RAM testing software wanted

SvenBent

2[H]4U
Joined
Sep 13, 2008
Messages
3,323
As a follow up on my testing of 20+ CPU testing software and finding only prime95 and linpack to be really worth their merrits, I am now looking into testing RAM testing software. So if anyone has a good suggestion of what to use please dump it here.

memtest86 7.3
memtest86 4.07
memtest86+ 5.01
Windows 7 memory tester
HCI memtest
Prime95 blend mode
7-zip benchmark



Instabilty in the memory will be controlled by adjusting voltage as well as timings to unstable settings
I am thinking to use around 4 hours test time per iteration. anything taking longer than 4 hours will be determin as not faults found for the current iteration.
any other input on this is welcome.
 
use to live by memtest86 but these days its all but useless for finding anything but severe instability
i found the info here very usefull
http://www.overclock.net/t/1569364/official-intel-ddr4-24-7-memory-stability-thread
in the end i found the linux GSAT test by far the best for quickly finding memory related errors and HCI best for finding memory controller related errors
Thank you for the info im looking into it


I don't think 4 hours is enough for ram testing.
You are missing the point. We are not testing ram. We are testing the ability of software to find errors.once one iteration is done. the memory is adjusted to bee more unstable and goes on to next iteration.
4hours is plenty full enough because each iteration will make it easier to find errors. We just Need a time cut off line because we cant wait infinite time for an error that is not present.

or said in another way. after 4 hours of running the iteration without erros the iterations is judgeds inconclusive on its own because all software has shown the same. So we bump op the test rinse and repeated until a pattern emerges that shows which software shows errors first (or if at all)


I'ive also add'ed Intel XTU to the test
 
Goldmemory used to do the trick for me. At least back in the day when I was still into this.
 
Memtest86 is good for finding obscure errors if you're willing to wait 10-20 iterations (each pass changes the test pattern slightly--more passes=more coverage, for most of the tests anyway).

It's not a good test if you only plan to run one or two passes and be done with it, at least not on it's own. For just testing for minimum stability to boot an os, running moving inversions 16-bit, block move, and modulo-20 for three passes will usually pick up any instability, but I'd run more for peace of mind.

Thank you for your time, but did you even read the post?

1st: metest86 is already mentioned as on the test
2nd: You talk about minimum testing time when its already told the testing time is going to be 4 hours.
3nd: You are talking about peace of mind when we are not testing memory for error in a system here. we are testing the software abiltiy to detect errors

I appreciate you intensions but the lack of actually reading what is going on makes it just noise.
 
Anyway what i came into post was. that windows memory diagnostic is absolut horseshit of the worst kind and is not to be trusted for any kind of stability.
It ran 13 hours finding no errors in extenden mode on a system that was unbootable and had graphical clitches on the load screen from memory errors. This is the most uttersh#t Ive seen.
 
Well, if you run the three tests I mentioned it'd take less than 20 minutes. You could run them 12 times in 4 hrs which should give pretty good coverage. You should see errors well before then if there is any instability or a bad chip. Maybe that makes my post more relevant?

Of course if you don't understand the tools you probably shouldn't use them.

So i guess you way of dealing with you made a mistake, is now to try insulting people to try to sound smarter ?
is it really so hard to admit that you didn't read the post to begin with ?

It seems close to meaningless to test a software in some arbitrate configurations when the mission is to figure out which is the better software so proper recommendations can be made.
recommende software A but only if you adjust it in numerous way would be the very reason not to use it to being with since it does not do a proper job to begin with.
I might be curious enough to test it anyway, but I bet you, just running block moves test would find a error on this test settings before your method for the simple reason that, as stated in the original post, errors is made by adjusting timings, and block move is way more intensive for latancy issues than any other pattern in the test.

also to you information i was the one that got one of the memtest86+ releases withdrawn due to false positives and i have actively supported metest86+ development by providing defective memory for regression test of new release.

But yes lets argue i don't know the tool because heck pride right?
 
