Stability test · CPU
CPU stress test.
A long load on all cores, then on every core on its own. Every chunk of work has a known answer and gets checked, so the test also catches a CPU that does not freeze but simply calculates wrong. An error names the exact core and CCD.
- 1 min to 1 h for free
- No time limit in PRO
- Temperature guard
CCD 0
CCD 1
expected 0x9F3A…, got 0x9F2A…
An illustration of the test window with a sample error. The real window shows tiles by the cores and CCDs of your processor.
What the test does
Three workloads and every core alone.
Every thread is pinned to its own logical processor, so you know which core gets it wrong. The workloads rotate, because different kinds of work run through different clocks and voltages.
Big number multiplication
Multiplying 524,288-bit numbers with an NTT transform. The result is checked with an independent method, the way y-cruncher does it.
AVX2 FMA
Eight independent FMA chains on 256-bit vectors with data in the L1 cache. The highest power draw and temperature.
Sorting
2 million 64-bit numbers (16 MB) per thread, with an order and checksum check. It also loads the L3 cache and the memory controller.
One core at a time
The first thread of every core works alone. With one core loaded the boost is at its highest, and that is exactly when an aggressive Curve Optimizer fails most often.
When to use it
Why test, if games run fine?
An unstable CPU rarely freezes right away. More often it drops a game to the desktop once every few hours, corrupts a file or leaves a WHEA error in the log. The test shows it under controlled conditions.
PBO and Curve Optimizer
After every change of the negative offset. An error on a specific core tells you which core needs a smaller value.
Undervolting
Lower voltage means less heat and noise, but too low and the CPU starts calculating wrong. The test catches that point.
Fixed clock overclock
See whether the clock holds under full load and under the heaviest FMA work.
A new PC
A new CPU, motherboard or cooler. A short test shows right away whether something is off, before you start gaming.
Crashes in games
Random drops to the desktop, freezes and blue screens. The test helps tell an unstable CPU apart from a problem in the game.
After a BIOS update
A new BIOS can change voltages and default limits. You can read your settings in the BIOS tab.
How to read the result
Passed, failed or stopped.
A stability test gives no points and does not go to the ranking. The full report with the error list stays in the app’s history.
No chunk returned a wrong result and no new hardware error event (WHEA) appeared in the Windows log.
A calculation error or a new WHEA event. The report names the logical processor, the physical core (P or E on hybrid CPUs) and the CCD.
You stopped the test, the CPU hit 95 °C in three readings in a row, or a laptop switched to battery. This is not a stability verdict.
- An error on one coreMost often the Curve Optimizer value for that core is too strong. Make its negative value smaller and test again.
- Errors on many coresMore often the whole CPU voltage is too low, the clock too high or the temperature too high. Go back to your last stable settings.
- An error only in sortingSorting also loads the L3 cache and the memory controller. Run the RAM stress test as well.
- A WHEA event without a calculation errorThe CPU itself reported a corrected or uncorrected error to Windows. That also means your settings are on the edge.
How to test
Short after a change, long before you keep it.
- Save your workWith unstable settings the PC can freeze or restart. Close games and heavy programs.
- A short test after every changeA few minutes is enough to catch a clearly too aggressive setting before you go further.
- A long test at the endWhen the settings are meant to stay, run the test for an hour, or in PRO with no time limit, for example overnight.
- Check the speed tooStable does not always mean faster. The benchmark shows whether a change gave you anything in points.
To be honest: a pass means there were no errors during that time and under those workloads. It is not a guarantee that different work at a different time will never hit an error. A longer test gives more confidence.
Questions
FAQ
How long does the test take?
From 1 minute to 1 hour in every version of the app. With no time limit, until you stop it (24 hours at most), in PRO.
How is this different from just heating the CPU?
Every chunk of work has a known answer and gets checked. With too little voltage a CPU often does not freeze, it just calculates wrong. Heat alone will not show that.
Why is there a “one core at a time” phase?
With one core working the boost clock is at its highest. An aggressive Curve Optimizer often fails exactly then, not under full load on all cores.
The test found an error on one core. What now?
That core is unstable with your current settings. With Curve Optimizer, make the negative value for that core smaller and test again. With a fixed overclock, lower the clock or raise the voltage within a safe range.
What happens if the CPU gets too hot?
A guard checks the temperature 4 times per second. At 95 °C in three readings in a row the test stops with the result Stopped.
What is a WHEA event?
Windows logged a hardware error in the system log (WHEA-Logger). The test checks the log during the run and at the end. A new event ends the test as Failed, even if the calculations matched.
Check that your CPU calculates without errors.
The stability test is in the Benchmark tab in SmoothWizard. 1 minute to 1 hour in every version, with no time limit in PRO.
