Mixing RAM kits on Ryzen: reuse 48 GB or buy a matched kit?

Four sticks, two kits, 48 GB at 3200 MT/s. Mixing RAM worked on our Ryzen 5 5600G bench. The useful question is whether you should reuse the DIMMs you already have or buy one matched kit and spend the evening building instead of troubleshooting.

Measured on our bench; new matched-kit specifications below come from Corsair and Kingston.

On September 26, 2026, our recorded setup had run for 3 days 13 hours under GPU and LLM load with zero machine-check or memory-error log entries. The bench notes record the module labels, firmware settings and boot timings. Pair that real-load observation with a dedicated memory test for your own configuration.

For a new build with a memory-integrity requirement, the ECC RAM buying guide compares exact ECC UDIMM choices and checks CPU, board and DIMM support before purchase. Choose ECC support at the platform level if correction is part of your requirement.

Reuse your RAM or buy one matched kit?

Reuse working memory when it gives you capacity you need. Buy one matched kit when you are starting fresh or replacing a troublesome combination. Corsair validates rated performance using the modules inside one kit and recommends against combining separate kits—even ones carrying the same speed rating (Corsair mixing guidance). Our measured mixed configuration gives you a useful example of successful reuse; a matched purchase gives you one boxed set to configure and validate.

Your situation	Decision	Why it fits
You already own compatible mixed kits and can test them	Try reuse first	Keep the capacity you have; validate the complete installed configuration
Your working setup has 48 GB and your workload needs more than 32 GB	Keep enough capacity	Replacing it with 32 GB removes 16 GB; that is a capacity cut
You are building a 32 GB DDR4 desktop from scratch	Buy one 2 × 16 GB kit	Choose the intended capacity in one matched purchase
Mixed kits produce errors or refuse your chosen profile	Test each kit separately, then replace the combination if needed	Separate a faulty module from a profile or mixing problem
You need 64 GB	Plan one compatible kit at that capacity	64 GB is 16 GB above this bench configuration; verify the exact larger kit before buying

A matched kit still needs validation in your board. Use the motherboard’s supported-memory list for the full part number and CPU, then run the Memtest86+ checklist at the settings you intend to keep. If your existing 48 GB already does the job, there is no need to replace it just to make the labels match.

Matched 32 GB DDR4 kits worth considering

Corsair Vengeance LPX CMK32GX4M2E3200C16 is our starting pick for a fresh 32 GB build. Kingston FURY Beast KF432C16BBK2/32 is the alternative when its exact part number matches your board’s supported list. Both are two-module desktop DDR4 kits; choose by platform support, not an invented speed winner.

Kit	Capacity and format	Maker-tested profile	SPD/default profile	Best fit
Corsair Vengeance LPX CMK32GX4M2E3200C16	32 GB, 2 × 16 GB, 288-pin DDR4 UDIMM	3200 MT/s, CL16, 1.35 V; XMP 2.0	2133 MHz, CL15, 1.2 V	A fresh matched 32 GB desktop build
Kingston FURY Beast KF432C16BBK2/32	32 GB, 2 × 16 GB, 288-pin DDR4 DIMM kit	3200 MT/s, CL16-20-20, 1.35 V; XMP 2.0	DDR4-2400, CL17-17-17, 1.2 V	A board-supported alternative 32 GB pair

Those profiles are from Corsair’s specification and Kingston’s datasheet. The Corsair 32 GB product is a different kit from the recorded 2 × 8 GB Corsair modules below. Keep the part numbers separate when comparing our measurements with your shopping list.

For two modules on this particular Gigabyte board, the manual recommends DDR4_1 and DDR4_2. Follow the printed labels and your board’s manual; do not copy a slot pattern from an unrelated motherboard. The Gigabyte manual lists the memory arrangement and XMP support.

Know exactly what we mixed before you copy it

Here is the combination to compare with yours. The board is a Gigabyte B450M DS3H V2 on firmware F62b (2021-09-06), with an AMD Ryzen 5 5600G (bench notes).

