An old version of CoreCycler detects an unstable core, but the new one doesn't

[HerMajestyDrMona]

Hi.

I used CoreCycler in April 2022 to setup my voltage offsets. Since I found some WHEA errors in Event Viewer (around 1 per month), I decided to repeat some tests.

I accidentally ran the old version (0.8.2.4) instead of the newest one (0.9.5.2), and for some reasons it finds a problem within the first 5-10 minutes on Core 0. However, I can not reproduce this detection on the newest version of CoreCycler (PRIME95), even during longer tests.

The error is:

FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected running 720K FFT size, consult stress.txt file.

I wonder if there was some bug in the old version that could trigger errors / detect unstable cores with no reason, or it's actually the older version of PRIME95 that seems to catch errors faster? That core (0) was never in WHEA errors, but I am just trying to make sure it's fully stable :)

[sp00n]

So this is a "true" Prime95 error and not something that CoreCycler would be able to influence.
You could try to copy the Prime95 version from the old CoreCycler directory to the directory of the new one (test_programs\p95). Prime95 has gone through some config changes over the time, but it should still work, although I haven't tested this explicitly.

And besides config changes, Prime95 also had some changes in the way it does the stress test, so it may very well be that this specific version of Prime95 detects errors faster for you than other versions.

I do remember one instance where Prime95 itself was buggy and caused errors (Prime95 version 30.4 & 30.5 - issue #19), but the version used in CoreCycler 0.8.4.2 should already be the one that had fixed this problem (Prime95 30.6 build 4).

Also, the default config doesn't test 720K in Prime95, so if you were running the new CoreCycler with the stock configuration but the old one with a custom config, this would be another explanation why you didn't see the error in the new CC.
720K seems to be a pretty popular test size for Prime95 though.

[HerMajestyDrMona]

Thank you for the answer.

Copying the old "p95" folder is causing it to run on all cores, so I suppose there's some incompatibility.

I now changed to FFTSize = 720-720, but so far nothing is detected, even though I lowered the voltage by -1 (so it's -16 now, the same as before I changed it to -15 two days ago).

It's strange that on 0.8.2.4 it's always the same error:

FATAL ERROR: Rounding was 0.5, expected less than 0.4

In like 10 tests it was always the same error about 0.5 vs < 0.4 . I would expect it to be more random, but maybe I'm just not fully understanding the patterns it makes.

It's rather impossible that the old PRIME95 version had some bug that could error with no reason? I suppose it's impossible, because it isn't happening on the other cores so far :)

[sp00n]

Ha, I actually discovered two bugs with the compatibility with old Prime95 versions.
You could try again with this current development version.
script-corecycler.zip