Why is my server not recognizing its new ECC RAM?

1. 1.On a PowerEdge R740, a blinking amber system health indicator means a system fault. Dell's manual sends you to the System Event Log, or the LCD panel where the bezel has one, for the specific message, and to its Event and Error Message Reference Guide for what the message means. We have not read that guide, so this page prints no Dell message IDs.
2. 2.The System Event Log is kept by the BMC, not the operating system. If the server boots once the new modules are out, read the log from the host. This works on any server whose BMC speaks IPMI.

   sudo ipmitool sel elist

   ReadingPrints every SEL entry with the sensor ID that raised it, using the SDR records. A memory event names the sensor or DIMM it came from: note the slot before you move anything.
3. 3.Write down the exact POST message or code. On a Supermicro X11SPM-TF with a Xeon Silver 4110, 64GB 2Rx4 RDIMMs stopped POST with 'No memory DIMM detected', 'Memory DIMM incompatible with memory controllers' and 'No system memory is physically installed', while 32GB 2Rx4 RDIMMs worked. That kind of message points at a module the platform will not accept: go to p3.
4. 4.Know what a broken rule does on the platforms we have documents for. Dell's R740 manual says the system might not boot or might stop responding during memory configuration. Intel's S2600WF rules say mixing DDR4 DIMM types (RDIMM, LRDIMM, 3DS RDIMM, 3DS LRDIMM) within or across processor sockets produces a Fatal Error Halt during memory initialization.
5. 5.Do not assume a memory-looking code is memory. On a Supermicro H12SSL-i with an EPYC 7302P, the POST code changed with the memory kit fitted, yet a new board, BIOS and IPMI updates and a CMOS clear did not help, and the fix was a replacement CPU. If the modules pass the rules in p3, go to p5.

1. 1.On a PowerEdge R760 (the manual we read), press F2 at power-on and open System BIOS > Memory Settings. It shows System Memory Size, System Memory Type and System Memory Speed.
   ReadingA System Memory Size below the total you fitted means modules are being ignored, the 'operate with reduced memory' case in Dell's R740 manual. On the R760, Dell says unbalanced or odd configurations lose performance and the system may not identify the modules being installed. System Memory Type should read the platform's generation.
2. 2.Ask the BMC over Redfish, the DMTF standard interface, from any machine that can reach it. List the Memory collection first to get each module's MemoryId, then read one.

   curl -sk -u <user> https://<bmc>/redfish/v1/Systems/<ComputerSystemId>/Memory/<MemoryId>

   ReadingPer module: BaseModuleType (RDIMM, UDIMM or LRDIMM), MemoryDeviceType (DDR4 or DDR5), RankCount, OperatingSpeedMhz, ErrorCorrection, CapacityMiB, PartNumber and Status. We have not verified which of these properties a given vendor's BMC fills in.
3. 3.From Linux, dump the firmware's memory tables. -t memory selects DMI types 5, 6, 16 and 17; each Memory Device (type 17) block is one slot. The step-by-step inventory, with the Windows equivalents, is on the installed-memory page linked below.

   sudo dmidecode -t memory

   ReadingLocator is the slot name. Type is DDR4 or DDR5. Type Detail reads 'Registered (Buffered)', 'Unbuffered (Unregistered)' or 'LRDIMM'. Speed is the module's rating and Configured Memory Speed is what it runs at. Rank reads 'Unknown' when the firmware leaves it 0. Part Number is what you check against your platform's list. The dmidecode manual warns that DMI data is often inaccurate, incomplete or wrong, so confirm a surprise in the BIOS or BMC.
4. 4.Cross-check with lshw, which reads DMI among other sources. Run it as root.

   sudo lshw -class memory

   ReadingOne entry per bank with its size and description. Without root its own manual says it reports only partial information.
5. 5.If the count is right but Configured Memory Speed or System Memory Speed is below the modules' rating, check the documented speed rules before treating it as a fault. Dell says modules of different speeds run at the speed of the slowest one, and the R740 supports 2933 MT/s only at one DIMM per channel. Lenovo's SR650 V3 runs DIMMs at up to 5600 MHz at one DIMM per channel and 4800 MHz at two with 5th Gen Xeon processors. Intel's S2600WF BIOS attempts the highest common frequency.

