Author Topic: Reverse engineering a server CPU voltage regulator module  (Read 3655 times)

Andy Brown

  • Administrator
  • Hero Member
  • *****
  • Posts: 267
    • View Profile
    • Andy's Workshop
Reverse engineering a server CPU voltage regulator module
« on: September 24, 2016, 07:28:59 am »
Some fairly aimless ebay fishing recently turned up a voltage regulator module for older Xeon CPUs. Initial thoughts of breaking it open for parts were quickly replaced by a desire to reverse-engineer it and perhaps reuse it as a miniature PSU board.



Click here to read about how I got on. A video, extensive load testing results and lots of photographs accompany the writeup.

When you get to the end of the article you'll note that there are many unanswered questions. If anyone knows what the unidentified pins do then please feel free to post your ideas. Eventually I'd like to build a board around this module and it would be nice to expose all the functionality.
It's worse than that, it's physics Jim!

mr_goll

  • Jr. Member
  • **
  • Posts: 2
    • View Profile
Re: Reverse engineering a server CPU voltage regulator module
« Reply #1 on: October 09, 2016, 05:43:50 am »
Hello.

Got an earlier version of the board after seeing your video (made in USA - visibles almost identical) - nice sturdy module. Here are my observations and thoughts, having spent the night with it.

With a 4.7k trimpot on the (lm342 OUT2) -> (HIP6004 VSEN) in place of R11 (normally 3.4k), we can convince the HIP6004 to alter output within new range i.e. 4.1k=3v, 2.3k=4v, 1.1k=5v etc. Great with no load. But there is a pesky feedback loop (I think) HIP6004->LGATE  switching the Overprotection FET on creating a dead short to ground (therefore blow input fuse to protect the output). That feedback loop is fed by the HIP6004 monitoring the voltage drop across the Upper/Top FET (the TO220 FDP7030).

When reviewing my initial thoughts were to pop the HIP6004 and put another micro PWM controller on there, but in retrospect less is so much more. Mulling removal of the D-PAK (HIP6004 OVP) overprotection fet, simply seems most practical for my potential use case as a low current stable supply. There is a fair few passives on there obfuscating matters; but fundamentally cant foresee too much of a problem there.

After Lifting the gate of the overprotection fet (HIP OVP) - it all works like a charm (limited testing few hundred milliamp).

P.S. The through hole barrels securing the heatsink posts are solid - iron at ~ 330c, bit of fresh lead; screwdriver pry-bar; the heatsink came off a treat!
« Last Edit: October 09, 2016, 07:02:47 am by mr_goll »

Andy Brown

  • Administrator
  • Hero Member
  • *****
  • Posts: 267
    • View Profile
    • Andy's Workshop
Re: Reverse engineering a server CPU voltage regulator module
« Reply #2 on: October 09, 2016, 09:22:45 am »
We now know that it's an NXA66 and enough of a datasheet is available for download to work with it. It's capable of two fixed output voltages. 3.3V and whatever you get after grounding VSP. I now have two modules, the one you see in the blog post from a Dell server will output 5V with VSP grounded which is contrary to the datasheet's claim of 2.5V so I bought a second module that looks the same except it has a silver coloured heatsink and is labelled Rev A01. This new one outputs 2.5V with VSP grounded as the datasheet claims. So there are two variants out there, both with very useful output levels and spec'd for up to 20A continuous output current.

There's also a differential SENSE input that, although not properly documented in the datasheet, is highly likely to be used to compensate for voltage drop at the load. I intend to experiment with these to see what happens.

I'm currently designing a PCB to mount this board on with some typical PSU features. I hope to make it simple to construct (through hole) so anyone can pick one of these up from the 'bay and build a nice little PSU around a very solid module.
It's worse than that, it's physics Jim!

mr_goll

  • Jr. Member
  • **
  • Posts: 2
    • View Profile
Re: Reverse engineering a server CPU voltage regulator module
« Reply #3 on: October 10, 2016, 06:06:20 pm »
I am using Rev A01, Silver Heatsink ATM. With 16k R11 : 3.4k R10 divider it happily goes down to 500mv, but I don't think the soft-start likes it until up and around 1v (with a load on it at least) - that is with the std RT frequency. So VSEN manipulation gets you 0.5-6.5V out of the box.

Many thanks for your hard work. Inspiring, post consumer recycling at it's finest!

Andy Brown

  • Administrator
  • Hero Member
  • *****
  • Posts: 267
    • View Profile
    • Andy's Workshop
Re: Reverse engineering a server CPU voltage regulator module
« Reply #4 on: October 12, 2016, 01:28:31 pm »
I am using Rev A01, Silver Heatsink ATM. With 16k R11 : 3.4k R10 divider it happily goes down to 500mv, but I don't think the soft-start likes it until up and around 1v (with a load on it at least) - that is with the std RT frequency. So VSEN manipulation gets you 0.5-6.5V out of the box.

Does PGOOD stay high at all those voltages?
It's worse than that, it's physics Jim!

Andy Brown

  • Administrator
  • Hero Member
  • *****
  • Posts: 267
    • View Profile
    • Andy's Workshop
Re: Reverse engineering a server CPU voltage regulator module
« Reply #5 on: December 26, 2016, 06:41:20 am »
I've now completed my build of a power supply controller around this module and the write up will appear in a blog article in about 2 weeks when I've had time to fully write it up. It's looking great and performing really well!
It's worse than that, it's physics Jim!