[NinjaZX6R]

OK, this is bugging me. At 270mhz on my PC3200 OCZ Rev2 platinum, I can pass 10 tests of memtest #5 at 2.5-3-3-6-1t. However, I go to run 3d mark, and it crashes on the first test. So, I loosen the timings to 2.5-4-4-7-1t and it can run 3d mark endlessly. So, it's the ram...but why the heck does memtest say things are all fine and dandy when I can't run a single 3d benchmark? This makes me think memtest is useless....anyone else had this problem?

-Collin-

[{PMS}fishy]

Yea, I have. 3D is probally the hardset stress that the computer endures.

I can pass test 5 for 8 hours, just to load up 3D03 and have it crap out on me half way though wings.

I don't really bother with memtest anymore. Try SuperPi, seems to give about the same results in terms of stability as 3D. I also find that your 2d stability will always be at higher clocks than 3d, but I think everyone knows that.

[HousERaT]

One thing I've found is Test 5 is a good indicator. However Test 5 + Test 8 is an even better indicator. I've memtested Test 5 for hours without error only to try Test 8 and get errors after a few passes. If you can tweak the timings enough to get both tests stable without errors, I've had good success in windows afterwards. A really good quick test would be 20 passes with Test 5 then do 10 passes with Test 8.

[Sentential]

I agree, memtest has been slacking off considerably as of late. Not sure why. Same applies to the infamous P95

[Super Nade]

Collin,

Memtest is a rough indicator of stability. One may think of it as a first line of defence. Surely, all of you who like to bash memtest should know that!

If you cannot pass memtest, everything else may fail. What is memtest? You have blocks of data being read and written to different memory locations. If the checksums don't match, you get an error. Quite simple. It's sole purpose is to see if blocks of data can be written and read faithfully. There is no CPU intensive operations being performed. With 3D gaming/testing, we are stressing the FPU to the max. Also, the Video card comes into play. Now you have 3 variables i.e CPU,GPU and RAM. Are their individual tests mutually exclusive? Experiment tells us NO. You also have crapdows added to the mix.

So, really, there is no point in blaming memtest as you are attributing more to it, than it was designed for.

I believe the modulo 20 test (#8) are the most important. In my experience, I have yet to come across a case where my RAM passed test 8 and failed test 5.

[HousERaT]

Originally Posted by Super Nade

I believe the modulo 20 test (#8) are the most important. In my experience, I have yet to come across a case where my RAM passed test 8 and failed test 5.

My research is demonstrating the same thing. I've had memtest sessions in which I could pass Test 5 and not Test 8. However I haven't had a pass on Test 8 that couldn't pass Test 5. Not that it can't happen, it just hasn't happened to me.

[deception``]

Simply put, Memtest stable does NOT equal Windows stable. On the contrary, it is simply a good way to begin your overclocking ventures.

In my case, I simply use Memtest to benchmark a particular speed before I go into Windows. This way, I am able to rule out certain speeds before I go through tons of in-depth testing. Once I complete a couple of Test #5 passes (about 15-25), I then give it a shot in Aquamark3, 3D Mark, and SuperPi (probably the best test out of the three). However, I am also very realistic about my scores, as I recognize that I may have to do further adjustments once I hit Windows. For example - I can Memtest all day @ 300 Mhz 2.5-3-3-7, but I have to lax my timings to 2.5-4-3-7 to achieve 100% stability.

In the end, Memtest is a good starting point, but it's not everything - just like Prime 95.

deception``

[Sentential]

Originally Posted by deception``

Simply put, Memtest stable does NOT equal Windows stable.

I couldnt agree more. These soft errors are a real bitch.

[Eldonko]

Originally Posted by NinjaZX6R

OK, this is bugging me. At 270mhz on my PC3200 OCZ Rev2 platinum, I can pass 10 tests of memtest #5 at 2.5-3-3-6-1t. However, I go to run 3d mark, and it crashes on the first test. So, I loosen the timings to 2.5-4-4-7-1t and it can run 3d mark endlessly. So, it's the ram...but why the heck does memtest say things are all fine and dandy when I can't run a single 3d benchmark? This makes me think memtest is useless....anyone else had this problem?

-Collin-

Same experience here. I used to have to same RAM and it would run error free in memtests (all tests) for 24 hours+ at 310Mhz. However the max it would do stable in windows was 283Mhz @ 2.5-4-3.

Now I only use memtest when I get new RAM. After I have an idea of its limits I test in windows with clockgen, tweaker, and superPI 32M.

