HELP Wierd memory problem.

cymatty

cymatty

cymatty

Soldato
Joined
27 Sep 2004
Posts
13,305
Location
South Yorkshire
I have 2x512 Mushkin XP4000 memory and a dfi ultra NF4 D woth a venice 3200, now i am having major and strange problems with my memory.

Stick one

Won't pass mem test at ddr400 in the top yellow slot, it will trhow up errors in seconds @2.8v, it is now however running at [email protected] for 5 mins and counting without any errors. Edit it is now showing errors after 50 mins, putting all back to stock and trying this stick again.

Stick two

Wont pass mem test at any speed in the top 3 slots, took it round to a friends it ran in his machine for 10 passes of mem test without problem @2.8v DDR400

Whats going on?
 
Last edited:
Try the orange slots first mate. I've found they give better stability.

Doh, just read your comments again. Hmm, in that case, I don't know :p
 
Last edited:
Dutch Guy said:
What test produces the errors?
What BIOS version are you using?
Click to expand...

At the minute i am ignoring mem test, it is passing prime at 260mhz for 7 hours (without failing) runnig pi fine, running 3d mark fine and also playing games fine.

I suspect it is a problem with mem test rather than the ram. :confused:
 
Matty can it do 8+ hours of Prime95 Blend test (uses lots of memory)?

My setup at 8 x 333 HTT (2664 MHz) can do 8+ hours of the Small FFTs and In-place large FFTs with Vcore at 1.55v but for the Blend test it needs 1.65v (1.5v + 10% in BIOS) to do 8+ hours.

I came to the conclusion the Blend test was failing with Vcore at 1.55v due to the memory falling over (system freezing requiring a reset) or the CPU's memory controller falling over and the problem occurs around 25 minutes into the test.

To try to find the cause I did a Memtest86+ test with Vcore at 1.55v and Vdimm at 2.5v (minimum setting in BIOS) and errors showed up on test 6 so I progressively increased Vcore one step at a time and saw the number of errors reducing until 1.65v where only one error occured each pass. I know it seems crazy to increase Vcore to fix memory errors but I was guessing it would help and seeing as the errors reduced I guess I was right. I then switched back to 1.55v Vcore and started increasing the Vdimm and at 2.7v no errors occured in 5 passes of Memtest86+.

So in the end I learned 1.55v Vcore and 2.7v Vdimm results in Memtest86+ stability but for Prime95 Blend stability I had to increase the Vcore to 1.65v which suggests it was the memory controller falling over rather than a problem with the memory.

I'm amazed at how well the Freezer 64 Pro performs with my DFI (the heatsink orientation results in the hot air being sucked out the case instantly via the rear case fan) especially with Vcore at 1.65v and my DFI overvolts a little at full load. I even got the temps down some more with the help of an extra case fan so now it idles at 30c/36c/31c/40c (system/pwmic/cpu temp/on-die thermal diode temp) and at full load it's 32c/44c/49c/61c and my room temperature is 23c.

My next testing will be the 24 hour Prime95 Blend test and I'm reasonably confident it'll be 100% stable.

Have you tweaked any of the memory settings in the BIOS and also can you post an A64Tweaker screenshot of your overclocked setup please mate so I can see what settings you're using.
 
Last edited:
str said:
Matty can it do 8+ hours of Prime95 Blend test (uses lots of memory)?


Have you tweaked any of the memory settings in the BIOS and also can you post an A64Tweaker screenshot of your overclocked setup please mate so I can see what settings you're using.
Click to expand...

I ran the blend test for 7 hours yesterday and it passed fine, i will post a pic with A64 tweaker tomorrow when i am next on my rig. :)
 
Nice result on the blend test! :D

I've read about some CPUs or motherboards not liking some types of ram but your RAM isn't ever mentioned as a problem on any lists I've seen so unlikely to be that. So yeah that's a weird problem.

It could be your DFI isn't choosing good default memory settings and I'm assuming you haven't tweaked those settings but maybe your friends PC had different default memory settings causing it to pass the tests or possibly he had them configured optimally rather than at defaults?
 
An error in one of the first four tests means it is VCore or CPU FSB related, not memory related.

What you can do if you do't trust Memtest86+ is to run Prime95's Small FFT test as that stresses the CPU
 
Dutch Guy said:
An error in one of the first four tests means it is VCore or CPU FSB related, not memory related.

What you can do if you do't trust Memtest86+ is to run Prime95's Small FFT test as that stresses the CPU
Click to expand...

Well i ran mem test at default settings ie stock straight from the box, and it still failed straight away, but the other stick passed at 3.4v with the ram @ 265mhz and the cpu at 2650mhz.

If there was a problem with the cpu would it pass the blend test for 7 hours, amd also pass pi and 3d mark without even a slight hiccup? :)
 
cymatty said:
If there was a problem with the cpu would it pass the blend test for 7 hours, amd also pass pi and 3d mark without even a slight hiccup? :)
Click to expand...
Possibly, when testing my overclocks I found that I could run Prime95 blend for an hour yet it failed after 5 minutes on Small FFT's due to too little VCore.

Also SuperPI and 3DMark are not really a stress testers as Prime95 and Memtest86+ are.

But the bottom line is that if you are happy with the stability of your machine it's fine, the only time you would want (need) a fully stable machine is if you do Distributed Computing.
 
Dutch Guy said:
But the bottom line is that if you are happy with the stability of your machine it's fine, the only time you would want (need) a fully stable machine is if you do Distributed Computing.
Click to expand...

I agree with you whole heartedly there, i mean there are people who will say it is not stable blah blah, but it is not used for distributed computing or anything else important. I have a laptop for general web surfing and a file server for storage, my desktop is only used for a few (2-4) hours at a time maybe a couple of times a week for when i need something more powerful for a task or to play a game on.

If the machine starts playnig funny buggers and crashing and blue screen of deathing, then yes of course i will look into it further to see what the problem is, but if it can run for the length of time i need it for without glitch how can i say it is not stable, because for me it is. :)

I will at some point run the test you suggested and see what the results are Dutch Guy, and thank you for your help, in a side note would sis soft sandra burn in test (cpu and memory tests only) be a good check for faults, as i thought that would really stress a system?

Thanks again.
 
cymatty said:
in a side note would sis soft sandra burn in test (cpu and memory tests only) be a good check for faults, as i thought that would really stress a system?
Click to expand...
I have no experience with that, perhaps that only heats up the CPU when Prime95 checks the calculations it is doing and reports it if it finds an error.
 
Dutch Guy said:
I have no experience with that, perhaps that only heats up the CPU when Prime95 checks the calculations it is doing and reports it if it finds an error.
Click to expand...

I think the burn in test keeps running the cpu and memory tests from sandra (or whatever tests you chose) and runs them on cycle over and over again. :)
 
Have you configured Prime95 to priority 10 (Advanced menu -> Priority) as a means of severely stress testing? Might be worth trying the Small FFTs test with that priority level if you haven't already.
 
prime small.JPG


Small fft's at priority 10. :)
 
It works! :D

Thanks for the A64 Tweaker shot. My settings are a lot different but I'll experiment in a day or so with your settings to see if I can get to 2.8GHz (my setup can do 2.8 GHz with 1 stick of Crucial ram but it doesn't even boot past the BIOS at that speed with my Samsung ram).

I'm guessing the CPU memory controller doesn't like my Samsung ram but who knows it might with your settings. :)

It's all cheap stuff I've got so I shouldn't expect much from it anyway but it's still fun experimenting with it. :D
 
You must log in or register to reply here.
Back
Top Bottom