Does your memory pass rowhammer test?

L

Luke M

Gawd
Joined
Apr 20, 2016
Messages
618
Just curious...I ran the memtest86 rowhammer test on two near-identical machines with Crucial DDR3 modules (2x8GB non-ECC), one with memory made in 2012 and the other made in 2016. The 2016 memory passed; the 2012 memory failed.

(It's possible that other rowhammer tests could give a different result, I'm not suggesting that memtest86 is definitive).
 
Historically mem86 was always a complete waste of time for me. 1 day, finds nothing. Then IBT fails in 3 minutes.
 
Memtest86 is primarily a memory stability test and IBT is primarily a CPU stability test...

So your anecdotal results prove pretty much nothing.

I've had no problems with the rowhammer test in memtest86. Ran the whole memtest86 test suite on 10x 8-year old Opteron servers a couple months ago. Passed with flying colours.
 
Pandur said:
I've had no problems with the rowhammer test in memtest86. Ran the whole memtest86 test suite on 10x 8-year old Opteron servers a couple months ago. Passed with flying colours.
Click to expand...

Memory that old doesn't suffer from the problem. Plus if it's a "server" it might have ECC.
 
Panmaster said:
Historically mem86 was always a complete waste of time for me. 1 day, finds nothing. Then IBT fails in 3 minutes.
Click to expand...
I can say the same about IBT. IBT ran perfect but I would still got errors in running chkdsk ...


@OP
I haven't had a system with row hammer issues and if i did i would exchange the RAM.
 
Running the test on anything older than DDR3 is mostly meaningless, and even then it has to be some process or smaller.

ECC helps but does not make it immune. It does make crash the most likely outcome versus silent bit flips though, which is a good thing.
DDR4 is also not immune: https://arstechnica.com/security/20...4-memory-shown-to-be-vulnerable-to-rowhammer/

You won't be encountering problems with any normal use though, exchanging your ram is a waste of time unless you are a cloud provider. If you are trying to protect against drive-by exploits, surf smarter not harder and turn off javashit et al by default.
 
I found memtest (not memtest86) to be brilliant at testing while still in Windows.
It needs a new instance running for every 2GB of ram but it works.
Using this my 3000MHz C15 ram now runs at 3466MHz C17 and my PC is rock stable 24/7 with hard gaming and VR.
It showed my previous overclock of 3600MHz wasnt quite stable with a CPU overclock of 4.7GHz so backed off.

ps when pushing the boundaries of CPU overclock and Ram overclock, backing off one can help the other.
 
The DDR3 for the 3770K (Crucial), 4770K (Mushkin), and the DDR4 for the 6700K (Kingston) all passed. Even the 5820K (Crucial) I built for work passed. It may not be much as people imply, but testing has found problems with systems I have built and when the RAM was changed passed. But note, I never overclock and never run memory out of JEDEC specs.
 
Nenu said:
I found memtest (not memtest86) to be brilliant at testing while still in Windows.
It needs a new instance running for every 2GB of ram but it works.
Using this my 3000MHz C15 ram now runs at 3466MHz C17 and my PC is rock stable 24/7 with hard gaming and VR.
It showed my previous overclock of 3600MHz wasnt quite stable with a CPU overclock of 4.7GHz so backed off.

ps when pushing the boundaries of CPU overclock and Ram overclock, backing off one can help the other.
Click to expand...


I've never seen memtest detect errors when memtest86 didn't. However the opposite I have seen plenty.

I would not trust memtest as it appears weak
 
All the Kingston DDR3 HyperX Fury 8GB 1866 MHz, Blue (seems to be made from 1.35V chips) and Black (1.5V Micron chips w/o any visible speed grade).

8 of my 16 samples of old DDR3 Samsung 4GB 1600 MHz passed at least 2 hours of row hammer testing. I didn't try the rest.
 
You must log in or register to reply here.
Back
Top