So i guess you way of dealing with you made a mistake, is now to try insulting people to try to sound smarter ?
is it really so hard to admit that you didn't read the post to begin with ?

It seems close to meaningless to test a software in some arbitrate configurations when the mission is to figure out which is the better software so proper recommendations can be made.
recommende software A but only if you adjust it in numerous way would be the very reason not to use it to being with since it does not do a proper job to begin with.
I might be curious enough to test it anyway, but I bet you, just running block moves test would find a error on this test settings before your method for the simple reason that, as stated in the original post, errors is made by adjusting timings, and block move is way more intensive for latancy issues than any other pattern in the test.

also to you information i was the one that got one of the memtest86+ releases withdrawn due to false positives and i have actively supported metest86+ development by providing defective memory for regression test of new release.

But yes lets argue i don't know the tool because heck pride right?
No, because you hadn't explained why it wasn't sufficient or otherwise sucked in general. I know this isn't the thread for that so I won't ask why. Suppose I'll just have to take your word for it. [Edit: I did read it, btw (unless you meant dasa's link, didn't follow that). I just wanted to say why I thought memtest86 was good for some things, if you set it with particular constraints in mind. Then you jumped down my throat as if I had insulted your intelligence (sorry if that was the case, not my intention)]

I'll go ahead and delete my prior posts since they clutter the thread.
 
Last edited:
some people say u need to test ram for long periods like 24 hours + plus but I personally feel thats overkill and I usually only do it for 1-2hours max
 
some people say u need to test ram for long periods like 24 hours + plus but I personally feel thats overkill and I usually only do it for 1-2hours max

I have seen several times at home and at work where testing ram required more than 8 hours. To me a single pass will not detect marginal ram.
 
I have seen several times at home and at work where testing ram required more than 8 hours. To me a single pass will not detect marginal ram.

I mean if there is a problem it will usually detect it quickly though as you say if you want to be 100% safe you may want to test for longer i agree
 
I mean if there is a problem it will usually detect it quickly though

I have personally witnessed at least 1/2 dozen times where this was not the case. In all cases took several hours to get bad ram (ram that crashed windows on boot or exhibited other instability that was easily seen...) to detect a single error in memtest86+.

In all cases after an RMA the new ram was fine and passed a much longer ram test.
 
Last edited:
No, because you hadn't explained why it wasn't sufficient or otherwise sucked in general. I know this isn't the thread for that so I won't ask why. Suppose I'll just have to take your word for it. [Edit: I did read it, btw (unless you meant dasa's link, didn't follow that). I just wanted to say why I thought memtest86 was good for some things, if you set it with particular constraints in mind. Then you jumped down my throat as if I had insulted your intelligence (sorry if that was the case, not my intention)]

I'll go ahead and delete my prior posts since they clutter the thread.

You did in facts try that "Of course if you don't understand the tools you probably shouldn't use them."

The problem, is you started up with a complete wrong assemnt what the thread was about which i picked up to begin with and even thank you for you effort but its was not what the thread was about i was not looking to test my memory but to test the memory tester ability to find errors
offcause now you have that posted deleted to even more hide that fact.
and then you came back with an indirect insult.


Anyway this is an effortless debate.


some people say u need to test ram for long periods like 24 hours + plus but I personally feel thats overkill and I usually only do it for 1-2hours max
We are not testing memory, that is not the goal of this. that why we do;nt need to go the the same extend to make sure our RAM is stable. Simple because we really don't care if the ram is stable or not. we care if the software picks up errors.
There is no need to search longer when we can just amplify the severity of the errors. so the pop up within the time frame we work with.



I have personally witnessed at least 1/2 dozen times where this was not the case. In all cases took several hours to get bad ram (ram that crashed windows on boot or exhibited other instability that was easily seen...) to detect a single error in memtest86+.