Part	What the source says
Corsair CMK16GX4M2B3200C16	2 x 8 GB, single-rank. Tested 3200 MT/s, CL16, 1.35 V. SPD 2133 MHz, CL15, 1.2 V (Corsair)
Samsung-labelled 16GXMP3200CL16	2 x 16 GB, dual-rank (bench notes). Identity and rank come from the recorded module labels.
Gigabyte B450M DS3H V2	4 DDR4 DIMM sockets, up to 128 GB, XMP support (Gigabyte)

Both recorded kits are labelled 3200 CL16: Corsair confirms its rating on the spec page; the Samsung-labelled rating comes from its part label. Matching ratings give you a sensible starting profile. Install the combination, read back the settings and test it as a complete configuration.

The complete two-GPU B450 build shows how these DIMMs fit alongside the CPU, storage and expansion slots; the memory result here is one part of that platform.

Turn on XMP or you’re running 2133, not 3200

Without a profile, the Corsair kit falls back to its SPD speed of 2133 MHz (Corsair). 3200 MT/s is its tested XMP speed, not its default.

With XMP enabled, all four DIMMs report a configured speed of 3200 MT/s on our test bench (bench notes). Read back the configured speed for each DIMM, not just the summary line.

Watch out: The firmware reports a configured voltage of 1.2 V (bench notes). Corsair tests this kit at 1.35 V and lists 1.2 V as its SPD voltage (Corsair). Treat that as a firmware report; verify the applied setting or measure voltage separately when investigating instability.

Use this checklist before you trust mixed kits

1. Both kits carry the same rated speed and CAS latency (ours: 3200, CL16).
2. Your board and CPU support the memory type, capacity and intended profile. The B450M DS3H V2 manual lists XMP support (Gigabyte).
3. Total capacity is within the board’s limit (128 GB here; we run 48 GB).
4. XMP is enabled, and every DIMM reads back the target speed rather than the SPD fallback.
5. You have noted the reported DRAM voltage and compared it with each kit’s tested voltage.
6. Run a full memory test pass before putting the machine to work.
7. You have run days of real load with no machine-check or memory-error entries in the kernel log.

Example (illustrative): You own a 2 x 8 GB 3200 CL16 kit and add a 2 x 16 GB 3200 CL16 kit. Items 1–3 pass on paper. If step 4 shows one pair stuck at 2133, you have your answer before you spend a day stress testing.

Judge stability by days under load, not by a clean boot

The bench ran for 3 days 13 hours continuously under GPU and LLM load. The kernel log shows zero machine-check or memory-error entries over that period (bench notes).

Boot is unremarkable too: 24.6 s to fully up, with 8.2 s spent in firmware (bench notes). To investigate training overhead, compare the firmware stage with two DIMMs and four DIMMs.

Tip: Keep the kernel log check in your routine after any firmware update or DIMM reseat. A clean result from last month does not cover a changed memory configuration.

System-RAM stability and model fit in GPU memory are different checks. The RTX 3060 Ti versus RTX 2060 comparison shows where a model spilled onto the CPU, which matters when choosing a representative inference workload for stability testing.

For your validation run, the Memtest86+ mixed-RAM testing guide shows how to record the selected tests, address range, passes and errors separately from the kernel-log observation above.

If you are choosing the processor as well as the memory, the 5600G versus 5600X platform guide separates display, PCIe and firmware requirements from this configuration’s RAM result. Repeat validation when you change the processor.

The verdict: keep useful capacity, buy a kit for a fresh start

Keep the measured 48 GB mixed-kit approach in mind when you already own the DIMMs and can validate them. For a new 32 GB DDR4 build, start with Corsair Vengeance LPX CMK32GX4M2E3200C16, with Kingston FURY Beast KF432C16BBK2/32 as another matched 32 GB option. Pick your capacity first, apply the intended profile, then test the whole installed set. Matching the box is convenient; keeping enough RAM for your workload is the real win.

Choose the board around the whole build

Reusing your DDR4 in a new AM4 build? Our B550 homelab motherboard picks compare ECC conditions, storage connectors and the slots left for a GPU or HBA.