Intermittent fault on X48 - F3-128000CL7-4GBXM

[Gskilldogz]

Hi all,

I'm getting Intermittent faults on my config, the motherboard is a P5E3 Pro (latest BIOS), RAM is F3-12800CL7-8GBXM (Ripjaws X - 2 sticks of 4Gb).

Although I found a way to stabilize my CPU OC to 3.6Ghz on prime95 Blend (24hours) I discovered I was still experiencing random crashes.

I put on HCI Memtest Pro, and found an Intermittent fault that occurs after hours of testing (Something like 6-7 hours).

Any suggestion would be gladly welcome !


Here are my settings

Ai overclock tuner : [Manual]
CPU ratio : 9
FSB Strap to Northbridge [333]
PCIE Frequency [100]
DRAM Frequency [DDR3 - 1600 Mhz]
DRAM Command Rate [Auto]
DRAM CLK Skew Channel A [Auto]
DRAM CLK Skew Channel B [Auto]
DRAM Timing Control [Manual]

CAS Latency 7
RAS to CAS 8
RAS Pre Time 8
RAS Act Time 24

RAS to RAS Delay [Auto]
REF Cycle Time [Auto]
Write Recovery Time [Auto]

Read to Write [Auto]
Write to read S[Auto]
Write to read D[Auto]
Read to Read S [Auto]
Read to Read D [Auto]
Write to Write S [Auto]
Write to Write D [Auto]

Dram Static control [Auto]
Dram Dynamic write control [Auto]

Ai Clock Twister [Auto]
Ai Transaction Booster [Auto]

CPU Voltage [1.33750]
CPU PLL Voltage [1.50]
FSB Termination Voltage [1.20]
DRAM Voltage [1.60]
NB Voltage [1.39]
SB Voltage [Auto]
Clock Over Charging Voltage [Auto]

Load line Calibration [Auto]
CPU GTL Voltage [Auto]
NB GTL Voltage [Auto]
PCIE Spread Spectrum [Auto]

[Gskilldogz]

Anybody help ? This error occurs only after a few hours of testing.
Thanks

[Tradesman]

Try NB at 1.41 and raise DRAM voltage to 1.63

[Gskilldogz]

Still experiencing some bluescreens.

Like you told me, I raised vNB and vDRAM by 1 step, and tried again. Still the same problem, it's been months now. BUT, my Cpu is OC and stable at 3,6Ghz, 24h prime, OCCT...

Here are my settings, I got to the point where the vDRAM is 1,70 and vNB is 1,41....

Any help Mister Tradesman ? What are vNB and vDRAM max safe (24H) voltages ?

[supahos]

I would back the dram voltage back down to 1.6, 1.7 seems a bit steep for long term use.
Maybe bump the NB voltage to 1.42 or .43 though I think your best bet is dropping the timing of the ram to 8-8-8-24, that may not fix it but it just might. Also sometimes on the older processors A stable OC of the CPU can still cause things like memory controller instablity if its maxed as well. Might back down a tad on the CPU OC and see if that fixes it as well.

[Gskilldogz]

Thank you for your quick response Supahos.

I'll bump the vNB for a start, and I'll get back to 1.6v for vDIMM, before loosening the timings.

will keep you posted very quickly.

[supahos]

going from 1.6 to 1.7 won't help stability at all, but could help the memory not fry like bacon . that is a 1.5v kit so I would think if you're not OCing it (which you're not) that going above 1.6v won't fix the issue.

[Gskilldogz]

I wanted to make sure it was the northbridge voltage causing problem, in order to do so, I went back to low voltage and yes, I got my answer quickly :
at 1.31 vNb : Error after 3 hours

Now I knew that with 1.33 vNB, it achieved 12 hours without any errors.

Let's see with 1.35 shall we ? Will tell you the results !

[Gskilldogz]

Getting nervous, it will be 6 hours without a single fault :-)

I know HCI memtest is really good to test RAM but I don't know how much coverage % I should achieve to make sure I won't get any Bluescreens.

[supahos]

Assuming your cpu cooling is good enough to handle it, I think prime95 blend test is probably the most taxing on the setup, it maxxes out the computer and runs some pretty memory intensive stuff as well. I would think if you passed 6 hours faultless however, that you SHOULD be perfectly fine. That is stable enough that it would likely be a driver crash that caused a blue screen more so than memory fault.

[Gskilldogz]

Hello Supahos,
I cannot agree more on the fact that Prime95 is the best for CPU stress testing.

I've achieved 24h on blend test, however, it would not validate the fact that the ram is stable. I found that Memtest86+ also achieved 12 (long) passes !

However, HCI was able to find some errors.

I'm really glad to say that that all these softwares helped me to find that it was the VNb that was causing the problem.

[supahos]

The regular Prime test is almost purely a cpu test, there is a prime BLEND test that will use as much memory as you tell it to use and stress it to the max at the same time, if it survies the blend test for a couple of hours I'd call it stable.

[Gskilldogz]

Funny I must do something wrong with my Prime95, only stress a few % of the RAM in blend mode

What settings do you have to change so that it stresses 100% of the RAM ?

[supahos]

You can set it to do a custom test (at work so I don't have prime in front of me) and input how many MB of memory you want it to run. Stop maybe 2.5 gigs short of your whole set, if it ties up the ammount being used to run windows it will either 1) insta lock, or 2) start using the hard drive for virtual memory.


as my sig says I have 16 gb memory and have managed to get blend to run at ~ 87% total memory usage, any more things go weird but I think windows doesn't like it, never seen a fault in any other memory testing software. Of course in the real world, the odds of having 100% cpu useage, and 100% memory useage at the same time for more than a few seconds is nearly zero, one will always be faster than the other on a task.


Edit: http://www.playtool.com/pages/prime95/prime95.html Shows how to get to the screen, 3rd pic down, use the custom mark and enter how much memory you want it to use in the test.

[Gskilldogz]

Brilliant, I'll try that, will tell you how it is going