Abit KD7-RAID and Memory

[John Corson]

In the Abit KD7-RAID manual, the implication is that the motherboard will accept 4 unbuffered PC-2700 DDR - ECC memory chips. However, on the Corsair website, they suggest that Abit suggests not having more than two sticks of RAM. 4 sticks of PC 1600 or 2100 would be acceptable.

I currently have 4 sticks of 512 mb of PC-2700 DDR and have one major issue, viz. explorer.exe closes in Windows XP then re-starts (Error message 1002 Winlogon needing to close explorer.exe and restart). There usually is a message pop up discribing an error in a certain memory address. Is this issue related to having more than two PC-2700 memory chips on the motherboard? If so, will removing the chips on banks 3 and 4 eliminate the problem is could this be another hardware/software issue.

Abit KD-7 RAID
AMD Athlon XP +2700 Thunderbird 333 FSB
4 - 512 Mb PC 2700 DDR = 2,048 mb
Windows XP Pro / SP 1
ATI All-in-Wonder Radion 9700
Creative Sound Blaster Audigy 2 Platinum
4 - 80 Gb Maxtor HDD - on raid controlers (not raided)
Sony DVD-/+R -/+RW
Plextor CD/RW
Iomega 250 Mb Zip Drive

---

John C

 

[Satan's Little Sister]

Thus spake John Corson:

In the Abit KD7-RAID manual, the implication is that the motherboard will accept 4 unbuffered PC-2700 DDR - ECC memory chips. However, on the Corsair website, they suggest that Abit suggests not having more than two sticks of RAM. 4 sticks of PC 1600 or 2100 would be acceptable.

I currently have 4 sticks of 512 mb of PC-2700 DDR and have one major issue, viz. explorer.exe closes in Windows XP then re-starts (Error message 1002 Winlogon needing to close explorer.exe and restart). There usually is a message pop up discribing an error in a certain memory address. Is this issue related to having more than two PC-2700 memory chips on the motherboard? If so, will removing the chips on banks 3 and 4 eliminate the problem is could this be another hardware/software issue.

The problem with using multiple sticks of RAM is capacitance. As the northbridge transitions a data or address line from 1 to 0, or vice versa, it must charge/discharge the input gate capacitance of all the memory chips plus the parasitic capacitance of the motherboard traces.

As you add more chip/sticks, you make the northbridge work harder. Instead of nice clean square wave transitions, you end up with rounded corners as the chipset struggles to charge and discharge everything.

At PC1600/2100 speeds, sloppy rounded signals aren't a big problem - there's time to get the signal to its proper high or low value. With PC2700 or higher, things get trickier - with too much capacitance the signal won't reach its proper state before being acted upon.

To reduce that pesky capacitance, a register/buffer can be used (one register per DIMM side). Instead of the chipset driving the inputs of each RAM chip, it just drives the registers and the registers drive the individual chips. Problem is, buffering means a delay, register chip are power hungry, registered and unregistered can't be used together. By the way, ECC and buffering/registering are separate, independent technologies.

Enough theory! You have four sticks and they are *almost* working correctly. Your options are: pull one stick, three will most likely work; swap those DIMMs for four registered ones ($$$); slow down the memory bus speed a bit, giving them thar signals longer to transition; increase the voltage to the northbridge, so it can drive all that RAM.

Download yourself a copy of Memtest86 (http://www.memtest86.com/). It could save you from corrupting your OS installation.

 

[John Corson]

Hey SLS,

Thanks for your response! You are right about using ($$$) Registered memory and I was going to make that the next move. I realize registered mem and ECC are unrelated technologies, but I think the reason I wasn't getting BSODs was because of ECC. Instead I received the popup windows about a memory address problem.

Until I win the big Lottery and have the money for registered mem, I guess I will use only two sticks of PC-2700 and be happy with Swap Filing if I need to use more memory.

Thanks for your note! I was really helpful -- especially the lead on Memtest86. I forgot that tool was out there!!

---John C---

Thus spake SLS:

The problem with using multiple sticks of RAM is capacitance. As the northbridge transitions a data or address line from 1 to 0, or vice versa, it must charge/discharge the input gate capacitance of all the memory chips plus the parasitic capacitance of the motherboard traces.

As you add more chip/sticks, you make the northbridge work harder. Instead of nice clean square wave transitions, you end up with rounded corners as the chipset struggles to charge and discharge everything.

At PC1600/2100 speeds, sloppy rounded signals aren't a big problem - ...