Picking server memory looks simple until you're staring at a listing with a dozen specs you need to match exactly. Get one of them wrong and you've either bought something that won't boot in your system, or something that works but leaves performance on the table. Here's what actually matters, in the order it matters.
1. Start With What Your Server Supports, Not What's On Sale
Before you look at a single listing, check your server's documentation for its supported memory type, speed, and maximum capacity per slot. This is non-negotiable — a great deal on memory your server can't use isn't a deal at all. Dell, HPE, Lenovo, and Cisco all publish compatibility lists (often called QVLs — qualified vendor lists) for their server models. Check yours first.
2. Match Capacity to What You Actually Need
More isn't automatically better — buying capacity you won't use for two years ties up budget you might need elsewhere. A reasonable approach:
- Database or virtualization workloads: size memory to comfortably hold your active data set, with headroom for growth.
- General application servers: match or slightly exceed current usage; check your monitoring tools for actual peak usage before guessing.
- New builds: plan for 12-18 months of growth, not just day-one needs.
3. Get the Speed Right — But Know Your Ceiling
Every server has a maximum memory speed it will actually run at, usually tied to the CPU and how many DIMM slots per channel are populated. Buying faster memory than your server supports doesn't hurt anything — it just runs at the server's ceiling — but it does mean you paid for speed you're not using. Check your platform's spec sheet for the actual supported speed before choosing a module.
4. RDIMM vs LRDIMM
Most servers use RDIMM (registered DIMM) — it's the standard choice for the vast majority of builds, offering a solid balance of speed, capacity, and cost. LRDIMM (load-reduced DIMM) exists for one specific case: when you need very high total memory capacity per server and populating every slot with high-capacity RDIMMs would overload the memory controller. If you're not chasing extreme capacity, RDIMM is almost certainly what you need.
5. Don't Skip ECC
ECC (error-correcting code) memory catches and corrects small memory errors before they become data corruption or a crash. For any production server — especially one running a database — this isn't optional. Non-ECC memory has no business in a server environment, regardless of price.
6. Match Rank and Voltage to Your Existing Modules
If you're adding memory to a server that already has RAM installed, matching rank (single, dual, or quad) and voltage to your existing modules matters. Mixing mismatched ranks can work in some configurations and cause instability in others, depending on the platform — when in doubt, match exactly what's already installed.
7. Buy From a Source That Shows You the Real Spec
The last step isn't about the memory itself — it's about who you buy it from. A listing that gives you the exact manufacturer part number, full spec table, and clear compatibility notes saves you from guessing, returning, or re-ordering. If a listing can't tell you the part number, rank, and validated platforms up front, that's a sign to look elsewhere.
The Quick Version
- Check your server's supported memory type and max speed first.
- Size capacity to your actual workload, with reasonable headroom.
- Don't overpay for speed your server can't use.
- Default to RDIMM unless you specifically need LRDIMM's capacity advantage.
- Never skip ECC on a production server.
- Match rank and voltage to existing installed memory.
- Buy from a listing that gives you the full spec, not just a price.
Get these seven right, and you'll order once instead of twice.