: Proper Synchronization of the Power Supplies
To provide 2.4 kW of power, we use two synchronized server units, distributing the load across a dedicated circuit:
- First unit (Master): Powers the motherboard (via the PicoPSU into the 24-pin slot and the stock cables into the 8-pin CPU connectors), the system drives, and the first two graphics cards (strictly together with their riser cards).
- Second module (Slave): Powers only the third and fourth graphics cards (strictly in conjunction with their riser cards).
- Synchronization: A synchronization cable (SYNC) connecting the backplanes of both units is used to ensure simultaneous startup. The power-on signal (PS-ON) is transmitted instantly. If the slave unit delays startup or shuts down before the master unit, the graphics cards will be left without power while the PCIe bus is active, which will burn out the controllers on the GPU boards.
4. The Deadly Trap of Adapters: SATA and Molex Are Off-Limits
Remember: The flagship graphics card and its high-speed riser card should be powered only directly from the motherboard using the original, thick cables.
Specifications for Connector Current Loads:
| Power connector | Maximum Safe Power | Actual consumption of the component | Consequences of Using an Adapter |
|---|---|---|---|
| SATA (Power) | up to 54 W | A PCIe x16 riser requires up to 75 W | The thin SATA contacts overheat, the plastic melts, and the card short-circuits. There is a 99% risk of fire. |
| Molex (4-pin) | up to 156 W (high-quality only) | High-Quality PCIe x16 Riser Card | Powering the riser is permitted only if the original cables are missing and strictly according to the diagram: 1 connector per line. |
| PCIe (6-pin / 8-pin) | 75 W to 150 W | PCIe x16 Riser / GPU Tails | The perfect choice. The safest way to power the power circuits of the riser cable and the graphics card itself is directly from the breakout board. |
Rules for forced switching:
- Powering the risers: Connect each riser to the power distribution board using a separate 6-pin PCIe cable. Do not use a daisy chain of cheap adapters.
- Power Requirements for Graphics Cards (RTX 4090/5090): Each card requires 3–4 separate 8-pin cables (or a single cable using the new 12VHPWR / 12V-2×6 standard). Each connector on the graphics card’s adapter must be connected via an individual, dedicated wire from the motherboard. The use of “splitter” cables (where two connectors protrude from a single wire) for cards of this caliber is strictly prohibited—the wire strands will not withstand the current and will melt.
- Single-Circuit Rule: The graphics card and the riser card into which it is inserted **must be powered by the same power supply**. A mismatch (for example, a riser from the “Master” and the card itself from the “Slave”) will cause a ground potential difference, and the power current will flow through the motherboard’s signal lines to equalize the potential, instantly burning out the PCIe lanes.