MemTest64 causes instabilities

[tj91]

I've been testing my new system for a few days, now. I'm running the Ryzen 3600 processor on an Asus Crosshair 6 Hero (x370) mainboard, BIOS version 7106 and 32 GB RAM (Crucial Ballistix LT Sport Red, 2 x 16 GB Dual Rank, 3200 MHz XMP).

When running MemTest64, my system usually becomes unstable. Programs just dont work correctly anymore and the system does not recover. Screens might become black, too. In contrast to <this thread here>, I've only allocated 8 out of 12 threads and only around 80% RAM. So it can't be the problem that Windows runs out of memory completely. Unless it's a weird fragmentation issue with the Windows scheduler/allocator.

I first thought my RAM is defective, but ruled that out. All other stress tests and memory test tools worked without any problem so far. To be completely honest, I can't exactly remember whether or not I had a freeze/crash/whatever at the beginning where MemTest64 was not involved. However, I'm quite confident now that I only have issues with MemTest64.

<I posted a longer version of this on Reddit>

Initially I Thought it might be a compatibility issue or simply just a BIOS bug since many issues get reported with the new Zen2 BIOS versions.

Now a friend of mine could reproduce the same behaviour with MemTest64 (also running the Asus C6H board but a Ryzen 1800X). His system has been stable for a long time now. So that settles it for me and I have to say that MemTest64 just is not stable.

One thing that I have to add: I've already ran MemTest64 for 8 hours straight, but was not using the computer at that time, mostly. So, MemTest64 only seems to be unstable when other programs are running. Also quite a few times the issues arised when I opened a program and a UAC request was shown to click "yes" or "no". I'd have a black background (instead of the desktop background), and I'd be able to click "yes" but then the system would be in a weird, instable state already.

 

[W1zzard]

Don't set it to allocate that much memory. That memory is effectively gone from other applications, so if you have several other apps open they will get paged out to the pagefile = run ULTRA SLOW, this includes the OS

This is expected behavior. MemTest64 doesn't even run with administrative privileges, it has no low-level hardware access, so it can't do anything that makes the system unstable.

Edit: you are running OCCT and MemTest64 at the same time?

 

[tj91]

Well I only allocated like 25 GB out of 32 GB. I'd say 7 GB is plenty for Windows to operate. MemTestPro for example doesn't show similar problems when I allocate that amount or even more.

Edit: you are running OCCT and MemTest64 at the same time?

Yes, I wanted to stress my system while testing the memory. However running OCCT is not a necessity to make the system unstable. It's just been fast to trigger it. Sometimes even before I start the stress test, e.g. while starting the program or more specifically while the Windows UAC request asks me if I want to start the program.

In general, if I do tasks (not too heavy on either memory or CPU, considering it's 4 threads and around 7GB of memory left), the system will usually become unstable after a while if MemTest64 is running. Even when the task manager shows 80-90% free memory. For reference, 90% means there is still 3.2GB left.

MemTest64 doesn't even run with administrative privileges, it has no low-level hardware access, so it can't do anything that makes the system unstable.

That's what I thought too, so this is even stranger.

Maybe the following has to do something with it? MemTest64 usually fails to allocate the memory in its first (or second, or third, ...) try, even though there is enough memory left.

Example - imaginary - run:

1. Around 30 GB free memory
2. MemTest64 tries to allocate 25GB, fails when it has allocated 8 GB
3. MemTest64 tries to allocate 25GB, fails when it has allocated 15 GB
4. MemTest64 tries to allocate 25GB, fails when it has allocated 20 GB
5. ...
6. MemTest64 runs.

It's not always like this and it's not always that MemTest64 allocated increasing amounts of RAM with each try. Sometimes it works at its first or second try. I'm not sure if this behaviour is expected and why it is occurring. Maybe this hints to some underlying problem, I don't know.

 

[W1zzard]

Check page faults with Task Manager, that shows you when Windows is making use of the page file

 

[tj91]

It's not like every page fault automatically is a read/write to the pagefile. However, I think I'm ready to let this rest and am not really looking forward to investigating the issue further.
If you are certain that it's expected behaviour then it might very well be. I was just not experiencing anything like it with e.g. MemTestPro and there always was free memory left for Windows to allocate. That's what keeps me wondering if something else is going on. Maybe the way MemTest64 works is triggering an edge case in Windows - Or it is as you say and it's simply expected behaviour

Thanks for your replies.

 

[pdrier]

(Yes, I know this is a 2 year old thread)

I'm seeing the same trouble with MemTest64 (v1.0) myself. More specifically, I can't get it to allocate even 60% of my memory even when I boot in safe mode, it will usually do 50% or less, though maximum mode rarely finds >10 on my 64gb machine even when otherwise idle.

