Table of Contents
[-0-] Version History, Disclaimer & Legal Info
Problems, Workarounds and other stuff
Fixing the Regulator Misery - German Approach
Those of you who don't like to compromise and want to prevent problems rather than fix them later on, may consider this very different approach I've tried recently.
Sitting in "the lab" one morning at a very early hour, being quite uninspired and bored, I've started digging in my boxes full of "unsorted stuff". During that a bag fell into my hand containing no less than 4 switch-mode voltage regulator modules (VRMs) for the bigger IBM Netfinity machines (5600 or 6000R). They are designed for a single +12 V DC input and have a programmable 1.3 - 3.5 V DC output @ up to 16 A, intended for the Pentium III or Xeon processors.
Remembering the problem I had with the overheating LT-1084CT 3.3V regulator on the P90 platform, I thought it may be a good idea to try substituting the linear regulator with the switching VRM.
20 minutes later my 9595-B06 RAID box ran with a P133 CPU (I had no faster spare at that time) on the P90 platform, and with the Netfinity switching VRM. The module gets merely hand-warm. The bigger one of the two coils heats up a little - but that's it. No comparison to the "space heater" LT-1084 with its flimsy heatsink.
As you can see above, the pinout is pretty simple. The table shows ID codes for the different output voltages.
Three Steps To Heaven
Step 1: Removing the LT-1084 and its heatsink
Start by unsoldering the heatsink holder tab.
It may need a bit more heat and a little pulling force to come
free. The heatsink tab isn't electrically connected to anything on the board
- so it's mot fatal if you manage to rip the pad or pull the via out off
the board. But be careful anyway...
Step 2: Preparing the Switchmode Voltage Regulator
On my regulator most of the upper copper surface is ground, and
most of the bottom surface is either the +12 V input or the output.
I've decided not to bother with any connectors and go with just a direct
wire connections. So, I've scratched a 2 by 6 mm spot clear of the solder
mask of the GND plane, and of the +12 V input plane, and soldered two thick
wires directly to the board. On the other end of these wires is a standard
"molex" plug that fits directly to the 9595 power supply. Later I've replaced
it with a male-female extension (as for these silly processor fans) - that way
I still have all 3 PSU connectors available to me for other purposes.
Step 3: Installing on the P90 Platform
Now, there is no provision for fixing the regulator anywhere.
The simplest method was just to strap it onto the upper side of the
processor complex. It's relatively flat, there is enough room (given you don't
have a long MCA board installed in the bottom-most slot #8) and the way down
to the power supply as well as to the solder pads of the removed LT-1084 is
The result will probably look similar to that shown below:
With a voltage module of this caliber, any mistake in wiring or part selection/installation, may cause massive damage! The little LT-1084 regulator would collapse at around 5 A, but a baby like this can pump 16 A (or more!) into the circuit and it's being fed directly from the power supply - unit that can deliver 35 - 45 A at 12 V. That will blow the CPU out of the socket if it isn't oriented correctly. And will - most likely - cause massive damage to the PCB itself as well.