By manufacturer
Dell PowerEdge HPE ProLiant Lenovo ThinkSystem Fujitsu Primergy Supermicro IBM System x / Power Acer AltosECC RAM memory: bank-by-bank diagnosis, targeted replacement.
ECC errors above threshold, MCE events (Machine Check Exception), kernel panics tied to specific banks, DIMMs disabled by the BIOS: diagnosis starts from the BMC logs and ends at the single module at fault. Replacement with a DIMM matched on rank, voltage, frequency and type (Registered vs Load-Reduced).
The culprit is almost always already in the logs.
Enterprise servers log ECC errors from the very first event. The operational problem is that, up to a certain threshold, correctable ECC errors produce no visible symptoms — the system keeps running. By the time the symptom shows up (crash, reboot, kernel panic), the offending bank has already piled up evidence in the SEL.
A typical read is via ipmitool sel elist or the vendor equivalents (Dell racadm, HPE hponcfg, Lenovo xclarity): you look for a recurring Correctable ECC pattern on a specific bank, then any logging limit reached, and finally the Memory Device Disabled that marks the bank as no longer usable.
"A compatible DIMM" is not enough.
- Rank: 1Rx4, 2Rx4, 2Rx8 — not every rank is allowed in every combination of populated slots.
- Voltage: 1.5V vs 1.35V (DDR3), 1.2V vs 1.1V (DDR4/5). Mixing is possible, but the system falls back to the higher voltage.
- Frequency: the system falls back to the lowest frequency present. The mix depends on CPU + population.
- Registered vs Load-Reduced: cannot be mixed. RDIMM and LRDIMM have different electrical topologies.
- Vendor SmartMemory: HPE Gen10+ requires HPE-branded DIMMs to avoid persistent warnings on iLO. Technically compatible, but flagged as "non-HPE Smart" if generic.
The slot, not the module.
In rare cases the problem is not in the DIMM but in the motherboard slot (bent pin, degraded PCB trace). You diagnose it by moving the suspect DIMM to a known-good slot: if the error follows the module, it is the DIMM; if the error stays with the slot, it is the motherboard. In the latter case the slot can be isolated via BIOS (populating only the healthy slots), or it calls for motherboard replacement.
The questions we get most often.
Correctable ECC errors: do I need to act right away?
It depends on the volume. Sporadic events (1-2 a month) are normal and handled by the system with no impact. Recurring events that exceed the logging limit set by the vendor (typically 1000 in a day) are the signal that the bank is degrading: the sooner you act, the lower the risk of it turning into an uncorrectable ECC error and crashing the system.
Can I mix DIMMs from different vendors?
Technically yes, as long as they match on the same parameters (rank, voltage, frequency, registered/LR). In practice, on vendor-locked systems (HPE SmartMemory, Lenovo TruDDR) the system works but logs persistent warnings. For support-critical environments our recommendation is homogeneity per populated slot.
What does a typical DIMM replacement cost?
It is one of the lowest unit-cost jobs in the server hardware repair range. Variables: module capacity (16/32/64/128 GB), type (standard RDIMM vs Vendor SmartMemory), speed, procurement channel. A detailed estimate comes with your quote request.
Do you run an extended memtest86+ before/after the job?
Yes, when the symptom calls for it or when the system allows offline execution. On business-critical production servers we rely mainly on the SEL/MCE logs, which have the same diagnostic sensitivity without requiring downtime. After replacement: a stress test (extended memtest86+ or stress-ng --vm under Linux) for 4-12h before returning to production.