You bought a 3600MHz kit, enabled XMP in BIOS, and now the machine either won't POST at all, loops on the memory training screen, or boots into Windows just long enough to throw a WHEA_UNCORRECTABLE_ERROR or IRQL_NOT_LESS_OR_EQUAL BSOD. Turn XMP off and everything works. That's the exact pattern, and it's maddening because it feels like you got sold bad RAM.
I've hit this on a 5800X with a 3600 CL16 kit, on a 12700K with DDR4-4000, and on a 3900X that refused anything past 3200. Same story every time.
What's actually happening
XMP and DOCP are just pre-validated overclock profiles stored on the RAM sticks. They set frequency, timings, and voltage to the kit's rated spec. But the rating is validated on one specific test bench with one specific CPU and motherboard. Your board and CPU aren't that bench.
The bottleneck is the memory controller inside the CPU, not the RAM itself. On Intel 10th/11th gen it's rated for DDR4-2933. On AMD Ryzen 3000 it's 3200. Anything above that is technically overclocking the IMC, and some chips hit the silicon lottery better than others. The other factor is the Infinity Fabric on Ryzen, which runs 1:1 with memory up to a point, then drops to 2:1 and tanks latency.
DOCP is just ASUS's name for XMP on AMD boards. Same mechanism, different branding.
So when you enable the profile and the system BSODs at 3600MHz, the RAM isn't defective. The CPU's memory controller or the board's signal integrity can't hold that speed at those voltages.
The fix, in order
- Update your BIOS first. This is the single highest-impact step and people skip it constantly. AGESA updates from AMD and microcode updates from Intel add memory compatibility with every release. On a B550 board from 2020 that's never been flashed, a 2023 AGESA can mean the difference between stable 3600 and boot loops. Flash it. No excuses.
- Reset to defaults and enable XMP cleanly. Clear CMOS with the jumper or pull the battery for 30 seconds. Boot into BIOS, load optimized defaults, save, reboot. Now enable XMP/DOCP and nothing else. A lot of instability comes from stacked settings (fast boot, PBO, manual timings) fighting each other.
- Drop to the next speed tier. If 3600 won't POST, manually set the frequency to 3466, then 3200, then 3000. Leave XMP timings and voltage on. On Ryzen, 3200 CL16 with FCLK at 1600 in 1:1 mode often beats 3600 CL18 with the fabric forced to 2:1. The lower latency wins in real workloads.
- Bump the memory controller voltage. This is where most people stop too early. On Intel, that's VCCIO and VCCSA. On AMD it's VSOC. Do not go crazy here.
Set VSOC to 1.10V and try XMP again. If it works, stop there. If BSODs persist, drop frequency.Intel DDR4: VCCIO 1.20V, VCCSA 1.25V (max safe ~1.35V) AMD Ryzen 5000: VSOC 1.10V (max safe 1.15V) AMD Ryzen 3000: VSOC 1.10V (max safe 1.15V) - Loosen the primary timings. If the kit is 3600 CL16-16-16-36, set it to 3600 CL18-18-18-38 manually with XMP voltage. Loosening tCL and tRCD by 2 ticks often resolves borderline instability with near-zero performance loss. Skip tRFC tweaking for now unless you've already got everything else stable.
- Check DRAM voltage. XMP typically sets 1.35V. If the kit is rated for 1.35V at 3600 and you're hitting instability at 1.35, nudge to 1.38V or 1.40V. Don't exceed 1.45V on DDR4 for daily use. B-die can take 1.5V, most other ICs can't.
- Verify with MemTest86. Boot MemTest86 from USB and run at least two full passes. One pass isn't enough. Errors within the first pass mean your settings are broken. Errors on pass 2-3 mean borderline. Zero errors on four passes and a clean 2-hour OCCT run means you're good.
- Check the QVL list. Your motherboard manufacturer publishes a Qualified Vendor List of memory kits they've tested at each speed. If your kit isn't on it for your specific CPU generation, you're on your own. Cross-reference it. A kit validated for Ryzen 5000 at 3600 might not be validated for a 5950X with all four DIMM slots populated.
If it still fails
Run one stick at a time in slot A2 (second from CPU). If a single stick at XMP works but both together fail, you've got a dual-rank or four-DIMM loading problem, which caps most consumer platforms around 3200-3466. That's a hardware limitation, not a bug.
Check that the DIMMs are in the correct slots. On most boards that's A2 and B2, the second and fourth slots from the CPU. Populating A1/B1 first breaks signal integrity at high speeds and is a stupidly common cause of this exact symptom.
Run sfc /scannow and chkdsk /f if you've been BSODing for a while. Memory errors corrupt files, and you can end up chasing a stable system that still crashes because Windows itself is damaged.
Finally, if 3200 is the highest stable speed and you paid for 3600, that sucks but it's not the retailer's fault. The IMC on your specific CPU just can't do it. Return the kit and buy a 3200 CL14 or CL16 B-die kit instead. Tighter timings at a lower speed will feel identical in real use.