Uniqcli

DDR4 ECC server memory

Every DDR4 module whose manufacturer states error correction — registered RDIMMs, load-reduced LRDIMMs, unbuffered ECC UDIMMs and ECC SO-DIMMs. Filter by form factor, capacity and speed.

Showing 24 of 2,088 DDR4 ECC modules

Browse the full catalog

Specifications are read from the manufacturer product title as supplied to us; where a value is not stated it is left blank rather than estimated. Every line is screened for TAA country-of-origin and NDAA §889 status before checkout.

About DDR4 ECC server memory

Error-correcting memory carries extra DRAM devices beyond the 64 bits of data a rank presents — 72 bits in place of 64 on a DDR4 module — and uses them to detect and correct the single-bit errors that occur in any large memory array. The correction is transparent, the event is reported to the platform rather than silently swallowed, and the practical result is a machine that logs a failing module instead of corrupting a database or panicking at three in the morning. It is not a performance feature and on most server platforms it is not optional equipment: they will not post without it. Both the processor and the board have to support ECC for any of this to happen, which is why it is standard on server and workstation platforms and absent from most consumer ones.

The form-factor filter is the one that matters most here, because the three server module types are not interchangeable. An unbuffered ECC module (UDIMM) wires the address and command lines straight to the DRAM chips — simple, low latency, and limited in how many modules a channel can carry, which makes it the entry-server and workstation option. A registered module (RDIMM) puts a register between the memory controller and those chips, cutting the electrical load so a channel can carry more modules and larger capacities. A load-reduced module (LRDIMM) buffers the data lines as well, which is what makes the very largest per-channel capacities possible. A platform accepts exactly one of these; mixing registered and unbuffered is not a configuration a server will run; and the speed you actually get depends on how many modules per channel you populate, since two per channel commonly clocks lower than one. Read the server's own supported-memory list, then filter to that form factor here — every specification on these cards is read from the manufacturer's title, and anything the title does not state is left blank rather than inferred.

FAQ

DDR4 ECC server memory questions

What is the difference between RDIMM, UDIMM and LRDIMM?
It is where the buffering happens. A UDIMM (unbuffered) connects the address and command lines directly to the DRAM chips — the simplest arrangement, with the lowest latency and the tightest limit on how many modules a channel supports. An RDIMM (registered) inserts a register on those lines, reducing the load the memory controller drives so a channel can carry more modules and larger capacities; it is the mainstream server module. An LRDIMM (load-reduced) buffers the data lines too, which is how the highest per-channel capacities are reached, at some latency cost. A platform supports one type, stated in its documentation, and the three cannot be mixed in a system.
Can I mix ECC and non-ECC memory in a server?
No. A platform that requires ECC generally refuses to post with a non-ECC module present, and a platform that merely tolerates ECC will fall back to running everything without error correction — which quietly removes the feature you paid for. Registered and unbuffered modules cannot be mixed either, regardless of ECC. Keep one system on one module type, and when you expand, match the existing modules' type, rank structure and rated speed rather than only their capacity.
Will DDR4 ECC memory work in a desktop?
Sometimes, and only for unbuffered ECC. An ECC UDIMM runs with error correction active only when the processor and the board both support it, which is common on workstation-class platforms and rare on consumer ones; in a board without support it will typically run as ordinary memory or not post at all. Registered and load-reduced modules will not work in a consumer board in any configuration — the register requires platform support that is simply not present. Check both the CPU and the motherboard specifications before ordering.
Why is my memory running slower than its rated speed?
Usually population rules rather than a fault. A platform clocks memory at the slowest rate shared by every installed module and by the processor's memory controller, so one slower module pulls a whole set down, and a processor with a lower supported rate caps everything regardless of what the modules are rated at. Beyond that, filling two slots per channel commonly forces a lower speed than one, and higher-rank modules can do the same. The server's memory-population guide states the supported speed for each arrangement.
Can you confirm TAA compliance on memory modules?
Every line is screened for TAA country-of-origin and NDAA §889 status before checkout. Memory is a case where this needs saying explicitly with the request: country of origin can differ between production runs of the same part number, so state the requirement when you ask for the quote and we confirm it against the exact modules being allocated to your order rather than against a catalog record.

Populating servers, or a fleet of them?

Send the server model or the memory part number with quantities. We confirm the module type the platform takes, check stock across the quantity and quote the whole population rather than a single module.

DDR4 ECC Server Memory Modules — Uniqcli