1. 1.Generation. Dell's R740 takes DDR4 only; the R760 takes DDR5 only and its manual says DDR4 memory is not supported in the R760. Supermicro's H13SSL-N specification lists ECC DDR5 RDIMM. Compare dmidecode's Type, or Redfish MemoryDeviceType, with your platform's specification.
2. 2.Class. Dell's R740 manual says RDIMMs and LRDIMMs must not be mixed. Intel's S2600WF and S2600ST rules say mixing RDIMM, LRDIMM and 3DS types halts memory initialization. Lenovo's SR650 V3 supports no LRDIMMs or UDIMMs at all. On a Supermicro X11DPH-T with LRDIMMs and RDIMMs mixed, POST showed an 'Incorrect DIMM Population' warning and the OS still saw all the memory; replies called it unsupported and warned a later BIOS could break it, and the poster went back to an unmixed layout. In the H12SSL-i thread, an unsupported ECC UDIMM gave a different POST code from the RDIMMs.
   ReadingIn dmidecode, every slot's Type Detail should read the same class: 'Registered (Buffered)' beside 'LRDIMM' or 'Unbuffered (Unregistered)' is a class mix. In Redfish, compare BaseModuleType across modules.
3. 3.Rank, DRAM width, density and capacity. These rules differ by platform, so read your own; the cross-vendor matrix is at /server-ram/mixing-rules. Dell R740: x4 and x8 may be mixed, up to two RDIMMs or LRDIMMs per channel regardless of rank count, at most two different ranked DIMMs in a channel and no more than two capacities in the system. Dell R760: no mixing of capacities, of x4 with x8, or of 3DS with non-3DS RDIMMs. Lenovo SR650 V3: no x4 with x8, no 9x4 RDIMMs with 10x4 RDIMMs or with 3DS RDIMMs, no mix of DRAM densities, and mixed rank counts are supported, but only two different capacities across all channels. Intel S2600ST: at most 8 logical and 10 physical ranks per channel.
   ReadingRead each module's Rank in dmidecode or RankCount in Redfish, and its Part Number, and check them against your platform's list. The fields named on this page do not include DRAM width (x4 or x8) or 3DS, so check those against the module maker's data for the part number.
4. 4.A module the processor will not address looks the same as a broken rule. In the X11SPM-TF thread, 64GB 2Rx4 RDIMMs would not POST on a Xeon Silver 4110 where 32GB 2Rx4 worked; disabling 'Enforce POR' and setting the speed by hand did not help, and replies blamed the CPU's per-rank density limit and suggested 32GB 2Rx4 RDIMM, 64GB 4Rx4 LRDIMM or a newer CPU. No fix was confirmed, so check your CPU's memory support as well as the board's.
5. 5.Population order and balance. The vendor-by-vendor slot order is at /server-ram/population-rules; the per-platform checks from the documents we read are these. Dell R740 and R760: white release tabs before black, and only sockets where a processor is installed; with two processors the R740 manual says the configurations must be identical and the R760 manual says they should match. Intel S2600WF and S2600ST: start each channel at the blue slot farthest from the processor; on the S2600WF a single DIMM in a channel goes in the blue slot. Lenovo's SR650 documentation says the order depends on the configuration.
   ReadingCompare dmidecode's Locator names, or the BMC's slot list, with the population table in your server's manual.
6. 6.HPE ProLiant owners: we have not found an HPE document stating HPE's memory population and mixing rules or its position on non-HPE memory, so this page gives none. Use the population and mixing pages linked below, which cite HPE's QuickSpecs, and your server's own documentation.

