Quick answer: Your CPU's memory controller can't handle the XMP profile's voltage or timings at stock—manually bump VDDQ/VDDIO to 1.35–1.4V, relax tREFI, or drop to a lower XMP profile.
What's actually happening here is that DDR5 XMP profiles are aggressive. They're tuned for a specific motherboard and CPU combo, but the memory controller on your chip might be weaker. When you enable XMP, the board applies voltages and timings that the controller can't sustain, so it fails to train the memory during POST. That's why you get a boot loop or a black screen. It's not a faulty stick—it's a mismatch between the profile and your silicon's capability.
I've seen this a lot on 12th and 13th gen Intel with DDR5-6000 kits. The XMP profile sets VDDQ to 1.35V and tREFI to 32767, but the controller wants more voltage or looser refresh intervals. On AMD Ryzen 7000, it's often the opposite: the Infinity Fabric can't keep up with high memory clocks, so you get WHEA errors or random reboots under load.
Fix it step by step
- Clear CMOS and boot with XMP disabled. Pull the battery or use the jumper. Get into BIOS with default settings. This confirms the system is stable without XMP.
- Update your BIOS. Memory compatibility improves with AGESA or ME updates. Check your motherboard vendor's site for the latest version—don't rely on the one that shipped.
- Enable XMP but drop the frequency one step. Instead of 6000 MT/s, try 5600. This often fixes training failures while keeping most of the performance.
- Manually set VDDQ and VDDIO to 1.35V–1.4V. These are the I/O voltages for the memory controller. On Intel, they're called VDDQ TX and VDDQ CPU. On AMD, it's VDDIO. Bump them in 0.05V increments.
- Relax tREFI to 32767 or lower. High tREFI (like 65535) causes instability at high temperatures. Setting it to 32767 or even 16384 can stabilize things.
- Increase SoC voltage on AMD (max 1.25V) or VCCSA on Intel (max 1.3V). These feed the memory controller. Too low causes training failures; too high degrades the CPU. Stay within safe limits.
- Test with MemTest86 or TM5. Run at least one full pass. If you get errors, back off the frequency or loosen timings further.
The reason step 3 works is that DDR5 training is sensitive to signal integrity. Dropping 400 MT/s reduces the strain on the controller and the motherboard traces. You lose maybe 2–3% performance, but you gain stability.
If that doesn't work, try these alternatives
- Use a different XMP profile. Some kits have two profiles—one for Intel, one for AMD. Make sure you're using the right one.
- Set command rate to 2T. This gives the memory more time to respond and can fix boot loops.
- Loosen primary timings. Increase tCL, tRCD, and tRP by 2–4 cycles. It's a band-aid, but it works.
- Try a different slot pair. On four-DIMM motherboards, slots A2 and B2 are usually the best for two sticks. Check your manual.
- RMA the kit. If nothing works, the sticks might be marginal. I've had kits that only ran stable at JEDEC speeds—that's a bad bin.
On AMD Ryzen 7000, don't exceed 1.25V SoC. Higher voltages can cook the CPU. I've seen chips degrade from 1.3V+ within months.
Prevention tip
Before buying DDR5, check your motherboard's QVL (Qualified Vendor List) and your CPU's official memory support. For Intel 12th gen, DDR5-5600 is the safe bet; 6000+ is a gamble. For AMD Ryzen 7000, DDR5-6000 CL30 is the sweet spot, but only if your board's BIOS is updated. And always test with MemTest86 before trusting your data to it. If you're not comfortable tweaking voltages, stick to JEDEC speeds—the performance difference in real-world tasks is smaller than benchmarks suggest.
One more thing: if you're running four sticks of DDR5, expect to run at much lower speeds. The memory controller struggles with four ranks. Two sticks is always better for stability.