Test errors with Intel 265k

[espinaler]

I have an Intel 265k and an Asrock Z890 LiveMixer WiFi motherboard with the latest BIOS (3.0

I'm getting errors with 2 different memory kits from Kingston and G.Skill. The Kingston kit is on the motherboard QVL, the G.Skill kit is not. Both kits have passed testing in another computer. Turning XMP/EXPO off doesn't seem to change anything.

It's only ever Test #7 that fails. The number of errors can range from double digits to thousands. Occasionally, a test run will pass with zero errors and I'm still trying to figure out how to repeat that reliably.

What's puzzling is that the computer seems stable in use, and overnight testing with other software (y-cruncher, AIDA64, HCI MemTest Pro) hasn't turned up any issues. It's just weird to me that memtest86 can find so many errors but there's no apparent problem otherwise and no other tests are failing.

I've attached a log from a recent test run /w the Kingston kit. I only realized this morning that I'm a minor version behind but a quick test with just #7 and v11.4 did not seem to make a difference.

[David (PassMark)]

Posting the final HTML report is actually better than posting the debug log (unless you think there is a bug in MemTest86 itself).

Final HTML report contains a nice summary and it much more readable.

See this page to start with
https://www.memtest86.com/troubleshooting.htm

MemTest86 is a better memory testing tool that those other programs.

Errors in log does look like a RAM failure. But if you have a failure with two different sets of RAM you need to start to suspect the motherboard slot or CPU or BIOS bug.

What's puzzling is that the computer seems stable in use

But something going bad prompted you to run MemTest86 right? I am guessing you aren't doing this for fun

[espinaler]

I'm apparently not allowed to post files with an HTML extension, so I've renamed it.

I was previously trying to diagnose some visual glitches, but that was solved with a gpu driver rollback.

[David (PassMark)]

I had a look at the HTML report.

Some notes:

1. You are using the Pro version of the software, and so MemTest86 identified the bad memory module. But it claims both modules are bad (slot A2 and B2). Which isn't what you normally get for a RAM error. Normally errors are localized to a single stick of RAM, or sometimes random everywhere.
2. All the errors are in Test #7 (Moving inversions) and the number of errors was about the same on both RAM sticks (~5000 each). This also implies the error isn't on a single stick.
3. There are a large number of errors (> 10,000). So many that MemTest86 gave up testing. But they only get triggered on Test #7. So maybe there is some access pattern or data pattern that acts as a trigger. If this happened all the time the machine wouldn't even boot.
4. All the errors seem to be on the same bit (#14). So you see results like this, Expected: 00004000, Actual: 00000000. So a single bit that should be a "1" is being flipped to a "0", every time. So this isn't totally random. There is a strong pattern. So this isn't something like the power supply, as that would be way more random. Also probably not the slot on the motherboard nor the track on the PCB to the slot. This makes is less likely that it is a motherboard fault.
5. The error isn't localized to a single CPU core. The list of cores that detected memory errors is core 0 to 17. So most of them. So probably isn't a single bad CPU core.
6. You are running at 4788 MT/s & 1.1V, which is on the slow side for this CPU & RAM. In theory this RAM can run at 6400MT/s (but at 1.4V). So the problem probably isn't due to running the RAM to fast / hard.
7. Temperatures are fine. So not too hot.
8. Errors are spread across the entire memory range. (5MB to 67569MB). So again, not a localized error in a single chip on a single RAM module.

So what's left?
Needs to be something that impacts just a single bit (#14) and also impacts both RAM slots. Maybe the memory controller or L3 cache in the CPU?
You are using a 256K CPU which uses a LGA1851 socket. Can you borrow a compatible CPU for testing?

[espinaler]

Unfortunately, I don't know anybody with an Arrowlake cpu. Just not a very popular generation among enthusiasts. I'm going to try borrow a third memory kit and see how that goes.

[foobar]

i saw similar failure on test7 block move (movs) with memtest86+ v8.00.
Issue solved when D2D clock was set to 29 instead of 32.
200S boost offers better performance without voiding hardware warranty, but it's not guarantee system stability.
Verified system: Z890 LiveMixer WiFi BIOS 3.21/ 265K/ CMK64GX5M2B6000Z40 *2 = 64GB