1. 1.Check that Linux EDAC is loaded and found the memory controllers. The per-DIMM counters in the next steps sit under each memory controller EDAC found.

   edac-util -s -v

   ReadingReports whether EDAC drivers are loaded and how many memory controllers were found.
2. 2.List the errors EDAC has recorded.

   edac-util -v

   ReadingWith no report option it lists any uncorrected (UE) and corrected (CE) error counts recorded by EDAC, with a DIMM's label where EDAC holds one in dimm_label.
3. 3.Read the per-DIMM counters directly. Some systems name the directories rankN instead of dimmN; change the path to match.

   grep -H . /sys/devices/system/edac/mc/mc*/dimm*/{dimm_label,size,dimm_ce_count,dimm_ue_count}

   Readingsize is in MB, so a module you fitted that has no entry, or a 0, matches the reduced-memory case. A dimm_ce_count that keeps climbing is, in the kernel documentation's words, an early indication that the DIMM is beginning to fail. A non-zero dimm_ue_count records uncorrected errors: errors above the correction threshold.
4. 4.Check what the EDAC driver logged at boot and since.

   dmesg | grep -iE 'edac|ecc'

   ReadingShows whether an EDAC driver loaded and any error lines it logged. The wording depends on the driver; the counters above are the record to go by.
5. 5.Dell's R760 manual lists a System Memory Testing option in BIOS Memory Settings that runs memory tests at boot. On the R740, Dell's embedded ePSA diagnostics start from F11 > System Utilities > Launch Diagnostics or F10 > Hardware Diagnostics > Run Hardware Diagnostics. The ePSA page we read does not say which tests cover memory.
6. 6.Run Memtest86+ for a test outside the operating system. It boots directly from BIOS or UEFI, from USB, CD or PXE. Its authors say that once an error is found, working out which module failed is not clear-cut, so narrow it down by testing fewer modules at a time.

1. 1.Read the BMC's slot list. On a Supermicro H12SSL-NT with eight identical 64GB 2Rx4 RDIMMs, the BMC showed DIMMA to DIMMD at 65536 MB and 3200 MT/s and DIMME to DIMMH as 'Not Present'. Swapping modules between banks showed the bank, not the modules, was at fault.
   ReadingIf the missing slots stay missing whichever modules sit in them, the fault follows the slot: look at seating, the CPU or the board. If it follows one module, that module is the suspect.
2. 2.Check the seating. Replies in the H12SSL-NT thread advised checking the CPU's seating and testing the board outside the case to rule out a short against a standoff. Dell says to populate sockets only where a processor is installed, so check that every bank you filled belongs to a fitted processor.
3. 3.Check the board's memory support for your exact modules. Supermicro's H13SSL-N page sets the supported speed by CPU series and board revision. On an ASRock Rack EPYC3251D4I-2T, two 64GB LRDIMMs that were not on the board's qualified list stopped POST at code b04b with no console video, ASRock Rack support said they 'theoretically should be' compatible, and the thread ended unresolved.
4. 4.If the modules meet the platform's rules, the slots are seated and the failure still does not follow the memory, consider the CPU. In the H12SSL-i thread the fix was a replacement CPU after the board had already been replaced.

We have not found a Dell document stating what iDRAC reports for non-Dell memory, or an HPE document stating HPE's position on non-HPE memory, so this page makes no claim either way. The forum threads cited here are all on Supermicro and ASRock Rack boards.

edac-util -s reports whether EDAC drivers are loaded and how many memory controllers were found, and EDAC's per-DIMM corrected and uncorrected counters sit under each controller it found. A BMC that follows the Redfish Memory schema can report each module's ErrorCorrection as NoECC, SingleBitECC, MultiBitECC or AddressParity, though we have not verified which BMCs fill that property in.

dmidecode prints Unknown when the firmware leaves the rank field at 0, and its manual warns that DMI data is often inaccurate, incomplete or wrong. Read RankCount from the BMC over Redfish if your BMC fills it in; we have not verified which BMCs do.

We have not found a vendor document saying a BIOS update changes which modules a platform accepts. In the threads cited here it did not help: BIOS and IPMI updates did not fix the H12SSL-i, disabling 'Enforce POR' did not fix the X11SPM-TF, and on the X11DPH-T replies warned that a later BIOS could break a mixed layout that booted.