Quick answer: Pull every stick, test one at a time in slot A2, and if that boots but A2+the second stick in B2 doesn't, the problem is the slot, the CPU memory controller, or bent socket pins — not the RAM itself.
What's actually happening here is that most 'RAM errors' people chase aren't RAM. They're the channel. A modern board runs dual channel across specific slot pairs (usually A2+B2 on a four-slot board, counting from the CPU). If one slot's traces are damaged, or a single LGA1700/AM5 pin under the socket is bent, one channel goes dark. The system either won't POST at all, throws a DRAM debug LED, or boots into single-channel mode with half the bandwidth you paid for. The stick is fine. The path to it isn't.
I've seen this most often after a CPU cooler swap. You torque an NH-D15 down, the retention bracket flexes the board, and a pin that was borderline makes contact with nothing. Same story after a case move where the DIMMs unseat by a millimeter — DDR4 and DDR5 need a genuinely firm click on both latches, and 'it looks seated' is not the same as seated.
Numbered fix steps
- Clear CMOS and boot minimal. Pull the battery or short the CLR_CMOS jumper with the PSU unplugged, then boot with one stick in A2 only. No GPU if you have iGPU, no drives. If it POSTs, you've isolated it to memory or a slot.
- Test each stick solo in A2. One at a time. If stick #1 works and stick #2 doesn't in the same slot, you have a dead stick. If both work solo, the issue is dual-channel config or the second slot.
- Now test A2 + B2 with matched sticks. This is where it gets interesting. If solo works but paired fails, swap to A1+B1. If A1+B1 boots and A2+B2 doesn't, you've got a damaged B2 slot or a channel-specific CPU pin problem.
- Reseat with real force. Open both latches fully, line up the notch, and press until both latches snap shut on their own. If you're pushing with your thumbs and the board is flexing more than the stick is moving, you're not seated.
- Drop XMP/EXPO and test at JEDEC. Disable the profile in BIOS and run at 2133/4800 (DDR4/DDR5 default). If everything is stable at stock and crashes with XMP on, you have a memory controller or voltage problem, not a hardware fault. More on this below.
- Check for bent CPU pins. Pull the CPU and look at the socket with a bright light and your phone camera zoomed in. On AM5, look at the CPU's gold pads. A single bent pin in the memory address bus kills one channel. This is the fix nobody wants to hear, but it's the actual cause in maybe 15% of no-POST-RAM cases I've dealt with.
- Run MemTest86 for two full passes. Not one. Two. Errors in the first 10 minutes mean a bad stick. Errors after 40 minutes usually mean unstable voltages or a marginal slot.
If the main fix doesn't work
Boot with one stick, flash the latest BIOS from the vendor, and re-test. DDR5 boards especially — early AM5 and Z690 firmware had memory training bugs that made perfectly good kits fail to POST. The 2023-era AGESA updates fixed a lot of this. If your board is on a launch BIOS, that's your problem.
Next, verify against the QVL. Manufacturers publish a Qualified Vendor List per board, and a kit that isn't on it can still work — but a kit that isn't on it and fails is on you, not the board. Dual-rank DDR4 kits are the usual offender on budget B660/B550 boards.
If XMP keeps failing, bump SoC/VDDIO voltage manually. On AM4, 1.10V SoC is usually enough for DDR4-3600. On AM5, VDDIO around 1.35–1.40V for DDR5-6000 CL30. Don't go past 1.45V without knowing what you're doing — you'll degrade the IMC.
Still nothing? RMA the board. If a slot works one boot and not the next, the slot is dying, and no amount of reseating fixes a cracked trace.
Beep codes are still useful: three short beeps on AMI BIOS = memory failure. Continuous beeps = no memory detected at all. On boards with a Q-Code display, 55 (MSI/ASRock) or 53 (some Gigabyte) points straight at memory init.
Prevention
Mount your cooler with a torque driver or by hand until snug — don't crank it. Both Intel LGA1700 and AM5 have documented flex issues when coolers are over-tightened. Also: never move a case with the DIMMs installed if you can avoid it. Those latches don't hold under a 20kg shock load. And keep a spare known-good stick around. Two minutes swapping in a working DIMM saves hours of guessing.
# Linux, after you're booted and stable — confirm both channels are live
sudo dmidecode -t memory | grep -E 'Locator|Size|Speed'You want to see populated entries in both ChannelA and ChannelB (or DIMM_A2 and DIMM_B2). If one shows 'No Module Installed' when a stick is physically there, the slot or the CPU memory controller is your culprit. That's the tell.