My latest test: arm_freq=3000 and no over_voltage_delta, stable over 8 hours using Pi-Under-pressure.And then perhaps even longer than 30 minutes, if it doesn't explode in 30. See here.
Code:
Step arm_freq over_voltage_delta ResultCPU_temp max---- -------- ------------------ ------------------1 2400 MHz 0 Baseline (stock)63.1°C2 2450 MHz 0 PASSED68.1°C3 2500 MHz 0 PASSED68.6°C4 2550 MHz 0 PASSED68.6°C5 2600 MHz 0 PASSED68.6°C6 2650 MHz 0 PASSED70.8°C7 2700 MHz 0 PASSED73.0°C8 2750 MHz 0 PASSED73.0°C9 2800 MHz 0 PASSED74.1°C10 2850 MHz 0 PASSED75.7°C11 2900 MHz 0 PASSED76.3°C12 2950 MHz 0 PASSED76.8°C13 3000 MHz 0 PASSED76.8°C14 3050 MHz 0 PASSED77.9°C15 3100 MHz 0 PASSED81.2°C16 3150 MHz 0 <rebooted itself during test>17 3150 MHz 10000 <segmentation fault>18 3150 MHz 20000 <segmentation fault>19 3150 MHz 30000 <segmentation fault>20 3100 MHz 0 PASSED80.1°C21 3100 MHz 0 1h test → rebooted???22 3050 MHz 0 PASSED76.8°C23 3050 MHz 0 1h test - PASSED24 3050 MHz 0 1h test - <segmentation fault>25 3000 MHz 0 1h test - PASSED81.8°C26 3000 MHz 0 8h test - PASSED79.0°COne of my favorite torture tests. My last test on this pi5, building gcc-10.2.0 (c, c++ and fortran) ran in 1 hour 21 minutes. We'll see what a 25% increase in clock speed gives. Just over 1 hour? And, after that, the REAL torture test -Note that being able to run the full suite of tests provided by any over clocking tool does not prove that errors won't occur when running the different instruction sequences of other programs.
Strange as it sounds, after finding optimal settings based on Pi-Under-Pressure, I'd suggest trying a reproducible build of gcc to check whether it is still reproducible or not.
Code:
./pichart-openmp -r8 -w30Statistics: Posted by tinker2much — Sun Jan 04, 2026 3:24 am