16GB DDR4 PC4-2400T ECC RDIMM
16 GB DDR4 ECC — 2400 MT/s
- Gen9
- Gen10
Server memory is always ECC: it corrects a single-bit error before it ever reaches your application. On a workstation that is a nice extra; on a server running for months without a reboot it is the difference between “a bit flipped” and “the database is corrupt”.
7 parts
16 GB DDR4 ECC — 2400 MT/s
32 GB DDR4 ECC — 2400 MT/s
16 GB DDR4 ECC — 2666 MT/s
64 GB DDR4 ECC — 2666 MT/s
32 GB DDR4 ECC — 2933 MT/s
64 GB DDR4 ECC — 2933 MT/s
32 GB DDR4 ECC — 3200 MT/s
Prices are based on the global market average in euro.
RDIMM is the default: lower latency, lower price. LRDIMM starts to matter when you want to fill all 24 slots with high-capacity modules; its extra buffer lets total capacity go higher at the cost of slightly more latency. For most servers, which populate half the slots, RDIMM is both cheaper and faster.
This is the detail most often missed, and the one that wastes the most performance. A Xeon has four to six memory channels, each contributing its own independent bandwidth. Six modules spread across six channels deliver close to twice the bandwidth of the same six modules on one channel. The correct order is in that server model’s memory-population guide and is also printed on the board itself.
Mix a 2400 and a 2666 module and everything runs at 2400. No error is shown; you simply lose the speed you paid for. Actual speed also depends on the processor: Gen9 tops out at 2400 even when the module is rated 2666.
No. ProLiant servers accept only registered modules (RDIMM or LRDIMM) with ECC. Desktop UDIMM will not boot the server even if it physically fits the slot; iLO reports unsupported memory.
In practice memory runs out before CPU does, not the other way round. Budget 4 to 8 GB per ordinary virtual machine and reserve about 16 GB for the hypervisor itself. A server with 256 GB comfortably holds 30 to 40 mid-sized VMs — and the same server with 64 GB cannot, however many cores it has.
Technically yes, provided type (RDIMM/LRDIMM), rank and speed match; the chip brand itself does not matter. It is still worth keeping modules within a single channel identical, because diagnosing mixed memory once ECC errors appear is considerably harder.
Online — here to help