Use IPMI and iDRAC When the Server Screen Is Dead Remote Eyes Before You Open the Case

The BMC's state is the first branch of the whole fault tree, and it costs one ping to learn.

1. Browse or ping the management IP from any machine on the network. If the login page loads, standby power is present, the BMC processor runs, and part of the board is alive — the fault is downstream: main rails, VRM, or CPU.
2. If the address is silent, check the physical signs: the BMC heartbeat LED on the board (blinking means alive) and the link light on the dedicated management port.
3. A dead BMC plus a dead server points early: cords, PDU feed, PSU standby output, or the board's standby section. That is a power-input investigation, not a CPU one.
4. Forgotten IP? The BMC address is often shown on a chassis tag, in the router's DHCP list, or settable through the front-panel LCD where fitted. The BIOS setup — if reachable — also shows it.

Reachable BMC equals a witness on the stand. Unreachable BMC equals an earlier crime scene. Either answer is progress.

The System Event Log is the BMC's diary, timestamped and stored through power loss. It is the highest-value page in the interface for this fault.

1. Filter around the failure time. Voltage warnings name rails; power-supply events name modules; a power-fault assertion at the exact power-on moment convicts the power path.
2. Scroll backwards days, not minutes. Correctable memory error storms, thermal warnings, or a PSU predicting failure often announce the death before it happens.
3. CPU entries matter: 'processor presence detected' versus a missing or failed processor status separates a board that sees its CPU from one that does not.
4. Export or photograph the log before clearing anything. Cleared logs cannot testify later — for you, or for a warranty claim.

Technicians who read the log first often skip half the physical tests. The machine frequently knows what killed it.

The BMC watches the board's rails and component presence continuously. Its sensor page is a multimeter you already own.

1. Check the main rails against the healthy ranges the interface shows. A rail reading zero or wildly off during a power-on attempt is direct evidence, not inference.
2. The CPU core voltage line is the star witness for a cold CPU: input rails present but core voltage absent means the regulator stage never delivered — suspicion lands on the VRM and its enable path.
3. Presence and status lines for PSU modules show which modules the system believes exist and whether one is flagged failed or predicted to fail.
4. Honest limit: sensors report what the board's monitoring points see. A sensor reading nothing can mean a dead rail or a dead monitoring path on a sick board. Corroborate with one physical check before final verdicts.

Sensor evidence ranks below a meter on the actual pins but far above guessing — and it is available from your desk.

Beyond reading, the BMC can act — power-cycling a hung machine, and on licensed tiers, showing you a screen the dead monitor never will.

1. Use remote power control to attempt clean power cycles while you watch sensors and the log live. A rail that collapses at the same moment each attempt is a repeatable, recordable fault.
2. The virtual console (KVM over IP) shows POST and setup screens remotely. On Dell it needs the right iDRAC licence tier for full use, and HPE's iLO similarly gates advanced console features — the free tier may show less. Check your licence honestly before counting on it.
3. Even without a console licence, power control, logs, sensors and hardware inventory are typically available — more than enough for this diagnosis.
4. Inventory pages confirm exactly which CPU and DIMMs the board detects — a CPU missing from inventory while physically present is itself a strong clue.

If remote power plus live sensors reproduce the fault on demand, you can run the strip test and watch the BMC's answers change — diagnosis as a conversation, from another room.

The witness can also be a victim. Keep the interpretation honest:

1. A BMC that reboots, drops off the network, or shows garbage sensor values while the host is dead suggests the standby feed or the board region it lives on is unwell — evidence about the board, again.
2. A locked-up BMC sometimes revives by fully draining the chassis (cords out for several minutes — the BMC only dies when standby dies). A revival restores your witness, not the server.
3. Default or forgotten BMC credentials are common on second-hand servers; the vendor's documented reset procedure (physical access required) is the legitimate route back in.
4. If the BMC is alive and reports healthy rails, presence, and no faults while the host stays cold and silent, the fault hides in the sequence between 'power good' and 'CPU running' — board logic territory, and the strip and swap tests take over.

Alive-and-happy BMC plus dead host is not a contradiction. It is a map: everything the BMC can see is fine, so the fault is in what it cannot.