Can You Mix RAM With Different Speeds?

A 3200 stick and a 3600 stick can share a system — they will both slow down to match. That sounds like a bad deal for the faster stick, and in one sense it is: you paid for speed you will not use. But it is also safe, stable, and often the cheapest way to add memory. What else changes when speeds mix, and how to make it stable.

Updated 11 Nov 2026 · 8 min read

What the memory controller does

There is one memory clock and one set of primary timings per channel, so the controller negotiates down to what every stick can do. Voltage also unifies: mixing a 1.35V XMP kit with a 1.2V JEDEC stick usually means running both at JEDEC settings, because enabling XMP would overvolt the plain stick’s profile.

Timings follow the same rule, and this is the part people miss. A DDR4-3600 CL18 stick and a DDR4-3200 CL16 stick do not just meet at 3200 MT/s — they also meet at timings the 3200 stick can handle at 3200, which may be looser than CL16, closer to CL18. The result is usually 3200 MT/s at CL18 or looser, not 3200 MT/s at CL16.

How mixing speeds works, in plain English

Every RAM stick carries a small chip called SPD that lists the speeds and timings it supports. On first boot, the board reads the SPD from every stick, works out the fastest combination they all share, and starts there. That shared starting point is the JEDEC speed — usually 2133, 2400 or 2666 MT/s on DDR4, or 4800 MT/s on DDR5 — which is why a mixed system often boots at a surprisingly low speed even though both sticks are rated faster.

XMP and EXPO complicate this, because they are not part of the JEDEC standard. An XMP profile is a tested overclock stored on one stick, and the board can only load it if every stick in the system can run it. If one stick has an XMP profile for 3600 MT/s and the other has no XMP profile at all, enabling XMP will usually fail, or the board will fall back to JEDEC speeds. That is normal and safe. See our XMP vs EXPO guide for how the profiles work.

Making a mixed setup stable

1. Install both sticks, enter BIOS, and load defaults first. Do not enable XMP on the first boot — let the board find the safe JEDEC speed and confirm both sticks are detected.
2. Try enabling XMP — if it boots and passes a memory test, keep it. On some boards this works, if both sticks are from the same maker and the slower stick’s profile matches.
3. If not, set the speed manually to the slower kit’s rating and timings to the looser kit’s values. For a 3600 CL18 stick and a 3200 CL16 stick, that means 3200 MT/s at CL18 to start.
4. Run MemTest86 or Windows Memory Diagnostic overnight; even one error means loosen timings or drop 200 MT/s. One error in an overnight run is not “close enough” — it is an unstable system.

What you actually lose: the numbers

Latency in nanoseconds is CL × 2000 / MT/s. A DDR4-3200 CL16 stick responds in about 10 ns, and a DDR4-3600 CL18 stick also responds in about 10 ns. When they mix at 3200 MT/s and CL18, the latency becomes 18 × 2000 / 3200, or about 11.3 ns. You have lost about 1.3 ns of responsiveness and the bandwidth of the faster kit, which is a small but real cost.

Bandwidth is the bigger loss, and it is easy to count. Dual-channel DDR4-3200 gives about 51.2 GB/s and DDR4-3600 gives about 57.6 GB/s, so mixing down from 3600 to 3200 costs about 6.4 GB/s, or roughly 11% of the bandwidth. Going from 3200 down to 2666 costs about 17%. In everyday use that is not something you will feel. In gaming, the difference between 3200 and 3600 is a few percent in frame rate at most. The capacity you gain by adding the second stick is worth far more than the bandwidth you lose.

The exception: DDR5

DDR5’s on-stick PMICs make mixing kits noticeably fussier, and board training times balloon. Mixing DDR5 speeds works, but if you value your weekends, a single matched 2-stick kit is the sane choice on AM5 and LGA1700.

DDR5 training is already slow with a matched kit — first boots can take 30 to 90 seconds — and mixing speeds makes the board test far more combinations. A mixed DDR5 pair may train for several minutes and still fall back to the 4800 MT/s JEDEC speed, throwing away the reason you bought a 6000 MT/s kit.

If you must mix DDR5 speeds, keep the speeds close — 5600 with 6000, not 4800 with 7200 — buy both sticks from the same maker if you can, update the BIOS first, and plan to run at JEDEC speeds or at the slower kit’s rating. DDR5 instability often shows up as random reboots or failed training on a cold boot, not as an immediate blue screen. Our DDR5 compatibility guide covers the platform limits in full.

When the answer changes

On DDR4, mixing speeds is a routine, low-risk upgrade. The controller falls back cleanly, the penalty is one step of speed and slightly looser timings, and most mixed pairs are stable at JEDEC speeds or at the slower kit’s rating. On DDR5, the same mix is possible but far less rewarding: training takes longer, XMP/EXPO is much less likely to hold, and the fallback to 4800 MT/s wastes more performance.

Laptop BIOSes rarely offer XMP or manual speed settings, so a laptop with a DDR4-2666 stick and a new DDR4-3200 stick will simply run both at 2666 MT/s. Check the laptop’s maximum speed before buying, or the faster stick’s premium is wasted.

Intel boards are generally more tolerant of mixed XMP profiles. AMD boards, especially AM5, are pickier: a mixed pair that works on Intel may fall back to JEDEC speeds on AMD. The rule is the same on both — slowest speed, loosest timings.

Verdict

Mixing RAM speeds is safe, and for most people it is good value. You lose the faster stick’s advantage — one step of speed and a little latency — and you gain the capacity of a second stick, which is worth far more in practice than the bandwidth you gave up. Mix when the speeds are one step apart and you need memory now. Replace the slow stick when the gap is two steps or more, and buy a matched kit when you are on DDR5, building from scratch, or chasing high-refresh-rate gaming performance.

Frequently asked questions