[tresmonos]

I too, only use memtest as the first step in deciding a stable overclock. Passing memtest, getting into windows, runing superPi then prime95 is the sequence of testing I use to determine if my pc is going to crash midway through a css game or solidedge design. And I've been in your shoes with my OCZ plat rev 2. I can get 260 at 2.5-3-3-6 prime stable. Anything beyond that, prime refuses. Memtest will run overnight at 275...

[tresmonos]

Oh, almost forgot:
Its really nice to see you back with AMD, NinjaZX6R!

[Reefa_Madness]

What running Memtest will do for you is to give you some assurance that if the ram passes the test it will not corrupt your hard drive when you go into Windows.

It is the high end of any possible overclock, your actual overclock will most surely be less.

[NinjaZX6R]

Wow, it seems uniform that people START with memtest, and end with 3d tests in windows. I guess I just got the wrong idea of memtest. I was under the impression that Memtest was the end all be all of an overclock. Looks like 3d stresses it much more!

The input is much appreciated guys!

-Collin-

[deception``]

Originally Posted by NinjaZX6R

Wow, it seems uniform that people START with memtest, and end with 3d tests in windows. I guess I just got the wrong idea of memtest. I was under the impression that Memtest was the end all be all of an overclock. Looks like 3d stresses it much more!

The input is much appreciated guys!

-Collin-

No problem. Sadly, no such program exists when it comes to measuring stability of ANY kind. It takes a fair amount of time and patience to conclude whether or not your system is stable. If you need any advice on timings, drop me a PM.

deception``

[Big Lar]

I used to use a proggy called Stability Test, It had different levels you could choose, not unlike Prime95, but for some reason, if I passed that, I could do anything with my setup without changes to the setup, and never fault.

Edit; Spelling

[Category 5]

I have been wondering if voltage has somethign to do with this phenomenon. I notice on my Powerstream that the light for 3.3 is lit red in memtest, and sometimes in windows at idle, but drops to green when the system is under load.

This leads me to belive that when other system components rae heavily loaded, perhaps there is ram voltage droop causing the subsequent instability.

Of course it could also have to do with RAM temperature too.

[Silver]

I am very happy with memtest86. It allows me to ballpark the ram rather quickly, adjust timings and voltage to attain a very good ballpark setting. Clockgen, along with superpi 32 and a benchable setting is attained. I do alot of folding these days and the peak settings are usually off by a couple mhz but definately ballparked rather quickly. 10 passes? A little on the light side as frequently errors occur in the 15 to 25 pass area.

Memtest was of great help to me in getting my xms 3200 to 250 1:1 at 2,2,2,6. It would have taken a good long time to do without it. Folding 24/7.

[felinusz]

Memtest86 is a great and invaluable tool, but by no means a "be all, end all" to memory stability. I find it invaluable for "scouting out" timings and settings for optimal stability, especially with it integrated into the motherboard's BIOS. 3D stability is a common scourge of an otherwise clean overclock, and always needs to be taken into account, memtest86 aside.

I use memtest86 for stability testing, and I use Prime95 for stability testing, many people don't like or trust these two programs - I personally firmly stand by them.

But there's no way on earth I would trust them as an indication of proper and thorough stability by themselves!

True stability involves an overclock passing memory specific, processor specific, and GPU specific/blended stability testing. For me, that means Prime95, memtest86 tests #5/#8, and the 3DMark2001 Nature test looped.


I can generate overclocks on my current machine that will pass any memtest86 test for ~20 hours, but fail Prime95 or SPi. I can generate overclocks that will pass Prime95, but lock the 3DMark2001 Nature test. I cannot generate an overclock that will run unstable after passing all three

My two cents.

[HousERaT]

Originally Posted by deception``

No problem. Sadly, no such program exists when it comes to measuring stability of ANY kind. It takes a fair amount of time and patience to conclude whether or not your system is stable. If you need any advice on timings, drop me a PM.

deception``

If ever my thoughts were crystalized in a statement this would be it. There is no quick fix to getting a computer overclock stable in your OS. It takes time and lots of testing.

I firmly agree with felinusz on this also, as it takes a few programs to determine stability. If you can torture your computer with benchmarks, tests, burnin programs and other useful applications in situations that go way beyond normal usage and still maintain a stable overclock, you can feel pretty good that your computer is stable enough for normal usage.

[Super Nade]

Hints to using memtest in the last post here:-
http://www.ocforums.com/showthread.php?t=391391