To fix this problem, add a 200 to 220 Ohm resistor in parallel to the 56 Ohm ZP resistor. This makes the ZP resistance about 46 Ohms and generally solves the data corruption problem. See the fix guide for details. Look here for the complete FAQ and discussion of this problem.

Problem background:

It appears that some Athlon CPU's, when combined with some (of these) motherboards, have pattern dependent data corruption in the communication between the CPU and 735 chip. The data corruption can easily be fixed by simply telling the CPU to send stronger signals (lower ZP resistance causes stronger signals).

Problem History: I, like many others, had no problems with any of my first boards (in mid 2001, with stickers numbered 0-2 by the PCI slots). I think this is because these early boards have 40 Ohm ZP resistors. Things changed when I got new boards (toward the end of 2001, and continuing to boards in early 2002), that had a pair of "XP" stickers by the Socket A. I, like many others, noticed that something was very wrong. Windows seemed pretty stable, but I couldn't make CD's without errors when using an Athlon Thunderbird 1400 CPU.

Not only did Windows have problems, Memtest86 also reported many errors.

Nothing I could do would help until I lowered ZP by placing a 200 Ohm resistor in parallel with the existing 56 Ohm resistor. Then, everything worked fine on every single board I had!

When using the Athlon 1400, ZP was the only thing I was able to find that seemed to make the problem better (or with higher ZP value, worse). I tried many things before I tried ZP, and nothing else would really make the problem better.

Many others have reported the same problems and the same success with the ZP change (10-20 at least).

A change to 56 Ohm ZP value was discussed in a preliminary AMD document, but if you read carefully, it seems to be subject to verification. From the listing on AMD's site at then end of 2001, it does not appear that this board was ever tested with the Thunderbird chips for stability or compatibility with the 56 Ohm ZP.

The lower ZP value changes the drive strength of the data signals going from the CPU to the 735 chip. This is for "impedance matching" to make sure the signals get to the 735 clearly.