Multiple RAM failiure

[Thorben]

Hello forum people,

 

I've bought a new system using 4 Corsair CM2X1024-6400C4 modules on an ASUS P5N-E SLI motherboard.

 

As the system kept crashing, I decided to do the memory test with Memtest86+ on the system. As the test of the 4 RAM simultaneous always crashed, I started testing them one by one.

 

(The system volatge was put to Auto, I assume thats fine)

 

The resuls were strange (every module was tested twice)

Module 1: +- 311 errors

Module 2: +- 150 errors

Module 3: +- 25 errors

Module 4: okay

 

I know that the chance of buying 4 modules and having 3 "defective" modules tends till zero. Still, if there Is soomething systematically wrong with the mainboard, how could it be that module 4 is okay?

 

Furthermore, it just looks weird that from test to test, the errors are getting less and less. What also strikes me is, that, now testing for example the module 1 a third time after I tested all the other modules, it only brings around 100 errors.

 

There is something rotten in the state of denmark

 

thanks

Thorben

[DerekT]

The memory controller can not keep up with the extra DRAM at the rated 6400 speed. You will need to drop the speed of the DRAM from 800Mhz to 667Mhz. If you had purchased 4 X 1024MB of PC8500 (1066Mhz) DRAM, then you would have had to drop to PC6400 (800MHz) DRAM, etc. 2 X 2048 will not issue this problem. This is a problem of all 4 banks being populated.

 

Think of it this way. If you have a small phone book, then when you go into the index to find the page where you will find the phone number you are doing so at a certain speed due to the pages of the Index. Then you have to drive through the pages to get to the number. Now if your index is twice as large and the pages twice as many, then it takes longer to access the data. Now DRAM has a Strobe and the length of the strobe is how long the rows and columns can be left open before they must be refreshed. There is not enough time for the dram to be refreshed and then accessed with 4GB at the full access speed of the DRAM. The chipsets are optimized for 2GB, not 4GB and for 2 DRAM slots, not 4 DRAM slots populated. You can overclock the FSB (and hence the Memory Controller Hub = MCH) to gain some extra bandwidth and thus access the capabilities of the DRAM since the chipset is now clocked up. Usually when you clock up the FSB and concurrently the Memory Controller Hub (MCH) you need to raise the voltages of the CPU/MCH a bit as well.

 

Try these settings and retest:

Advanced:

 

JumperFree Configuration

AI Tuning = Manual

Overclock Options = Disabled

 

System Clocks

NB PCI-e Frequency (MHz) = 100

 

Voltage Control

Vcore Voltage = Auto

Memory Voltage = 2.1v

NB Core Voltage = 1.35

 

FSB & Memory Config

FSB - Memory Clock Mode = Unlinked

FSB - Memory Ratio = 4:5

FSB (QDR) = 266

 

Chipset:

 

Memory Timing Setting

tCL = 4

tRCD = 4

tRP = 4

tRAS = 12

Command Per Clock = 2 Clocks

tRRD = Auto

tRC = Auto

tWR = Auto

tWTR = Auto

tREF = Auto

tRD = Auto

tRFC = Auto

Async Latency = Auto

CPU Spread Spectrum = Auto

PCIE Spread Spectrum = Auto

SATA Spread Spectrum = Auto

LDT Spread Spectrum = Auto

LDT Frequency = 4x

 

USB Configuration

USB Controllers = Enabled

USB Legacy Support = Disabled (Disable when you memtest. Enable afterwards if you have usb keyboard)

USB 2.0 Controller = Enabled

 

Results?

[RAM GUY]

It is likely that these modules would give you errors when leaving the voltage at the auto setting. Set the memory voltage to 2.1 and follow the rest of Dereks settings and let us know what happens!

[Thorben]

Hey folks,

 

I just changed the settings according to Dereks instructions. There are nevertheless some flaws: The voltages do no exactly fit (MemV: 2.085V vs. 2.1V, NB CoreV: 1.393V vs 1.35V), the FSB memory ratio can not be changed and the FSB (QDR) value can not be set to 266 MHz as there is a minimum level at 533 MHz. As you said, i set the MemDDR voltage to 667 MHz.

 

What's most important: IT SEEMS TO WORK. I only had the time to test one RAM brick until now, but as it displayed no error in Memtest so far, and also putting in the other three bricks did not hurt, I suppose the test tomorrow will run fine. I hope at least.

 

Still, only question remaining: If the RAM voltage can not be set to the exact value, which to choose? The closest value, or the closest value just below or just above standart. And what to do with my FSB QDR?

 

thanks so much,

you guys are great

Thorben

[DerekT]

I just changed the settings according to Dereks instructions. There are nevertheless some flaws: The voltages do no exactly fit (MemV: 2.085V vs. 2.1V, NB CoreV: 1.393V vs 1.35V), the FSB memory ratio can not be changed and the FSB (QDR) value can not be set to 266 MHz as there is a minimum level at 533 MHz. As you said, i set the MemDDR voltage to 667 MHz.

Good interpretation Thorben. Setting the FSB to 533 must be the 1066 (QDR) and for some reason they double it. LOL

What's most important: IT SEEMS TO WORK. I only had the time to test one RAM brick until now, but as it displayed no error in Memtest so far, and also putting in the other three bricks did not hurt, I suppose the test tomorrow will run fine. I hope at least.

Very likely.

Still, only question remaining: If the RAM voltage can not be set to the exact value, which to choose? The closest value, or the closest value just below or just above standart. And what to do with my FSB QDR?.

I choose the DRAM voltage that is close to the lower end, but if it fails in memtest, then I move it up one increment. It's finding what fits. :) Your FSB is correct I would say. For some reason they turn the 266 X 4 = 1066 into 533 X 2 = 1066 I would think. Download CPU-z and check this out.

 

http://www.cpuid.com/download/cpu-z-141.zip