If the label says 5600 and the BIOS shows 4400, the memory is not faulty. A server trains DDR5 RDIMMs to the speed its memory controller can hold with the population you installed. Two DIMMs per channel or one slower stick in the mix pulls the whole system down. Here is how each happens.
The BIOS Trains Memory Speed, Not the Label
A DDR5 RDIMM marked 5600 is a maximum, not a setting. At every boot the memory controller on the CPU tests the installed population and picks the highest speed that holds across every populated channel. Current Xeon Scalable and EPYC parts expose eight memory channels per socket, and each channel can carry one or two DIMMs depending on the board. The number you finally see on the BIOS memory page is therefore a property of the whole configuration — controller, board, slot population, rank count — and not of any single stick.
That is why two servers loaded with the same 5600 RDIMMs can boot at different speeds. Nothing is broken in either box; the populations are simply different.
Two DIMMs per Channel: the Derating Rule
Every memory channel has up to two slots. One DIMM per channel, usually written 1DPC, is a light electrical load, and the controller can hold the full rated speed. Two DIMMs per channel, 2DPC, doubles the load on the same signal lines. Signal integrity gets harder to hold at speed, so every platform publishes a derated maximum for that case. On many current platforms the step is one or two notches — for example, 4800 MT/s at 1DPC becomes 4400 MT/s at 2DPC. The exact values live in the platform’s memory population table, and that table is the document to read before you buy capacity.
| Population per CPU | Slots used | What the controller trains |
|---|---|---|
| 8 DIMMs, one per channel (1DPC) | 8 of 16 | Full rated speed — the number on the label |
| 16 DIMMs, two per channel (2DPC) | 16 of 16 | One or two speed steps lower, e.g. 4800 to 4400 on many platforms |
| 16 DIMMs including quad-rank modules | 16 of 16 | Often derated further than the standard 2DPC value |
| 6 DIMMs placed outside the manual’s slot order | 6 of 16 | Empty channels carry no traffic, and populated channels may still train low |
| One slower or higher-rank stick anywhere | any | The whole system runs at the slowest common trained speed |
The last two rows are the mistakes that cost the most. An uneven population not only leaves bandwidth on empty channels, it can also train lower on the channels that are populated when the board’s rules are not followed. Slot order matters: fill the slots in the sequence the server manual gives, and keep the layout symmetric between sockets.
One Slow Stick Drags Down the Whole Server
Speed training is a lowest-common-denominator exercise. Mix 4800 and 5600 RDIMMs and every DIMM runs at 4800; the faster sticks never run faster. Rank count behaves the same way: a quad-rank stick sharing a channel with dual-rank sticks forces more conservative settings for that channel. If you must mix, keep identical part numbers within each channel and accept that the slowest member sets the pace for everything. We covered the speed-mixing case in more depth in DDR5 4800 vs 5600 RDIMM.
How to Check What Your Server Actually Trains
Three places tell you the truth. The BIOS or UEFI setup memory page shows the trained speed per DIMM. The out-of-band controller — iDRAC9, iLO, XCC or any Redfish inventory — lists each slot with its rated and configured speed. On Linux, dmidecode -t memory prints both a Speed field (what the module supports) and a Configured Memory Speed field (what it is running), and the gap between those two lines is the whole story of this article.
Read the slot map too. Sixteen DIMMs across two sockets means two per channel, so a 2DPC derate is expected and not a defect. Eight per socket booting at the same lower speed as a sixteen-DIMM system is a sign something did not train — check for a mismatched stick.
When the Slower Speed Does Not Matter
Bandwidth-bound work — HPC jobs, AI training, in-memory databases — feels every notch of derate, and the fix is to fill channels at one DIMM per channel before adding a second DIMM anywhere. Capacity-bound work — file serving, web front ends, most virtualization hosts — barely notices: 768 GB at 4400 does more useful work than 384 GB at 4800.
So plan in this order: fill all channels first, add second DIMMs only when you need the capacity, and match part numbers when you do. If you are configuring a DDR5 platform, the DDR5 RDIMM 4800–5600 we carry covers 16 GB and 32 GB modules at all three speeds, and a listing such as the Dell PowerEdge R760 8SFF with DDR5-5600 shows its memory layout up front so you can see the population before you buy. The PowerEdge R650xs 1U is the crossover case: depending on which CPU generation it is configured with, it takes DDR4 or DDR5, so confirm the CPU before ordering memory.
On the DDR4 side the same rules apply with different numbers — the derate table differs by platform, and DDR4 and DDR5 are not interchangeable because the key notch physically blocks the wrong module. For older platforms, the DDR4 RDIMM 2133–3200 covers 16 to 64 GB, and a machine like the PowerEdge R7525 with dual EPYC 7543 carries sixteen DIMM slots per socket on DDR4.
