Server No Display and CPU Stays Cold How to Prove the Motherboard Is Dead

Server PSUs hold more energy than desktop units, and their capacitors can keep a dangerous charge after unplugging. Set the ground rules before any cover comes off.

1. Pull both power cords on redundant-PSU machines and wait a full minute. Then press and hold the power button for 10 seconds to drain the board.
2. Never open a power supply unit itself. The capacitors inside a PSU can hold a lethal charge long after disconnection. A suspect PSU is swapped as a sealed module — never repaired on a bench by non-specialists.
3. Work with one hand where practical around live measurements, insulate meter probes, and treat every heatsink near the VRM area as potentially live until the cords are out.
4. Use an anti-static strap or keep a hand on the chassis. A static zap will not announce itself — it just quietly finishes off a weak board.

Skipping safety to save two minutes is how a repairable server becomes a dead technician story. Drain first, always.

A working CPU converts almost every watt it consumes into heat. Heat is not a side detail — it is the receipt proving the CPU received power and ran. A stone-cold heatsink after several minutes is the opposite receipt.

1. Give it time: run the server 3 to 5 minutes before judging. A CPU under no load warms slowly, but it does warm. Truly cold after five minutes means essentially no current is reaching the processor.
2. Compare against neighbours. The chipset and some board areas are fed by standby power and may feel slightly warm even when the CPU zone is cold. A cold CPU zone beside a warm board is strong evidence the CPU power stage specifically never started.
3. Touch briefly and carefully, or use an inexpensive infrared thermometer. A heatsink that is too hot to touch tells a different story (short or VRM fault) — note which one you have.
4. Do not let fans fool you. Fans and LEDs run happily on rails that have nothing to do with the CPU feed. Spinning fans prove almost nothing about the CPU.

Cold CPU + no display = the power-on sequence died before or inside the CPU voltage regulator stage. Diagnose power, not picture.

A server has a witness a desktop lacks: the baseboard management controller (iDRAC, iLO, IPMI). It runs on standby power the moment the cords are plugged in, completely independent of whether the server itself boots.

1. With cords in and the server off, look for the BMC heartbeat LED on the board — a small blinking light. Blinking means standby power is alive and the BMC is running.
2. Browse to the management IP (iDRAC / iLO / IPMI) from another machine. If it answers, standby power, the BMC, and part of the board are proven alive — suspicion moves to the main power rails and the CPU feed.
3. If the BMC is also dead and there is no heartbeat LED, the fault is earlier: cords, PDU or UPS outlet, PSU modules, or the board's standby section itself.
4. From the BMC you can often power the server on remotely, read voltage and presence sensors, and open the system event log — all before touching a screwdriver.

Two minutes with the BMC splits the fault tree in half. Never condemn a server board before asking its witness what it saw.

Elimination order matters, and the PSU is both the cheapest test and a very common culprit. Server PSUs add one gift: on redundant systems there are two modules to cross-check.

1. Check each PSU module's own status LED against the vendor's table in the manual. A solid amber or a dark LED on a plugged-in module is the module confessing.
2. On a redundant pair, run the server on one module at a time. A module that lights up yet collapses under load is a classic fake-healthy failure — the single-module test exposes it.
3. Measure the rails with a multimeter where your platform exposes test points or a backplane connector: the main +12V feed and the CPU's EPS 12V input should sit within about ±5 percent of nominal. A rail that sags or never appears is a PSU or short-circuit story.
4. Confirm the stupid things first, honestly: the PDU or UPS outlet is live, both cords are fully seated at both ends, and the PSU modules are pushed home until their latches click.

If a known-good PSU module brings the server to life, you never had a motherboard problem — and you found out for the price of a swap, not a board.

Now reduce the machine to the smallest set of parts that can possibly POST. Every extra card and stick is a suspect you are choosing to keep — remove that choice.

1. Power down, drain, and remove all PCIe cards, all drives, and every RAM stick but one. On a dual-socket board, the first CPU socket is the boot processor — the manual names it (usually CPU1).
2. Fit the single RAM stick in the exact slot the manual specifies for a one-stick configuration. The slot matters; the wrong slot imitates a memory fault.
3. Reseat the main power connector and the CPU EPS power connector — push until the latch clicks, then inspect for heat discolouration on the pins.
4. Clear CMOS exactly as the manual describes, then power on and watch: any change — heat, codes, beeps, a BMC log entry — is new information the strip test bought you.

A server that POSTs stripped has a healthy board and one bad add-on. Refit parts one at a time until it dies again — the last part fitted is the culprit.

The voltage regulator modules around the CPU socket step 12V down to the CPU's working voltage. When they fail, the CPU starves — cold — while the failed parts themselves may cook.

1. Look for bulged or leaking capacitors near the socket, darkened or shiny-discoloured PCB patches, and cracked or blistered MOSFET packages.
2. Smell the board zone (power off). A sharp burnt-electronics smell localised near the VRM is evidence, not imagination.
3. Feel carefully after a brief power-on: MOSFETs scorching hot while the CPU stays cold means power is being dumped in the regulator or a short — power off immediately and do not repeat the test for fun.
4. Honest grading: swapping a PSU, RAM or CPU is technician work. Replacing MOSFETs and capacitors is chip-level rework with a hot-air station and donor parts — a specialist's job, and worth it only when the board's value justifies it.

A motherboard earns a death certificate only after elimination, never after a hunch. You may conclude the board when every one of these is true:

1. Standby power state is known (BMC alive or proven dead at the PSU), and a known-good PSU or module did not change anything.
2. The strip test — one CPU, one RAM stick in the specified slot, no cards, no drives, reseated power connectors, cleared CMOS — produced no POST, no CPU warmth, no reaction.
3. The board's own reports (codes, LEDs, event log) either froze at the earliest stage or never moved regardless of configuration changes.
4. The VRM inspection or a CPU swap (next guide) removed the last alternative suspect.

If one of those boxes is unticked, the honest verdict is "not yet proven" — keep testing. The next guides take each branch of this tree in depth, and the final one covers rescuing the data and the repair-versus-replace decision before any money moves.