Worse yet, MemTest64 can consistently crash dwm.exe with an Event 1000 problem. Typically with 20gb of memory still free at the time.

Note: I've run a full windows memory diagnosis, passed, and aida 64 memory stress test was able to allocate 54gb without any problems. So I'm relatively certain the memory itself isn't the problem.

So, I'm left wondering what MemTest64 is actually doing, how it's allocating memory, what "Memory locking failed (might be reserved by other apps/kernel)" means in technical terms (been developing sw for 35 years now..) and how the heck is it possible for a user app to crash the windows 10 desktop window manager process.

Any thoughts?

-Peter

 

[W1zzard]

MemTest64 can consistently crash dwm.exe with an Event 1000 problem.

First time I'm hearing about that. Anything that makes your setup special?

So, I'm left wondering what MemTest64 is actually doing, how it's allocating memory, what "Memory locking failed (might be reserved by other apps/kernel)" means in technical terms (been developing sw for 35 years now..)

if (VirtualLock((PBYTE)memory + locked, lockSize) == FALSE)
{
Log(_T("Memory locking failed (might be reserved by other apps/kernel)"));
VirtualFree(memory, 0, MEM_DECOMMIT | MEM_RELEASE);

and how the heck is it possible for a user app to crash the windows 10 desktop window manager process.

This shouldn't happen. Do you have a super high number of page faults when this happens?
Do you have a pagefile? On HDD or SSD?

 

[pdrier]

Anything that makes your setup special?

Not particularly.. 24 core AMD 3900x, 64gb ram, 2tb ssd, page file set to auto, win 10 up to date through Apr'22. Software developer, generally running: visual studio'22 x 2 or 3, chrome, a bunch of chat apps, market data, ...

Thread over on LTT with the various system details... https://linustechtips.com/topic/142...ry-used-feels-like-swapping/#comment-15354309

This shouldn't happen. Do you have a super high number of page faults when this happens?

Not entirely sure. When this happens Task Manager says I still have north of 15gb of ram free.. Then the whole screen goes blank and things freeze up for a good 5-15 seconds. When they sort of come back much of windows isn't right at that point (lost my night shade, taskbar doesn't refresh, most apps blank till I forcefully minimize and resize them). I did see once a large amount of activity on the page file after things came back, but it's hard to know if that's a before or after the blank screen given the latency in those stats.

Do you have a pagefile? On HDD or SSD?

SSD, 2tb, which is about 90% full, but still has ~190gb free. Currently about an hour after a restart, I have a 11gb page file, but I'm not sure why. Commit charge on task manager is 24gb.

if (VirtualLock((PBYTE)memory + locked, lockSize) == FALSE)

I was hoping that would have been much more fancy to explain what's going on.. I'm assuming you've done the relevant SetProcessWorkingSetSize bits so it's not hitting some sort of process limit. I don't suppose you're calling GetLastError by chance? That could improve the Memory Locking Failed message.

Some threads on the ether attribute dwm problems like this to video driver issues. I've clean installed them (RTX 2060) just in case. Really not sure how your app allocating/using memory could trigger video driver issues, let alone crash a system app.

I should add a few notes:

• I've been able to get MemTest64 to allocate up to but not over 32gb.
• The screen blanking bit happens more often than the dwm crashing shows up in the event viewer.
• The problems have happened with as little as 15GB utilized in the test, typically in the last step (which is the longest running time, which may be the correlation)
• I typically turn the threads down to 12, so the rest of my system doesn't (poop) during the test.

Cheers,
Peter

 

[W1zzard]

Did a quick test on a Ryzen, Windows 11, 64 GB machine. Same problem. Will investigate more over the coming days

 

[delshay]

Did a quick test on a Ryzen, Windows 11, 64 GB machine. Same problem. Will investigate more over the coming days

I needed to ask this question for some time.

Is it possible when a fault is detected, can a alarm sound be outputted via the normal sound output or via the motherboard speaker? This goes for all program that test ram AIDA64, Prime95, ect, ect,

ie any program that test ram.

The reason am asking is, i use a external hardware ram tester & when a fault occurs it sounds a alarm via the motherboard speaker. This saves a lot of time monitoring watching the screen for faults. I can then pause the program & say the very first fault appeared at a specific location first.

 

[pdrier]

Did a quick test on a Ryzen, Windows 11, 64 GB machine. Same problem. Will investigate more over the coming days

I have a second machine that's a Ryzen 5400G, Win 10, 32GB, with no video card.. And it's fine. As a third data point.

Cheers,
Peter