Memtest86+, the little RAM tester, flexes FOSS muscles with v7.0

The revival in development of open source RAM-testing tool Memtest86+ continues with version 7.0 – and it's not just for Linuxy types. Memtest86+ is a FOSS tool for testing a computer's RAM. It's tiny, simple, and has no restrictions on use. If a computer becomes unreliable, a great first step, which doesn't even require …

[Jou (Mxyzptlk)]

I love memtest, but as for ECC...

For ECC, at least the real one and not the light-ECC included in standard DDR5 modules, consult your OS log.

For Windows:

Event protocol: System

Event id: 47

Event Source: WHEA-Logger

"Correctable error, component memory".

In XML view you even get the exact address. And it the BIOS does init it correctly: The bank/module number.

How that message looks in Linux or Mac: Please fellow RegCommentards, fill me with real examples. My Windows example was forced with overclocking the RAM.

[collinsl]

Re: I love memtest, but as for ECC...

On Linux you get MCECheck errors from the mcelog or rasdaemon processes which show up in your logs and on the console, and if configured also generate emails to your root account (which hopefully you've redirected to a useful email address you actually read).

[Jou (Mxyzptlk)]

Re: I love memtest, but as for ECC...

As for Ryzen ECC capabilities: No need to forward the mail. The machine suddenly gets very slow, about 10th of the speed (not measured, but it feels like that), but it keeps on going and allows a normal shutdown without data errors. Here it shows the difference between server grade ECC and consumer grade ECC: The ECC error correction circuit seems to be a lot slower on Ryzens compared to Epyc/Threadripper/Xeon which only slow down a bit. That distinction is fine by me, I don't need full speed with detected ECC errors on a consumer machine. On a server it would be a different story.

[Anonymous Coward]

Re: I love memtest, but as for ECC...

> I don't need full speed with detected ECC errors on a consumer machine. On a server it would be a different story.

I like that approach (where appropriate). "Slower" is still much faster than "crashed".

[GrumpenKraut]

It also does a speed test

Giving memory transfer rate for cache levels L1 and L2, and for out of cache.

[chuckufarley]

You can also...

...install Memtest86+ directly into your Linux desktop/server. Adding it to the grub boot menu is usually done automatically and them you have the option to boot into it without needing a USB stick.

[PRR]

> a few DIMMs suddenly go bad in the last decade

Do you try taking it out and putting it in again? Long ago I second-level supported a school room of Apple ][s, and the most common 'fix' was re-seating all the RAM. In DIP form you usually don't even take them out: pressing down scuffs the tarnish on the pins and frequently restores happiness for a semester.

I've also done this on decade-old netbooks in light 24/7 duty. They get goofy, I take-out put-back the RAM, un-goofed for another few years.

[C R Mudgeon]

That satisfying crunch of the chip settling back into its socket.

[ldo]

Do you try taking it out and putting it in again?

Could that be because the memory module got slightly loose over the years? Or is it because of a thin layer of corrosion on the pins, again building up over the years?

Either way, taking it out and putting it back is likely to fix the problem, in the latter case by rubbing enough corrosion off the pins to restore suitable electrical contact.

[John Brown (no body)]

Also, "thermal creep", especially with those old DIL sockets. Thermal cycling up and down makes the metal legs and spring contacts expand and contract ever so slightly such that over time the chips creep out of the sockets. Any mechanical connection that is a friction-only joint suffers this, including screw in terminals.

And yes, that "crunch" of the chip going back fully is a very satisfying sound, especially to a field engineer of many years :-)

[DS999]

Nice timing

As I'm about to build a new PC using ECC DDR5 when the latest Ryzens are available on Jan 31 - first PC I've built since the turn of the century that supported ECC since Intel and AMD decided it no longer mattered outside of server class CPUs/boards.

[Jou (Mxyzptlk)]

Re: Nice timing

Be careful to explicitly get ECC RAM, easy to identify by RAM chips being multiple of 9 instead of 8. Don't be fooled that all DDR5 is ECC, since they force-added ECC in the DDR5 spec. Albeit with a bigger hamming to require less bits, and not exposed to the CPU.

[DS999]

Re: Nice timing

They are a multiple of 10 for DDR5 - 2x32 bit channels means you need 80 bit wide DIMMs for ECC.

That's not necessarily "easy" to identify as they make all sorts of different combinations - I've seen some ECC DIMMs in the past that had three chips: two x32 and one x8. I wouldn't be shocked if there are some DDR5 DIMMs out there with four x16 and two x8 (can't do five x16 due to the split channels)

[Jou (Mxyzptlk)]

Re: Nice timing

> They are a multiple of 10 for DDR5 - 2x32 bit channels means you need 80 bit wide DIMMs for ECC.

You are right. Confirmed by picture search. Thx!

[Korev]

> has been criticized by noted industry diplomat Linus Torvalds

It's nice to see the El Reg of old pop up from time to time

A pint for Herr Proven -->

[Liam Proven]

[Author here]

Why thank you!

This humble scribe is presenting a talk at this year's FOSDEM in Brussels, and is on a panel discussion too. Real world pints of Belgium's finest are very welcome. ;-)

[BenMyers]

Memtest86+ is super

I've kept up with the latest Memtest86+ and I found the 6.xx versions very good, and far better than Passmark's product. Can't wait to get version 7.0. Will download it now.

[John Brown (no body)]

Turn the computer off, remove all but one memory module, and try again.

Noooooooo!!!1!!ONE!!11!!!!

I thought EVERYONE knew that you start testing in the middle and eliminate one half the system/device/circuit before or after the test point[1]. If RAM testing with modules of any type, or even old DIL RAM[2] in older kit, you pull half the banks first and test half the RAM. If it fails, you likely have a single SIMM/DIMM/DIL chip fault in that half and the half you pulled is all good. Likewise the converse if that test passes. If you REALLY know what you are doing, you can probably work out which SIMM/DIMM module is faulty from the address(s) of the reported failures RAM is a fairly rare failure, so more than one RAM failure at the same, while not unheard of, is fairly unlikely.

[1}, yes, sometimes there's more than one fault in two or more very different places.

[2] in the case of DIL chips, odds are you can identify which bit is fault from testing and just pull the bad chip.

[RAMChYLD]

Memtest86+ is definitely a very useful engine

I have it on my Ventoy too. I use it whenever I try tuning the memory in my current AMD machine which seems to have developed a weird flaw where the RAM becomes extremely unstable if I try to run it at 6000MHz.