Diagram 1
AMD went to a lot of trouble making the K6-III a drop-in replacement for the K6-2. They wanted to take advantage of the millions of SS7 boards already in use. FIC has surely dropped the ball as far as the drop-in replacement idea is concerned. Owners of older VA 503+ and PA 2013 are being told that their boards "may show some instability when used with the K6-III in extreme conditions or under high stress". They also say that, for a fee, your local dealer can perform a hardware rework that will allow you to use the K6-III. When FIC support is asked what the hardware rework involves, they will give the reply "It is traditional policy not to release any schematic question".
An e-mail reply from FIC support that was posted on the FIC newsgroup provided some insight. The resistors R148 and R149 make up a voltage divider across the current sense resistor. The current sense resistor consists of R140 & R143 in parallel. See Diagram 2.
The IC responsible for voltage control and current limiting is a Semtech Corp. SC 1101. The voltage developed across the current sense resistor is divided by R148 & R149 and the voltage drop across R149 is input to pins 2 & 3. Internally this voltage is applied to a comparator. Current limiting starts when the voltage on pin 3 becomes 70mV more positive than the voltage on pin 2. When this threshold is reached the output of the comparator is used to interrupt the drive pulses to the external switching transistor. The value of R149 is critical. Lowering the resistance will lower the voltage drop seen at pins 2 & 3. This will allow the CPU to draw more current before limiting begins.
The 3.3V power supply uses the same design. The difference is that the output voltage is not adjustable.
Why FIC will not guarantee the operation the newer AMD processors in older boards could be related to tolerances of the current limiting components. For example, from the Semtech specification sheet the current limit threshold of the SC 1101 can be anywhere from 60mV to 80mV. This translates to a tolerance of ±2A of CPU current. Taking this figure into account, a board designed for a CPU current of 14A could have an actual current limit threshold of below 12A. The K6-2 450-475 & K6-III's all have maximum current draws of more than 12A.
The component values and board markings are taken from a VA 503+, Rev. 1.1a, ECN 3502. In order to reduce costs, FIC probably uses the same design in other versions of the VA 503+ as well as the PA 2013.
The following assumes a certain amount of experience replacing surface mount devices. I take no responsibility for any modifications you make to your own equipment.
Current transients that could trip the limiter are difficult to measure. Forget about using an ordinary DVM. The best way to determine whether the current limiter is causing problems is to temporally disable it. Locate R140 & R143 on the board using Diagram 1 as a guide. They look like two loops of bus wire placed side by side. From #22 wire or component lead cut a piece 5/8 in. long and jumper the two resistors. I found that this is more easily done from the back of the board. Power-up and check to see if the problems persist. If they do, then the current limiter is not at fault. Remove the jumper.
If the problems go away after disabling the current limiter, then there are two options. The first is to leave the limiter disabled. This will leave no short circuit protection and is not recommended. The second option is to change the value of R149. Lowering the value will increase the threshold. On my board changing the value from 3.65K to 2.4K increased the current limit threshold point 10%. Different versions might have a different value for R149, but whatever the resistance value, the idea is to lower the value to a point where the limiter doesn't cause problems when the CPU is stressed.
The SC 1101 can be made to regulate down to 1.3V. A stock VA503+ has a base voltage of 2.0V. If AMD introduces a K6 that requires a core voltage lower than 2.0V I'll put up a page showing how to lower the base voltage. We'll just have to wait and see what AMD does.
Part sources:
References:
