Quick answer: The board can't complete memory training on the sticks in A2/B2, so it halts before video init. Reseat both DIMMs, clear CMOS, then boot with one stick in A2 to isolate a dead module or slot.
What's actually happening here is memory training, not a dead board. On every cold boot after a RAM change, the memory controller inside your CPU probes each DIMM, measures signal timing across the trace length to the slot, and locks in a working set of voltages and latencies. Slot A2 and B2 sit at the far end of those traces on most consumer boards, so they're the hardest to train. If the controller can't find a stable window — because a stick isn't fully seated, because the CMOS still holds an old profile, or because the CPU's integrated memory controller is marginal — it stops right there and lights the DRAM LED. No display, because the GPU never gets initialized. That beep pattern is just the board's way of saying "I tried, and I gave up before I got to you."
This shows up constantly on AM5 boards like the B650 Tomahawk and X670E Aorus with DDR5-6000 kits, and on Z690/Z790 boards with DDR4-3600 kits. A common real-world trigger: you had 2×8GB running in A1/B1, bought a 2×16GB kit, moved the new sticks to A2/B2 like the manual told you, and now the box just sits there with a red LED and a sad beep. The kit is fine. Your slot choice is fine. The board just hasn't been told to start over.
Fix steps, in order
-
Kill power and reseat both DIMMs. Pull the PSU cable, hold the case power button for 5 seconds to drain the rails, then press the clip on each slot until the stick pops out. Inspect the gold fingers for a bright, even contact patch. Reinsert by pushing straight down until both clips snap shut — not one. Half-seated RAM in A2/B2 is the single most common cause of this exact LED. The board sees the SPD chip, tries to train, and fails on one channel.
-
Clear CMOS. The board may be holding a stale memory profile from the old kit. With power unplugged, short the CLR_CMOS jumper for 10 seconds, or pull the CR2032 coin cell for 60 seconds. On boards with a BIOS Flashback button, hold it for 3 seconds with the system off — that also wipes NVRAM. Reconnect power and boot. Give it 60–90 seconds before you panic; DDR5 training on a 2×32GB kit can take that long on the first POST.
-
Boot with one stick in A2 only. If it POSTs, power down and add the second stick to B2. If the second stick kills it, swap the two sticks between slots. That tells you whether the module is bad or the slot is. This is the fastest way to stop guessing.
-
Disable XMP/EXPO and set JEDEC defaults. Enter BIOS (Del or F2) once it POSTs, load optimized defaults, save, reboot. Then try enabling XMP/EXPO. Plenty of DDR5-6000 kits won't train on a given CPU's IMC at full speed — they'll run at 5600 or 5200 just fine. That's the IMC, not the RAM. Bumping SoC voltage to 1.25V (AM5) or VCCSA to 1.20–1.25V (Intel) often gets you to spec, but don't exceed 1.30V on AM5 SoC or you'll cook the chip.
-
Update BIOS. AGESA 1.0.0.7c and ComboAM5PI 1.1.0.0 fixed a mountain of DDR5 training failures on B650 and X670. If you're on the launch BIOS from 2022, you're asking for this. Use BIOS Flashback with a FAT32 USB stick so you don't need a working display.
If it still won't POST
- Check for bent CPU pins. On LGA1700 and AM5, one bent pin in the memory address or data group kills one channel. Pull the CPU, look at the socket under a bright light at an angle. A single bent pin can cause exactly this — POST with one stick, red DRAM LED with two. If you see one, don't try to bend it back with a fingernail. Get a proper pin tool or RMA the board.
- Loosen the CPU cooler. Over-torqued coolers bow the socket and break contact on the outer memory pins. Back each screw off a quarter-turn and retry. This is real, and it's bitten a lot of people who used a torque driver on an AIO.
- Try slots A1/B1 instead. It's a slightly worse signal path, but if A2/B2 is physically damaged (bent slot pin, cracked solder joint), A1/B1 will at least boot. You lose a bit of memory bandwidth, not stability.
- Test the sticks in another machine. MemTest86 from a USB stick for two full passes. If either stick throws errors, you have your answer. A stick that fails training intermittently is a dead stick, not a config problem.
Prevention
Always check your board's QVL before buying RAM, and match the kit to the slot layout printed on the board itself — not the manual's generic diagram. On a four-slot board, that's almost always A2/B2 for two sticks, which is where you already are, so the slot choice wasn't wrong. After any RAM change, clear CMOS before the first boot. It costs 30 seconds and saves an hour of LED-staring. And if you're on AM5 with a 2×32GB or larger kit, expect a 60–120 second training window on the first cold boot — walk away, don't mash the power button.
Rule of thumb: red DRAM LED plus beeps with no display after a RAM change is almost never a dead board. It's a training failure. Reseat, reset, then isolate.