ECC memory exists so that a server can survive memory errors, and most of the errors it reports are ones it has already fixed. The skill is telling routine corrected errors from a module that is starting to fail. This guide explains the two kinds of ECC error, where Linux, Windows and server management controllers log them, and a step-by-step process for deciding whether to reseat, repair or replace.
Correctable vs uncorrectable errors
| Correctable error (CE) | Uncorrectable error (UE) | |
|---|---|---|
| What happened | A bit flipped and ECC detected and fixed it. | More bits were wrong than ECC can fix. |
| Effect on the system | None: the system keeps running. | Typically a machine check: the process is stopped or the system halts or reboots to avoid data corruption. |
| What it means | Occasional ones are normal. A rising count on one DIMM is an early warning. | A hardware fault that needs action. |
Standard ECC uses SECDED: it corrects any single-bit error and detects double-bit errors in each data word, a 72-bit word on DDR4 and earlier (64 data bits plus 8 check bits). Server platforms add stronger schemes that correct the failure of a whole DRAM device: Intel calls it SDDC, HPE calls it Advanced ECC, and IBM introduced it as Chipkill. Supermicro's documentation describes SDDC as covering faults of up to 4 bits within one x4 DRAM device. For the basics, see the ECC vs non-ECC RAM guide.
Where ECC errors are logged
-
Linux: the EDAC subsystem keeps counters in /sys/devices/system/edac/mc/mcX (ce_count, ue_count, and per-DIMM dimm_ce_count and dimm_ue_count).
ras-mc-ctl --summaryand--error-countfrom rasdaemon summarize them, andedac-util -vreports the same counters. - Windows: Event Viewer, System log, source WHEA-Logger. Event 19 is a corrected hardware error, event 47 is reported for corrected memory errors, and event 18 is a fatal hardware error.
- Server management controllers: the BMC System Event Log names the DIMM slot. On Dell PowerEdge it is the iDRAC Lifecycle Log, on HPE ProLiant the iLO Integrated Management Log, and on Supermicro the BMC Health Event Log.
The BMC log is the most useful for hardware decisions, because it names the physical slot. Operating system logs often report a memory address or controller rather than a slot.
How to troubleshoot ECC errors step by step
- Find the slot and the type. Read the BMC log: which DIMM, correctable or uncorrectable, and how often.
- Judge the rate. Intel treats fewer than 10 correctable events per DIMM in 24 hours, with no crash, as within threshold; keep monitoring. A burst or a steady climb on one module needs action.
- Let the server repair itself. On servers with Post Package Repair, a threshold event schedules a repair at the next reboot. Dell says threshold events do not by themselves warrant replacing the module: reboot first, and do not move DIMMs until the repair has run.
- Update firmware. BIOS and BMC updates fix memory training and false error reports; HPE has issued fixes for false threshold messages on some systems.
- Reseat. Power down, unplug, and reseat the flagged module.
- Swap test. Exchange the module with an identical one in another slot. If the errors follow the module, the module is at fault. If they stay with the slot, look at the slot, the processor or the board.
- Run a memory test. Use the server's built-in diagnostics to confirm the result before you replace anything.
When to replace the module
- Correctable errors keep returning on the same module after reseating and a swap test.
- The server reports that Post Package Repair or memory retraining failed.
- An uncorrectable error recurs on the same module, or follows it to a new slot.
Our guides on how to identify a failing server DIMM and when to replace a server memory module go further. Replacement ECC memory: DDR4 RDIMM, DDR5 RDIMM, LRDIMM and ECC UDIMM.
Frequently asked questions
What is the difference between a correctable and an uncorrectable memory error?
A correctable error is detected and fixed by ECC, and the system keeps running. An uncorrectable error is beyond what ECC can fix; the Linux kernel documentation notes that the only reliable way to avoid data corruption is to halt or reboot, and servers typically crash or stop the affected process.
Are correctable ECC errors a problem?
An occasional one is normal and is the reason ECC exists. A rising count on the same DIMM is an early warning. Intel treats fewer than 10 correctable events per DIMM in 24 hours, with no crash, as within threshold.
How do I see ECC errors in Linux?
With an EDAC driver loaded, per-controller and per-DIMM counters are in /sys/devices/system/edac/mc (ce_count and ue_count). The rasdaemon tool ras-mc-ctl summarizes them with --summary and --error-count, and edac-util reports the same counters.
How do I see ECC errors in Windows?
Check the System log in Event Viewer for WHEA-Logger events. Event 19 is a corrected hardware error and event 47 is reported for corrected memory errors; fatal hardware errors appear as event 18.
When should I replace a DIMM with ECC errors?
Replace it when correctable errors keep returning on the same module after a reseat and swap test, when a server's repair feature (Post Package Repair) fails, or when an uncorrectable error recurs on the same module.
What is Post Package Repair?
Post Package Repair (PPR) replaces a failing row of DRAM cells with a spare row on the module. Dell, Lenovo and Supermicro servers can run it at boot or at runtime, and a successful repair can keep a module in service.
Buying for more than one system
IT teams and integrators that maintain server fleets can standardize on one kit per model. Request a quote with your models and quantities, or send us your system model and we will confirm the right module before you order.
Steps and rules are summarized from the vendors' published documentation; follow the guide for your exact model.