Inconsistencies while testing

[patriotdoge]

First congrats on the tool, it's amazing, keep up the great job.

I will report the following as a sanity check:
Imagine this I have an 5900X, Windows 11 24H2, currently using this curve values -4, -8, -24, -16, -12, -12, -30, -30, -30, -30, -30, -30.
RAM OC is stable, tested with TM5 1usmus_v3 profile, 30 cycles (3800Mhz - CL14-14-14-28-42 CR:2T - IF: 1900)

When I run OCCT with the above curve, be it on Auto, Fixed or Core Cycling, using Data Set: Large, Mode: Normal and Load Type: Variable (I already did all modes possible believe me) I don't get a single error reported.

I already did almost a week with CoreCycler (using Prime95 default, starting with different values from the above, to let the auto tune adjust it) and always get some errors that brings up the V/F value of some cores.

My final curve after all this time is 0, -9, -23, -13, -10, -12, -30, -30, -30, -30, -30, -30, for some cores the gap is significant, and I believe if I let the tool run more time it will up more the values for sure...

What can be the problem/issue between both tools, for one reporting no errors and CoreCycler always at some point finds one?
I always considered OCCT the standard to achieve a stable platform, just wondering why CoreCycler is always "spitting" an error.

[sp00n]

You can check in the log file which errors were thrown.

Also, did you run OCCT for one week as well to find your curve? 😀
Stability testing basically is like that if you deviate from the manufacturer defaults, the longer you test, the more likely is you'll find an error, you're always just on the path to full stability. There really isn't such a thing as "guaranteed 100% stability", always only a certain degree of stability, which can be closer or not so close to 100%.

And especially once you've reached a pretty solid degree of stability, encountering errors will take longer and longer, at which point many people already will have stopped testing. But finding these last 20% of errors can easily take up 80% of the testing time.

Regarding the RAM, 30 cycles of TestMem5 might also just not have been enough. I've treated RAM settings as unstable after finding an error after 12+ hours of stress testing, and did strive for a 24 hour stability test. Of course you'll always have to find your personal point of preference for balancing the desired stability vs. the required testing duration here.

But if you're not sure if your RAM is stable in the long term, you cannot be sure either if the error thrown by CoreCycler did come from the RAM or from the CPU undervolt, so it's kind of important to be pretty confident there (or disable the RAM overclock for getting a CO curve first, and then test both in combination later).

(Additionally, Ante777 Absolute or Extreme config is said to be slightly harder than 1usmus for TestMem5.)