Each stick passes alone, the pair refuses to boot or crashes. This is a training and matching problem — here is the fix path.
Updated 11 Nov 2026 · 7 min read
Quick answer
When sticks work individually but not together, the pair cannot agree on shared settings: mismatched kits, an XMP/EXPO profile one stick cannot do, four-stick DDR5 training limits, or sticks in the wrong slots. Load BIOS defaults, use A2/B2, disable XMP/EXPO, and if they boot together at JEDEC speed, raise speed manually in small steps.
Contents
Why pairs fail when singles pass
Two sticks must share one clock, one voltage and one timing set per channel — and dual-channel interleaving adds its own training. A stick that is marginal at 3600 passes alone (single channel, lighter load) and fails when the controller drives two. Four sticks double the electrical load again, which is why 4×DDR5 at rated XMP speed is the hardest configuration on current platforms. The four usual reasons are an XMP/EXPO profile one stick cannot run, mismatched timings the board cannot reconcile, the heavier electrical load of driving two sticks, and sticks in the wrong slots. A faulty stick fails alone in slot A2 — see one RAM stick not working for that test.
The fix sequence
- Slots A2 + B2. Confirm, don’t assume.
- Clear CMOS / load defaults. Boot the pair at JEDEC (DDR4 2133–2666, DDR5 4800).
- Stable? Enable XMP/EXPO. Unstable? Set speed manually 200–400 MT/s below the profile and retest.
- Still unstable? Raise DRAM voltage one small step (e.g. 1.35V → 1.37V DDR4) or loosen primary timings by +2 on CAS.
- Test with MemTest86, both sticks, 4 passes, after every change.
- Mixed kits that refuse all of the above: run them at the slower kit’s JEDEC profile, or replace with a matched kit — some IC combinations simply never train together on DDR5.
Work through these in order and stop as soon as the pair is stable. Step two is the most important: boot the pair at stock JEDEC speeds with XMP off. If the pair boots and runs stably at JEDEC, you have proved both sticks are healthy and the problem is the profile, not the hardware. From there, step three is a controlled climb: enable XMP and test; if it fails, set the speed manually a couple of hundred MT/s below the profile and test again.
| Step | What you change | What success looks like |
|---|---|---|
| Slots to A2/B2 | Physical placement | Pair boots at JEDEC speeds |
| Load BIOS defaults, XMP off | Settings only | Pair boots and passes MemTest86 |
| Enable XMP/EXPO | Profile on | Pair boots at rated speed, 4 clean passes |
| Manual speed 200–400 MT/s lower | Speed only | Stable at slightly lower speed |
| Voltage +0.02V or CAS +2 | One small step | Stable at or near rated speed |
Four sticks and DDR5: the hardest configuration
Four sticks are harder than two, and four DDR5 sticks are the hardest configuration on current platforms. Each extra stick adds electrical load on the memory bus, and the signal degrades with every stick, so the maximum stable speed falls as you add sticks. A board that runs two DDR5 sticks happily at 6000 MT/s may only manage four sticks at 5200 or 4800, and that is normal — it is a limit of the platform, not a fault in your RAM. Update the BIOS before tuning — makers steadily improve four-stick training — and buy four sticks as a single kit, not as two pairs. For the capacity question, see is 32GB RAM enough and is 64GB RAM overkill.
Mixed kits: sizes, speeds, and brands
Mixing kits is where most pairing problems start. Mixing sizes is usually the most forgiving: an 8GB stick and a 16GB stick can run together in dual-channel mode for the matched portion, with the extra 8GB running in single-channel — see can you mix 8GB and 16GB RAM for the details. Mixing speeds is more likely to cause trouble, because the faster stick must slow down to the slower one, and the XMP profile of the faster kit may not be achievable at all — see can you mix RAM with different speeds. Mixing brands is fine in principle, but it increases the chance of different chips inside — if the pair boots at JEDEC speeds, run it there or tune up manually.
Common mistakes that make it worse
The most common mistake is enabling XMP first and concluding the RAM is broken when the pair will not boot. Always test at JEDEC speeds first — a pair that boots at stock is healthy, and the profile is the problem. The second is buying a second kit to “match” the first and expecting it to behave like a matched kit: two kits bought months apart are not a matched kit, even with the same model number, because the chips inside may differ.
Buying your way out
If this is a new matched kit failing together at stock on the latest BIOS, return it — matched kits are sold as tested pairs, and yours isn’t behaving like one. If it is two separately bought sticks, the realistic fix is a proper 2-stick kit; resell the singles. See mixing sizes and mixing speeds for what usually does work. For help picking the replacement, see our RAM compatibility checker guide and XMP vs EXPO.
Never raise DRAM voltage above the kit’s rated maximum, and change only one setting at a time so you know what fixed it. Back up important files before memory testing — an unstable pair can corrupt data while the PC runs. Power fully off and unplug the cable before swapping sticks, ground yourself on bare metal, and never force a stick that does not seat easily.
Verdict: tune it or replace it?
Tune it if the pair boots together at JEDEC speeds: raise the speed in small steps, test with MemTest86 after each change, and settle at the fastest speed that passes four clean passes — even if that is a few hundred MT/s below the rated speed, the real-world difference is a few percent. Replace it if the pair will not boot together at JEDEC speeds in the right slots on the latest BIOS, because those sticks will never agree, and no setting will change that.
Frequently asked questions
They boot together at 4800 but not at 6000 EXPO. Is something broken?
Probably not — that is a training/voltage margin issue, very common with 4 sticks or early BIOS versions. Update the BIOS, try 5800, or accept 5600: the real-world difference is a few percent.
Does stick order within A2/B2 matter?
With mismatched sticks, sometimes. Put the weaker/slower stick in A2 — training typically starts from that slot’s SPD.
Can I use two different brands together?
Yes, if they boot together at JEDEC speeds and pass MemTest86. Different brands are fine in principle — the risk is different memory chips inside, which is why testing matters more than the label on the stick.
Is it safe to run RAM below its rated speed permanently?
Yes, completely. Running below the rated speed is easier on the memory controller and the sticks, and it is what JEDEC speeds are for. You lose a small amount of performance and gain stability.
Why did the pair work for months and then start crashing?
A BIOS update can change training behaviour, a driver update can change load patterns, or a stick can become marginal with age and heat. Retest at stock speeds first — if the pair is stable at JEDEC, the profile has become unstable, not the hardware.