Memtest versus RAM "stressing"

[Aenra]

Help this peasant comprehend pretty please.

I understand what memtest does (or so i think).
What i do not understand (as i have not seen it specified anywhere i've looked) is the RAM stressing testing.. be it with Aida x64, Intel XTU, or whatever.

If one passes two, three full passes of memtest, is OS-based RAM stressing still necessary?
Put differently, what's the difference between the two?

 

[jboydgolfer]

That depends whether the stress test stresses all factors of the ram, and all levels, like a memtest. meaning the cache etc...

 

[Aenra]

That depends whether the stress test stresses all factors of the ram, and all levels, like a memtest. meaning the cache etc...

How would i know that, lol...
Am already running it, just to be sure i'm stable, i'd just like to know why i'm running it!

edit: while i'm at it.. am currently stable at 3250MHz with 14-14-14-34, command rate at 2, all four sticks plugged (RAM is in my specs). Is it worth it trying to get the timings even lower? I don't do benchmarks mind, "worth it" as in my getting any RL benefits.
(in case you're wondering, my thinking was that as it turns out, running it at 3375+ put too much stress, IMc and voltage-wise. So decided to take it one click down to 3250 and start pushing the timings. Hopefully that's the smart thing to do?)

 

[natr0n]

Play games to test ram stability.

 

[Aenra]

Play games to test ram stability.

Seriously? And what would i be looking for?

 

[natr0n]

Seriously? And what would i be looking for?

For example a well coded fast racing game like Dirt 3.

Play something you are familiar with and if the game feels off or crashes then you know what to tweak/do. That should make sense I hope.

 

[Vya Domus]

Memtest is working to determine if there is a clear hardware failure which results in a error when doing low level types of operations in different patterns. Meanwhile tests like the one in AIDA64 typically are doing pretty basic stuff like allocating all memory available specified by the OS. Best stability test is just using your computer. Basically my point is unless you've got clear issues that point to bad RAM , it dosen't make sense wasting hours of testing.

 

[Aenra]

hmm.. looks like they -do- test different things, or at least they do the same tests differently?
Seems like i rushed in declaring my RAM as stable

Memtest could run forever, zero errors, tried Intel's XTU testing (yes, i use it, no, i won't use Prime; for my needs, i sincerely believe even XTU is extreme) for RAM only, got a watchdog violation in less than 15 minutes, lol
Increased VCCIO (was a lot lower than the advised VCCSA - 0.05volts) and started XTU again. One hour so far with no fail.. and about 393746452729 hours to go..

What i wonder now is if, assuming it was a low vccio that caused this/i pass the test, i need then lower the RAM vtt/vpp. This is a proper headache. Am consoling myself with the fact that this is the last bit, after it i will be finally done. Hopefully. All things being relative that is.

 

[Aquinus]

The problem is that the different parts of memory that are stressed depends on workload and the OS. For example, a bunch of memory accesses over a 32MB space is going to result in a lot more cache hits than misses and will stress cache more than memory. MemTest is going to be better for testing stability because it's going to run memory operations that are highly repetitive. With how DDR designed, it actually can execute these kinds of commands faster than ones that aren't repetitive which will put the most stress on DRAM because it's capable of issuing the maximum number of operations at a given time since no addition latency is being added by having to switch to different operational modes like read to/from write or, switching which bank is active, stuff like that.

So with that said, memtest is very likely stressing RAM in ways that can't be done when running a full blown OS because of all the magic that happens to handle paging and caching.

a watchdog violation in less than 15 minutes

Watchdog timer violations are related to core instability, not IMC or memory instability iirc.

 

[jboydgolfer]

If this is in regards to the operating system listed in your system specs I do recall reading a thread that w1zzard created in regards to a memory testing program he wrote which he was looking for testers for ,you might be able to find what you're looking for there with his program ... possibly

For what it's worth, more years than I care to mention worth of experience , has shown me that if a problem with your memory exists you won't have to look for it. You'll either get performance issues ,crashes ,or a message on shutdown

 

[Aenra]

Watchdog timer violations are related to core instability, not IMC or memory instability iirc.

Really? :S
Man i've had it run for 24hrs straight, core and uncore OCed, RAM at default 2133. Passed it..
It only gave me one now, when i changed the RAM frequency/timings; if you're right, i'm officially worried, lol, thought i had the other two set. Shit.

You'll either get performance issues ,crashes ,or a message on shutdown

Don't have your knowledge to judge, but this does sound reasonable. Only reason i'm bothering is that (hopefully) i can iron such issues out now, rather than needing to at some point in the future and probably in an opportune moment, such as say my getting a crash while i'm working.

If i may add a question, assuming i get it stable with all three OCed, do i need then clear CMOS, load optimized defaults and then re-enter my last values? Read various comments regarding RAM tweaking and the need to clear cmos when we change them -again-, but i'd rather ask here first.