If you are planning a population around a specific board, send us the model and the DIMM count you have in mind and we will check it against that platform’s population table before you order. For a worked example of planning around a barebone, see DL380 Gen10: what else you need to boot it.
Frequently Asked Questions
Why are my DDR5 RDIMMs running slower than rated?
Two causes cover almost every case. First, a mixed set runs at the speed of the slowest module, so adding a fast module to slow ones gains you nothing. Second, the platform applies a speed-versus-capacity curve — a larger-capacity module is often rated lower, and the platform will honour that rating. Both are normal behaviour rather than a defect.
How do I check the actual speed my memory is running at?
Read it from the BIOS or the server management interface, which report the running data rate alongside the populated channels. Compare that against the module part number and its rating. If the channels are only partly populated, that is the other thing to check, because an unbalanced set can leave bandwidth unused.
Does the platform always run memory at its rated speed?
No. The CPU memory controller and the platform’s population rules together determine the real rate. A 5600-rated module may run at 4800 in a configuration that uses fewer channels or larger capacities. This is why the supported configuration list for your specific model matters more than the module’s own rating.
Will a firmware update fix slow DDR5?
Sometimes. Firmware releases frequently refine memory population behaviour and the speed-versus-capacity curve. It is a low-cost thing to try, and we re-flash during testing so used units ship on a level we have verified rather than whatever the previous owner left behind.
Should I replace slow memory or live with it?
If the workload is memory-bandwidth-bound, aligning the set to a single part number is usually the cheapest fix and often cheaper than replacing modules outright. If the platform genuinely cannot reach the rated speed for your capacity, then no hardware change helps and you should plan around the actual rate. Tell us the platform and the modules and we will tell you which situation you are in.
How do you verify used DDR5 memory?
We stress-test the full set together on our own test line, which surfaces uncorrectable errors and marginal modules that pass a quick boot. We confirm the part number, speed rating and capacity per module, and we state the exact configuration that was tested. Ask for the test notes with the quote.
The DDR5 RDIMM models listed on zhongxindata.hk are quoted per part number, and availability is confirmed per model and per quantity.
Send the part number and quantity and we will check availability. WhatsApp +86 13068616914 · zhongxindata.hk
The two explanations that cover almost every case
- A mixed set runs at the speed of the slowest module, which is the single most common reason
- The platform applies a speed-versus-capacity curve, so a larger module can be rated lower
- Neither is a fault, and both are visible in the BIOS or the management console
Checking what your memory is actually doing
- Read the running speed and the populated channels from the BIOS or the server management interface
- Confirm the module part number and rating against what is physically installed
- Compare the population pattern against the platform’s supported configuration list
Fixing it and knowing when not to
- Aligning the module set by part number is usually the fix, and matching sets are cheaper than replacing
- Firmware updates sometimes adjust the speed-versus-capacity behaviour
- If the platform does not support the rated speed for that capacity, no change will reach it
Related reading
- DDR5 enterprise RDIMM — the memory covered
- DDR5-4800 vs 5600 — the related selection question
- Server memory guide — memory fundamentals
- Server memory — category hub
Frequently Asked Questions
Why are my DDR5 RDIMMs running slower than rated?
Two causes cover almost every case. First, a mixed set runs at the speed of the slowest module, so adding a fast module to slow ones gains you nothing. Second, the platform applies a speed-versus-capacity curve — a larger-capacity module is often rated lower, and the platform will honour that rating. Both are normal behaviour rather than a defect.
How do I check the actual speed my memory is running at?
Read it from the BIOS or the server management interface, which report the running data rate alongside the populated channels. Compare that against the module part number and its rating. If the channels are only partly populated, that is the other thing to check, because an unbalanced set can leave bandwidth unused.
Does the platform always run memory at its rated speed?
No. The CPU memory controller and the platform’s population rules together determine the real rate. A 5600-rated module may run at 4800 in a configuration that uses fewer channels or larger capacities. This is why the supported configuration list for your specific model matters more than the module’s own rating.
Will a firmware update fix slow DDR5?
Sometimes. Firmware releases frequently refine memory population behaviour and the speed-versus-capacity curve. It is a low-cost thing to try, and we re-flash during testing so used units ship on a level we have verified rather than whatever the previous owner left behind.
Should I replace slow memory or live with it?
If the workload is memory-bandwidth-bound, aligning the set to a single part number is usually the cheapest fix and often cheaper than replacing modules outright. If the platform genuinely cannot reach the rated speed for your capacity, then no hardware change helps and you should plan around the actual rate. Tell us the platform and the modules and we will tell you which situation you are in.
How do you verify used DDR5 memory?
We stress-test the full set together on our own test line, which surfaces uncorrectable errors and marginal modules that pass a quick boot. We confirm the part number, speed rating and capacity per module, and we state the exact configuration that was tested. Ask for the test notes with the quote.