Memtest shows 1 error on the 20th pass

vraa

Gawd
Joined
Mar 20, 2007
Messages
598
Who do you buy your ram from and how many errors do you have?

If ZFS needs a lot of ram, why not just max out mobos, ram is supposed to be making the largest difference?

Thoughts?

Happy Thanksgiving!
 
I have zero errors. A single error on any pass (even in 1000 passes) means that there is a hardware problem which could be the motherboard, CPU, memory or powersupply or of course the user. And what I mean by the user is that some users overclock their ram to crazy speeds on systems like an lga1366 which supports 1066MHz RAM max or AM3 and lga1055 which both support 1333 MHz RAM max. These OC profiles that memory manufacturers may not work on all configurations especially if you fill all slots.
 
I either get it from a local place that gives 100% cash refunds and test it like crazy, or buy Crucial stuff that has no heatsinks and is rated for standard voltage (1.50V or less for DDR3, 1.80V for DDR2). Don't trust anything if you can't see the actual chip manufacturer's markings. Actual chip manufacturer = Micron, Samsung, Nynix, etc., not Kingston, G.Skil, Corsair, etc.

If I got 1 error in 20 passes, I might accept the memory if it was rated for standard voltage AND it showed zero errors during another 40-100 passes at standard voltage, including many passes with a different diagnostic, like Gold Memory, and with another type of motherboard.
 
Last edited:
I don't accept anything that can't pass 1000% on HCI's memtest. It usually finds errors that standard memtest does not. But I also do computations that take weeks, so a crash/error is entirely unacceptable.
 
When the OP says "20th pass" exactly what is that implying or pointing out, because memtest86+ (and similar apps) do passes that complete in a full cycle, so... I run memtest86+ for 3 complete cycles and that takes some time even on the fastest machines I've had access to. One defective bit anywhere in those 3 cycles means the RAM isn't to be trusted.

As for running a Windows-based RAM testing app that is just a bad bad idea. I know a ton of people out there still do it but come on, that's just crazy stuff in so many ways I can't even begin to explain it.

The only true way to do a memory test is outside of any potential software related issues you can possibly remove from the testing process, and even though apps like memtest86+ (pretty much the standard) do that and work on the lowest possible level using the least possible amount of code they can, doing it inside Windows is... well, it's crazy. :)

The best potential would be a purely hardware-based RAM tester and those do exist but the cost involved is excessive just for a few sticks - unless you were a RAM retailer you wouldn't front that kind of expense.

Anything like memtest86+ that can run basically on top of the hardware itself without an underlying OS and drivers and other apps and all that other crap getting in the way - any of those having 1 problem can translate into false reports of RAM issues - is the way to go. You have to test the RAM from the lowest possible level from the software/code perspective which means not from inside a running high level OS like Windows, or any modern computer OS for that matter.

3 complete cycles of memtest86+ is more than enough for nearly every single person that bothers to go that far. Doing it for even longer periods of time - especially in situations where the hardware is pressed past the design specs (aka overclocking, etc) is a waste of time I'd have to say.

But, some people love doing testing and more testing then testing even more - seems people nowadays spend more time benching and testing for "stability" instead of actually using the computers which work just fine on spec. ;)
 
There's also something else to consider:

I used to have 2 x 2 GB OCZ sticks that could run memtest for 2 days without any errors. But if I ran memtest again after a SOFT reboot (programs starting to segfault due to errors), it would find at least 10 errors per run. Then the following day I could run memtest for hours without errors. I ran them at 1333 mhz and the case was well ventilated so temperature wasnt an issue.

I was able to reproduce that every single time. I'm sure its a 1 in a million hardware conflict and I can't find any explanation for it.
 
As for running a Windows-based RAM testing app that is just a bad bad idea. I know a ton of people out there still do it but come on, that's just crazy stuff in so many ways I can't even begin to explain it.

I was in that camp for years but now I recommend windows memory testing because it finds problems faster. I have seen this for myself. I think one of the reasons is that memtest86+ only tests ram in 1 thread at a time and does not stress the CPU or other hardware while windows memory testing the OS does its usual stuff. And with the free hcl memtest you need to run 4 or so copies of it anyways. I have seen cases at work where memtest86+ took a 72 hours (> 100 passes) to find a single error while booting windows or even windows and linux the problem was immediately evident.
 
Last edited:
Memtest is pretty good. No errors are acceptable really, a single recoverable error might in theory be acceptable after about a year of testing.

I use SEC or Hynix orginials or ATP these days. Bad experiences with unbranded, Crucial and Patriot.
 
3 complete cycles of memtest86+ is more than enough for nearly every single person that bothers to go that far. Doing it for even longer periods of time - especially in situations where the hardware is pressed past the design specs (aka overclocking, etc) is a waste of time I'd have to say.

But, some people love doing testing and more testing then testing even more - seems people nowadays spend more time benching and testing for "stability" instead of actually using the computers which work just fine on spec. ;)
We wouldn't have to do so much testing if the memory module companies would did more of it.

How long does 3 complete cycles of MemTest+ take? Because I had a 256MB or 512MB module that ran fine for the first 4-5 hours on MemTest86.
 
Decades ago it might have been reasonable to run memory tests.

Now, it is just a waste of time.
 
Now, it is just a waste of time.

Why? Memory still comes up DOA. Also since a lot of users here overclock their ram controller by buying ram that is faster than 1333 MHz which is not guaranteed to work on any current AMD or Intel processor. On top of this when you fill all ram slots memory may need tweaked to run.
 
Back
Top