Step 1

Start clean

• Load EXPO first, then apply only the preset changes you are testing.
• Keep CPU undervolting or Curve Optimizer out of the first pass.
• Test timing changes separately from clock changes when possible.

Step 2

Use short reject tests first

• Run a quick OCCT memory pass first to reject obviously bad settings.
• If a preset fails immediately, back out the most recent timing cluster first.
• Do not jump straight into a many-hour run on a fresh tune.

Step 3

Then move to timing-focused stress

• Use TM5 next for memory timing validation.
• If Safe passes, tighten toward Optimal one group at a time.
• Extreme values should be treated as individual experiments, not a bundle.

Step 4

Cold boot on purpose

• Power the system off fully and boot again every few cycles.
• AM5 can pass after a reboot and fail after a true cold boot.
• If training starts looping, step back before pushing more voltage.

Step 5

Finish with long mixed load

• Use a long mixed stress run such as y-cruncher VT3 after timing-specific tests.
• Watch DIMM temperature the whole time, especially with tight tRFC and high tREFI.
• If you use active RAM cooling, this is where the higher tREFI note matters most.

Step 6

What to record

• Preset name, voltages, MCLK, FCLK, Nitro values, and changed timings.
• DIMM temperature during the long run.
• Whether the setting survived both warm reboot and cold boot.