MemTest64 or MemTest86 ?

[80251]

They just do. It's a time-consuming process. There is no fast way to completely test large amounts of memory. What often happens is all of the RAM is swapped out for new RAM and then each stick is tested one at a time until the bad one is found. That's how we(the lab I work at) do it at any rate. This minimizes downtime. Our current servers have 384GB of RAM per server blade and the rest have 256GB or 128GB.

It already takes me 10 hrs. to run 4 passes of memtest86 v4.3.7 with my 32GiB's of DDR4 and that's only 4 passes! No wonder overclocking RAM is the most time consuming nightmare ever.

Can DDR5 at least test RAM a lot faster than DDR4? Say twice as fast?

Bill_Bright mentioned memory testers, if you buy one of those devices do you get the option to set the voltages, frequencies and all the timings for the RAM? Even if the RAM does pass at those frequencies though, that doesn't necessarily mean it'll work at those frequencies with an actual CPU+motherboard combo right?

 

[ThrashZone]

Hi,
If testing is for bad sticks just test one stick in single channel over night.
Repeat for the next until you're done that way it should be overnight and offline time and still do around 8 passes which really is the minimum not 2 or 3 lol

 

[80251]

8 passes of memtest86! Shhhhhhhhhhhhheeeeeeeeeeeeeeeeeeeiiiiiiiiiiiiiiiiiiiiiiiiittttttttttttttttt! That would take more than 21 hrs. for my 32GiB kit.!

 

[Bill_Bright]

Bill_Bright mentioned memory testers, if you buy one of those devices do you get the option to set the voltages, frequencies and all the timings for the RAM? Even if the RAM does pass at those frequencies though, that doesn't necessarily mean it'll work at those frequencies with an actual CPU+motherboard combo right?

I don't know if you can set voltages. The $2495 price tag (for the budget model) scared me away. My assumption is tests are only valid for the manufacturers published specs.

DDR5 might be a little faster than DDR4 (assuming same size) but no way twice as fast.

 

[ThrashZone]

8 passes of memtest86! Shhhhhhhhhhhhheeeeeeeeeeeeeeeeeeeiiiiiiiiiiiiiiiiiiiiiiiiittttttttttttttttt! That would take more than 21 hrs. for my 32GiB kit.!

Hi,
It's usually not the memory "unless it's corsair lol" and is more likely bios/ motherboard flaws or at least for myself
Cheap boards are cheap for a reason although same goes for cheap memory so best to buy good memory to start with.

 

[Halo3Addict]

How do people with 64 GiB or worse 128 GiB of RAM ever test it? It takes literally 8 to 10 hrs. to test 32 GiB of system RAM, I shudder to think how long it would take to test 64 GiB or 128 GiB of system RAM -- even with DDR5.

Currently testing my Dell R730xd with 256GB of RAM and looks like it's going to take about 2-3 days for 4 passes. I'll probably run it again before deployment just to be sure.

 

[droid-I]

They just do. It's a time-consuming process. There is no fast way to completely test large amounts of memory. What often happens is all of the RAM is swapped out for new RAM and then each stick is tested one at a time until the bad one is found

Right, if the RAM is in commercial use etc. it is damn important to see that the system(s) can run 24/7 & 365d. I have been in that area, and still have one foot in the mud, but it is not cheap.
But for a random PC overclock'er home user, if the computer does not want play nice, it will show anyway, and then start to look if the RAM is the culprit/cause.

Cheap boards are cheap for a reason although same goes for cheap memory so best to buy good memory to start with

This is the other thing, tech get outdated so fast, some want to switch for better performance every 2-3year or so, becomes costly too, in the long run (I see ppl are now running for eg. AM5 DDR5 just to keep up with the pace).

 

[freeagent]

I like TM5. I haven't used dos memtest in eons. I am a little surprised to see it still being recommended. I don't have anything against them, as I have used them before.

Isn't memory tuning fun

 

[lexluthermiester]

Yeah I've posted that one before and got slack around here saying it wasn't any good on ddr4

It's perfectly good for DDR4. Memtest86+ rocks.

to run 4 passes of memtest86 v4.3.7

That's a bit of an old version. Grab the latest version which is 6.10.

8 passes of memtest86! Shhhhhhhhhhhhheeeeeeeeeeeeeeeeeeeiiiiiiiiiiiiiiiiiiiiiiiiittttttttttttttttt! That would take more than 21 hrs. for my 32GiB kit.!

Sounds about right.

None are conclusive.

I've never had this experience. For 25+ years Memtest86 has worked perfectly to identify bad from good RAM. Of course I generally test one stick at a time.

 

[Bill_Bright]

I've never had this experience. For 25+ years Memtest86 has worked perfectly to identify bad from good RAM. Of course I generally test one stick at a time.

Well, I will not hold your obvious youth and lack of extensive experience against you!

I am not suggesting the problem is common. It definitely is not. But I have seen on a handful of occasions where RAM passed all tests, but still failed when put to practical use. And replacing the RAM (and only the RAM) with all new (and identical!) RAM cleared the problems.

