Why does running memtest seem to prevent kernel panics on my old iMac?

[Buadhai]

I've posted several threads on the kernel panic problems I've been experiencing with my 2017 21.5" iMac. The general consensus seems to be that it is a hardware problem. I'm convinced that it's not worth buying a new logic board for such an old machine. I plan to replace it "real soon now".

I previously ran memtest from an external USB boot drive several times when trying to figure this out, but it always passed 100%.

The other day I thought I'd run memtest from the GUI called Rember. (Note that this only checks free RAM.) I did an initial run of 25 loops. It passed 100%, but what surprised me was that the machine never had a kernel panic during the run. So, I thought I'd run it longer and did 100 loops which took almost 48 hours. Not a single panic.

But, once I had finished memtest, there was a panic a few hours later.

So, I started memtest again running 255 loops. That was 24 hours ago and still no panics.

As you can see, the day before I ran memtest there were 29 kernel panics. There were none in the 48 hours that memtest was running, one when I briefly stopped it and none since I started it again.

Kernel-2022-10-25-054149.panic
<snip 25 panics>
Kernel-2022-10-25-170958.panic
Kernel-2022-10-25-172315.panic
Kernel-2022-10-25-172545.panic

memtest-2022-10-26-100100 - Start
memtest-2022-10-28-113100 - Stop

Kernel-2022-10-28-152341.panic

memtest-2022-10-28-160100 - Start

I fully understand that this might be a fluke and I also understand that running memtest continuously puts a rather heavy strain on the cpu (nearly 100%). However, I don't actually notice a performance hit and it's nice to have the machine back without having to spend $$$ on a new logic board and without having to put up with constant panic reboots.

 

[Bodhitree]

Sounds like your memory is faulty, it is fine when “kept hot” by memtest and otherwise errors creep in which cause kernel panics.

 

[Buadhai]

Sounds like your memory is faulty, it is fine when “kept hot” by memtest and otherwise errors creep in which cause kernel panics.

That’s an interesting thought and makes a lot more sense than anything I’ve come up with.

It’s sort of a Heisenberg Uncertainty Principle of hardware testing. The act of testing the RAM renders the RAM untestable.

So, is there a less CPU intensive way to keep the RAM "hot"?

Oddly, when I booted the machine from a memtest86 boot thumb drive it ran RAM tests for hours and found nothing. I guess now I have a hint as to why.