Oh of course you can just turn it off. Never realized before.

TLDR - try slowing down “bad” ram

RAM is tricky these days. When you buy “fast” ram - in my case 6200 dominator sticks - what you really are buying is ram that the manufacturer says can be overclocked to 6200 speeds. But this is shady at best. It is entirely possible for that ram to run fine in your rig, and fail miserably in mine - showing itself as errors in memtest. This can be a result of the motherboard or say your cpu - and nothing is broken. Not the chip, not the mobo, not the ram - they just don’t work together at those speeds. Two sticks can work just fine, and then fail when you add two more (for a total of four). I’ve even had sticks that were on the edge, works for a year, and then started giving me trouble.

Most people just RMA those sticks - which I support - if it’s marketed to run at those speeds, then they should run at those speeds everywhere. But if you have “bad” memory laying around that was out of warranty, and given today’s prices - it might be worth pulling them out and trying to run them a little slower. It’s not like you have to take them all the way down to 3200 and turn off the overlock. For example, on many amd rigs, 6000 is the sweet spot. My “bad” stick of 6200 immediately fails a memtest at 6200 (even running that single dimm), but infinitely passes at 6000 (even with 4 sticks plugged in). I can’t even tell the difference between 6200 and 6000.

So ya - try slowing those bad sticks down manually. It might only be a little bit of a performance hit, but at today’s prices, could save you $1000

Also, check your warranty. I just found some bad ram I bought years ago and thought I was in trouble. Turns out g.skill has a lifetime warranty

Imho not worth the risk. Nothing except ECC is protecting system from memory corruption. And consequences of flipped bit can be huge. Unless it is in a system you do not care about, just do not use it.

Very cool to be able to do that even though I'm not sure it's worth it to run a RAM module that has a known defect.

If the fault spreads there will be corruption again.

I tried this for a bit last year, and I never got it to work. I'm sure it was user error, but I'm super glad I got a new kit before shit blew up.

Might have to track down the bad kit I had and give this another try for giggles.

First rule in computer building is to always buy twice as much RAM as you actually need so you can just remove the bad stick when this happens

This is amazing!! I had no idea this was possible!

One point of feedback: not having worked with memmap before, it is not immediately obvious to me which hex means what (I am guessing RANGE@START? ). Would be nice to have a link to an explanation there.

I am definitely sharing this, thank you for writing the blog!

Well done. I recently revived the BadRAM kernel patch in order to do something similar; memtest86+ supports that functionality too, using , , , .

Ooh, that rasdaemon looks interesting. thank for sharing!