In all cases after an RMA the new ram was fine and passed a much longer ram test.
during this testing ive seen windows memory test run for 13+ hours in a none-bootable system. so yeah sometimes even 13 hours is not enough :D

also have memtest86 running for 21 hours in a system that has minor ram issues that can be detected within 1hours in another software.. im actually rather surprised of some of these results
 
Right, you thanked me and in the same breath accused me of not reading your post. How should I respond to that? My point was if you used it as i suggested it would find errors faster, and that not every diagnostic tool is just push a button and wait. I'm sorry about putting it in an insulting way, but you should reevaluate how you respond as well.

The content is still in your quotes, I'm not hiding shit.
 
Last edited:
For overclocking, what I did was installed stressapptest, prime95, and linpack onto a linux live USB.


http://www.linuxliveusb.com/en/download Create a linux usb with persistent data (I recommend ~1 GB)
I used the latest linux mint-KDE 64bit: https://www.linuxmint.com/download.php

Boot to the USB. Install new applications > search for stressapptest (or using terminal: sudo apt-get install stressapptest)
To run in Terminal: stressapptest -W -s 3600
If you get an error that it can't alloc memory right after installing, restart the pc. If you installed via the install stressapptest button on the community mint website and get a permission denied error, you can usually ignore that too.


https://www.mersenne.org/download/ to download and extract mprime for linux.
To run, use the file browser and right click on the mprime executable > Actions > run in terminal/konsole.
To stop use CTRL-C.



For linpack, download and extract: https://software.intel.com/en-us/articles/intel-mkl-benchmarks-suite
The only files you need are in benchmarks_2017\linux\mkl\benchmarks\linpack:
lininput_xeon64
runme_xeon64
xlinpack_xeon64
Make a backup copy of lininput_xeon64 if you want, then edit the original lininput file to match your amount of ram:
Code:
Sample Intel(R) LINPACK data file
Intel(R) LINPACK data
1 # number of tests
12600 # problem sizes
12600 # leading dimensions
100   # times to run a test
4   # alignment values (in KBytes)

# Change the problem sizes and leading dimensions to suite your available physical memory
#
# Memory use = 8 x (problem size)^2
#           or
# Problem size = sqrt (memory use / 8)
#
# Consider the following as a conservative guide
# For 1 GB use a problem size and leading dimensions of 6300
# For 2 GB use a problem size and leading dimensions of 12600
# For 4 GB use a problem size and leading dimensions of 18900
# For 8 GB use a problem size and leading dimensions of 25200
# For 16 GB use a problem size and leading dimensions of 31500
# For 32 GB use a problem size and leading dimensions of 37800
# For 64 GB use a problem size and leading dimensions of 44100
To run linpack, use the file browser and right click on runme_xeon64 > Actions > run in terminal/konsole.
If you get an error saying ./xlinpack_xeon64: not found
Right click an empty place in the folder, actions > open terminal here and run the command manually: ./xlinpack_xeon64 lininput_xeon64
if lscpu shows 2 numa nodes and 9+ cpu cores, you can add on the parameter: numactl --interleave=all


You can open another terminal to monitor for kernel errors before tests are completed: sudo cat /proc/kmsg
Check cpu frequency: lscpu | grep MHz
System Monitor lets you end processes if needed.
You can view a single snapshot of temps by opening another terminal and entering: sensors
I haven't found a good gui that monitors voltage or watts. Powertop has estimates over a long period, but it's more applicable to battery devices.

On linux mint-kde, Open Settings > System Settings > Desktop behavior > Screen locking > disable screen lock and hit Apply to monitor for errors.

If you find another program you want to compile, the distro I used was missing a few dependencies and this command fixes that: sudo aptitude install build-essential
Then you can download whatever source.tar and use:
./configure
make
sudo make install
 
Last edited:
just a quick update

I'm about 150+ 4+hours benchmarks in.
Im probably around 50% done if o restrained myself to just 3 systems ( 2 Intels and one amd system)
There are some memory testing software that is straight out horrible and are coming from pretty big companies.. but are absolut turds to find errors.
 
Back
Top