memtest radom halt without bit error (2/5)

[Jessica]

memtest: 11.4 Pro
CPU : Intel Core Ultra 5 225H
Memory : Team Group – TE48GFSXV2TH
testCONDITION: 5600MT/s (2DIMM)

We encountered a problem that we did memtest testing 3-5 times (over night testing 12-16hr ), and found memtest halt without bit error. (1-2 time over 5 times testing)
How do we explain this situation?
(We have been testing 20 times, using 3 different sample boards and all without bit error but halt appear randomly)

[David (PassMark)]

There are these warnings in the log
"aborted waiting for CPU #5 to finish"

This is a timeout message indicting that some of the multi-threading code too way too long to finish.

There are three main possibilities for this,
1) A thread (CPU core) crashed, but the rest of the software continued running
2) One thread really was very very slow. This CPU has some "Low Power Efficient-cores" which are around 4x slower than a normal CPU core. So maybe that is the issue here. But it shouldn't result in the who system locking up. We are working on a new release that has better timeouts as well.
3) There is a bug in the BIOS firmware code for multi-threading. This was pretty common a few years ago.

So trying running with just 4 CPU cores as an experiment.

[keith]

Also, we made various fixes/improvements related to multicore testing since v11.4. Can you give this build a try:
https://www.passmark.com/temp/memtes...-11.5.1011.zip

[Jessica]

As soon as we get experement results, we'll let you know.
One of my colleague told me, in meterlake platform, he found if he set PL2 to lower value, memtest halt more often. (core Ultra 5 135U 2p+8e)

[Jessica]

We've been using 11.5 free version testing 8 times, all went good. (no halt, no error).
But there are a lot of Warnings (CPU completed but did not signal)

Can we say the memtest halt in 11.4 version (or lock up) is not relevant to hardware flaw?


P.S. Oririn logFile is too large to upload, so I delete part of content and replace some keywords. (origin log is 9M)

[keith]

Thanks for the logs.

There was a slight regression resulting in the warnings in the file. This should be fixed in the following build:
https://www.passmark.com/temp/memtes...6.1000.rc1.zip