Examples include a single stick passes all tests, but when running normal applications (like Photoshop, games, or videos), errors occurred. Replace the stick, no errors. Or two sticks (tested separately) pass all tests and individual sticks running normal applications have no failures, but when paired, errors occur. And again, replacing both sticks with new RAM cleared the problems.

Is there a chance the memory controllers on those systems just didn't like those specific sticks? I suppose, but that doesn't explain why swapping in different (but identical) sticks cleared the problems. It just adds to the inconclusiveness of the problem - to me.

I will emphasize again, if these software based testers report any errors, even 1, the RAM is bad. Period. So they have been very consistent and conclusive in that regard.

Having said all that, what are the alternatives? A $2500 tester? It is hard for most shops to rationalize spending that kind of money (even if they have the financial means) on a "unitasking" piece of test equipment. It would actually be much cheaper to have several sticks of spare RAM laying around just for testing purposes. That's what I do.

All I am saying here is "IF" your software based tester does not find any problems, "AND" you have eliminated all other possible causes of your computer problems, do not assume the RAM could not still be the issue. Swap it out, just to be sure - especially before having a stroke, or spending big money on a new motherboard, or CPU, something else.

 

[Calenhad]

Since Memtest86+ was updated for the first time in ages, I have gone back to using that. I used Memtest86 for years due to compatibility issues with the previous, ancient, version of Memtest86+. Honestly they do a pretty similar job, so it is mostly a personal preference.

Kind of a sidenote, AFAIK Memtest86 is a fork from the Memtest86+ source code (and there were some issues with naming and what not). So they were built from the same code some 15-20 years ago(?). This is one of the reasons I prefer the non-Passmark version. I like my open source software

 

[ThrashZone]

Hi,
Yeah after testing sticks in single channel second test is in dual channel
Using single channel is just a fast way to see if a single stick is bad since the test is a lot shorted plus if you get errors you'd have to do this anyway to see which one is bad.

But the way people mix memory without any hesitation that would also be separated and tested on it's own until last.

 

[Bill_Bright]

But the way people mix memory without any hesitation that would also be separated and tested on it's own until last.

To that (mixing memory without hesitation), with modern systems, I don't blame them.

Modern manufacturing techniques ensure, to a much greater degree, that all RAM (even from different makers/factories) coming off the production lines meet published specs with much more precise tolerances.

This is why RAM makers no longer test at the factory individual sticks, then match them in pairs (or trips) to sell as "dual" channel (or "triple" channel) memory "sets". These days, they package and "market" them as dual (or triple) channel sets just because it is cheaper and more convenient (read: profitable) to sell them that way.

At the same time, manufacturing techniques with motherboards and CPUs (memory controllers) have improved so significantly that they are much more capable of making slightly dissimilar RAM play well together.

So the old recommendation to ensure RAM is the exact same brand and model number no longer applies - at least not with newer RAM, motherboards and CPUs. As long as the published specs are the same they "should" be totally compatible.

 

[80251]

@Bill_Bright
Are there any triple channel CPU's anymore? I thought everything was dual channel or quad channel these days.

 

[Bill_Bright]

It is not triple channel "CPUs" but rather triple channel motherboards. And I am not aware of any currently in production. Dual and quad are, however. And there are still some triple channel still in the supply chain/retail outlets, but they definitely are not current generation technologies.

For HP IPMTB-TK motherboards LGA 1366 X58 DDR3 Triple channel M-ATX Motherboard 612503-001 - Newegg.com

 

[80251]

@Bill_Bright
I had thought the memory controllers were all located on CPU's now, not motherboards anymore?

 

[Bill_Bright]

This thread is about memory testers - software to test current and legacy RAM. Don't get hung up on the latest technologies when this software is also used to test legacy technologies too.

 

[Final Bulwark]

Always restore to stock. Never run any diagnostic utilities while overclocking. That is asking for problems. If you're OCing and you get errors, the very first thing one should do is set everything back to stock. If the problem goes away, you know your OC was unstable. If the problem persists, you know you need to test further.

Does that also apply to the Turbo boost frequency? My CPU is at stock settings - 3.4 GHz base clock and 3.8 maximum boost clock, with usual 3.5-3.6 GHz while gaming. I never touched the clock, but it is boosting when it has temperature headroom. Can that cause a problem while testing RAM?

 

[lexluthermiester]

Does that also apply to the Turbo boost frequency?

No. Turbo boosting is a part of the factory configuration for a CPU and such is a "stock" setting.

Can that cause a problem while testing RAM?

Not in any way that I know of, though I guess anything is possible. It's just very unlikely.

 

[chrcoluk]

HCI memtest
Karhu ram tester
Google stressapptest - probably the best one.

Better to test ram inside a booted OS.

 

[ThrashZone]

@Bill_Bright
Are there any triple channel CPU's anymore? I thought everything was dual channel or quad channel these days.

Hi,
Yeah I wasn't touching that one with a continental pole

 

[Bill_Bright]

@Bill_Bright
I had thought the memory controllers were all located on CPU's now, not motherboards anymore?

