What's the Difference Between RDIMM and LRDIMM Server Memory?
RDIMM vs LRDIMM explained: how registered and load-reduced memory differ, when each one wins, latency tradeoffs, and capacity ceilings for modern Xeon and EPYC servers.
What's the Difference Between RDIMM and LRDIMM Server Memory?
RDIMM (Registered DIMM) uses a single register chip to buffer command/address signals between the memory controller and DRAM, reducing electrical load on the CPU. LRDIMM (Load-Reduced DIMM) buffers both command/address AND data through an iMB (Isolation Memory Buffer), allowing much higher capacities per channel — 128GB or 256GB per module vs RDIMM's 64GB ceiling — at the cost of 1-2 cycles of added latency. Choose RDIMM for performance, LRDIMM for capacity.
How They Differ Electrically
A DDR4 or DDR5 memory channel carries 64 data lines plus 8 ECC bits plus command/address signals. As you add DIMMs to the same channel, electrical load increases, which forces the controller to lower the clock speed and increases the chance of signal-integrity errors.
RDIMM solves this for command/address signals only:
- A register chip on the DIMM PCB buffers the address/command before passing them to DRAM
- Data path is direct from controller to DRAM
- Slight latency penalty (~1 clock cycle for register)
- Maximum 2 DIMMs per channel at full DDR4-3200 / DDR5-4800 speed
LRDIMM buffers everything:
- An iMB (isolation memory buffer) chip on the DIMM PCB buffers both address/command AND the entire data path
- DRAM chips are isolated from the channel electrically
- Larger latency penalty (~2-3 clock cycles)
- Up to 3 DIMMs per channel at high speeds — and crucially, much denser DRAM stacks per module
Capacity Ceilings — Where LRDIMM Wins
The maximum capacity per module differs significantly:
| Generation | Max RDIMM Capacity | Max LRDIMM Capacity |
|---|---|---|
| DDR4 (Xeon Scalable Gen 1-2) | 64 GB | 128 GB |
| DDR4 (Xeon Scalable Gen 3, EPYC Rome/Milan) | 128 GB | 256 GB |
| DDR5 (Xeon Sapphire Rapids, EPYC Genoa) | 128 GB | 512 GB (with 3DS) |
For a Dell R740 with 24 DIMM slots, the math:
- 24 × 64GB RDIMM = 1.5 TB total
- 24 × 128GB LRDIMM = 3 TB total
If you need more than 1.5 TB in an R740, you must use LRDIMM.
Performance Comparison
RDIMM is consistently faster for typical workloads. The latency difference is small but measurable:
| Workload | RDIMM | LRDIMM | Performance Loss with LRDIMM |
|---|---|---|---|
| HPC compute (HPL/Linpack) | 2.85 TFLOPS | 2.79 TFLOPS | -2% |
| Database OLTP (TPC-C) | 145k TPS | 142k TPS | -2% |
| VM density (vmware vMark) | 95 VMs/host | 130 VMs/host | +37% (LRDIMM wins via capacity) |
| AI/ML inference | similar | similar | <1% |
For most workloads, RDIMM is slightly faster. For VM density, LRDIMM lets you pack more guests per host.
Cost Comparison (2026 prices)
Mid-tier capacity:
- 16GB DDR4-3200 RDIMM (2Rx8): $35-50 refurbished, $90-110 new
- 32GB DDR4-3200 RDIMM (2Rx4): $60-90 refurbished, $180-220 new
- 64GB DDR4-3200 RDIMM (4Rx4): $180-260 refurbished, $500-700 new
- 64GB DDR4-3200 LRDIMM (4Rx4): $150-220 refurbished, $480-680 new
- 128GB DDR4-3200 LRDIMM (4DRx4 3DS): $480-700 refurbished, $1,500-2,200 new
- 256GB DDR4-3200 LRDIMM (8Rx4 3DS): $1,800-2,800 refurbished, $5,000-7,000 new
LRDIMM is generally cheaper at 64GB and 128GB than the equivalent RDIMM because there's no equivalent — RDIMM tops out at 64GB on most platforms.
Quick Decision Tree
- Need >1.5 TB total per CPU socket? → LRDIMM (mandatory)
- Running high-density VM hosts (50+ VMs/host)? → LRDIMM (capacity wins)
- Running HPC, databases, or latency-sensitive apps? → RDIMM (performance wins)
- Need 256-512 GB total in a 12-DIMM socket? → Either works; RDIMM cheaper at this tier
- Mixed workload, undecided? → RDIMM (default for most enterprise)
Common Mistakes
Mixing RDIMM and LRDIMM in the same CPU's channels — most systems will refuse to POST. If they POST, they downclock everything to the slowest speed and run unstably.
Using LRDIMM when capacity per channel is low — paying the latency penalty without using the capacity benefit. If you only need 256GB per CPU, RDIMM is faster AND cheaper.
Buying 256GB LRDIMM for a 1st-gen Xeon Scalable system — 1st-gen tops out at 128GB LRDIMM. Always verify your CPU generation's max DIMM capacity before buying high-density modules.
Frequently Asked Questions
Can I add LRDIMM to a system that has RDIMM installed? No. Most platforms require all DIMMs on a CPU to be the same type. Some allow per-channel mixing but it's not recommended — performance degrades to the lowest common denominator.
Which has better signal integrity at full-population? LRDIMM by a clear margin. Filling all 12 slots per CPU at DDR4-3200 generally requires LRDIMM — RDIMM either downclocks or runs unstably.
Is DDR5 RDIMM faster than DDR4 LRDIMM at the same nominal speed? Yes — DDR5 has fundamentally lower per-bit latency and twice the burst length. A DDR5-4800 RDIMM beats a DDR4-3200 LRDIMM by ~50% on bandwidth-bound workloads.
Get the Right Memory
We stock both RDIMM and LRDIMM for every major server platform — HPE ProLiant, Dell PowerEdge, Cisco UCS, Lenovo ThinkSystem, Supermicro. Browse server memory or compare ECC vs Non-ECC.
Need spec-verified memory for a specific server model? Email sales@prodisknetwork.com with your service tag — we'll match the validated module.