MBP A1707 2017 RAM issue

[Beeldbuisje]

Short history of this machine...
Runs on Catalina, customer wants to upgrade to BigSur. Crash and not possible to install no matter which OS.
Reinstalled a TimeMachine backup from Catalina on an empty SSD. Succes.

Worked well, customer doesn't do heavy tasks. A few years later, he had to upgrade to Monterey, same problem.

I did the same (install a TM backup), but once booted in OS machine does weird things, but is operative.

Let run Memtest en gives a RAM error. Look on thumbnail
https://www.badcaps.net/forum/images/attach/jpg.gif

During the whole test one thing repeats: only when testing reaches 14GB-15GB errors are generated.
In the error list, "failing adress" displays always (14.5GB).

Is there a way to determine which RAM chip gives me the issue?

tnx

[piernov]

Re: MBP A1707 2017 RAM issue

Only 4 RAM chips on this board, shouldn't be a problem to replace them all at once.

However in this screenshot only the 62nd data bit is failing so that narrows it down to only 2 ICs on 2 different channels.

Finding which channel it is is very difficult (if not impossible) because the addressing is pretty complex and documentation about physical addressing with Intel memory controllers is not publicly available.
There is some reverse-engineered documentation but unfortunately it doesn't cover all situations and it could contain some mistakes. (e.g. https://cdn.badcaps-static.com/pdfs/...5e7875c4f4.pdf "DRAMA: Exploiting DRAM Addressing for Cross-CPU Attacks")

Btw Memtest86+ is a great open-source tool, but I'd recommend also confirming with the proprietary Memtest86. Apple has a bunch of quirks in their EFI implementation and the proprietary Memtest86 has several workarounds for that.

[Beeldbuisje]

Re: MBP A1707 2017 RAM issue

I don't have BGA rework tools by myself. But I have a donor board with 99% for sure good Ram chips.
I spoke to my friend with all the necessary BGA rework tools. He told me it's easier to replace the 4 chips at once as one by one, testing,... etc.

When it's almost impossible to determine which chip is defective, I'll change them all.

PS: thanks for the documentation provided!