Hi,
Yeah I wasn't touching that one with a continental pole

Why not?

It seems some here assume no one would want to test their old RAM from their legacy systems. ???

There are still millions and millions of DDR3 users (even DDR2 users) out there trying to eke out a couple more years of use from their aging systems. And we see many, even here at TPU, who drag legacy systems out of their storage closets who are trying to bring them back to life. Triple channel architecture, especially among AMD systems was quite popular not too many years ago. Why would they not want to test their RAM too?

 

[ThrashZone]

Hi,
Hard to take you seriously when you state "modern systems" then go to "triple channel"

 

[Bill_Bright]

Hard to take you seriously when you state "modern systems" then go to "triple channel"

Sorry you were not able to follow what I was saying. I thought I was pretty clear - at least in my mind.

I was talking about how, in the old days, RAM makers used to test every stick before it left the factory. This was because failure rates were relatively high so they tested to ensure it worked. This was actually cost effective compared to processing warranty returns.

But also, for dual and triple systems, they would test and bundle sticks that most closely matched each other's specs. This was because variances from the publish specs were much wider (less precise). Also, boards (memory controllers) of the day required more closely matched sticks in order to enable and run in dual and triple channel modes.

But in recent years, RAM modules have become extremely reliable and manufactured to more precisely match published specs. So testing and matching sets (which is an expensive process) is no longer necessary. Consumers can just pick two (or three) individual sticks of the same model, and be confident they will work together just fine. However, buying prepackaged sets is typically less expensive - just like buying in bulk tends to be cheaper.

I note these "modern" days, it is very difficult to even find RAM that is not warrantied for life.

Regardless, as I showed above, you can still buy triple-channel boards. And folks may still want to test those sticks - as well as their older RAM.

So whether you take the facts seriously, or not, they are still the facts - even in these "modern" days.

 

[Final Bulwark]

trying to eke out a couple more years

? I'm not trying anything. I'm using my RAM for what I want just as it was used when it was brand new in whichever PC it was originally in. I just bought it last year. In my mentality last year is very recently.

aging systems

? I should introduce some nice skin cream for it...

drag

? Uff that's heavy for sure... around 20 kg is too heavy for me.. I had to ask my girlfriend to lift it since she lifts 110 kg when training anyway so it was easy for her ...

trying to bring them back to life

? Really.. hmm I'm still not "trying" anything. I'm just enjoying my PC which seems to be running for a week non-stop now and I'm playing most of the day every day... so... ?

back to life

Wow. I didn't know I was (among those people) performing resuscitation... .... I gotta say, this is some really fascinating terminology you are using on a hardware forum. I'm impressed so much by it I almost want scientific proof and elaboration for physical degeneration of this almost FOOD like structure of these biological(?) organisms you must be describing. Wow.

Since RAM sticks are physical components like any other physical PC component, this part of your comment really makes it sound (at least to me) as if (in this case) RAM or any other physical component, is like a piece of FOOD which SPOILS over time and thus somehow loses the performance it had. (???) Any component that failed in my life was from physical damage caused either by me overusing a particular part, or another particular part had a factory defect which manifested only after certain time in certain sub-optimal conditions. That being said, I still have my RADEON X800 GTO from 2005 and it STILL WORKS and RUNS Windows 10 and plays Cryisi 1. All I ever had to do to it was change the thermal paste. So where is this... this derogatory, belittling, degrading, depreciatory, denigratory, disdainful, almost scornful even, language you u are using coming from? By the way, I tested my RAM with Windows Memory Diagnostic first, and then with the software which lexluthermiester recommended. The results were flawless and with no errors. The reason I asked in the first place was because I very rarely had a random freeze every two weeks or so specifically more so when I had HWINFO running along with a game. It may be a incompatibility with the SATA 6GB/s port that my SSD was plugged in. I lugged it in to a SATA 3GB/s port where also the rest of the HDDs are (I assume it's another controller) and now no random freeze ever happened since then. RAM tests OK so I don't see why I should spend my money on "new" systems (CPU,motherboard,RAM) if this system is working great for what I need and love. I spent a sum enough to buy a new PC on books for example and on other aspects of life. I don't see why I should consider new hardware, which would also possibly have various random issues maybe even more than well used, well tested USED hardware, or approve of a theory that makes it seem as if components are the same material as FOOD and that their performance somehow magically "DEGRADES" over time. I didn't know that. So my DDR3 RAM is not as fast as it was when it was "new" anymore? How much horsepower did I lose from "aging" components hahaha, 50%? 100%? Maybe my RAM is not 16GB anymore but it "aged" and transformed into less? Haha (That was a rhetorical question obviously.)

even in these "modern" days.

"Time". What is "Time" and what is "modern"? Very subjective terminology of phenomenons born in superficial conscientiousness of a metaphor of what is a supposed to be a "human mind", that has no real meaning or existence other than in superficial consciousness of human paradoxical perception of life, clarity, and individualism.



Edit: typos. sorry.