[Tfug] Interchangeability of VRM's?

[Bexley Hall]

Hi Adrian,

On 12/10/2012 3:44 PM, Adrian wrote:
> On Monday 10 December 2012 12:52, Bexley Hall wrote:
>
>> My assumption is that VRM's are just "very local" voltage
>> regulators designed to tweek the supply voltage to the
>> processor on a processor-specific (or even "on demand")
>> basis.
>
> More or less, there are 4 (or was that 5?) digital lines that select the
> output voltage.

Are they "static"?  Or, can the processor vary them while
executing?  (i.e., are the 4/5 signals effectively hardwired
to a specific processor-specific configuration?)

>> But, how universal are the VRM's?  I assume *less* universal
>> than "PC power supplies" (which don't care if you're running
>> AMD, Intel or WhoopdeeDoDah).  I.e., are the signalling bits
>> from the processor to the VRM somewhat standardized?
>
>> Similarly, does the VRM connect to the host/motherboard by
>> some standardized interface?  Or, is the VRM seen as an extension
>> of the motherboard?
>
> Excluding the manufacturer specific versions (Dell/HP/Compaq/etc. had variants
> specifically for their motherboards), VRMs were standardized, but not
> necessarily interchangeable.

So, they are more a function (responsibility) of the motherboard
and *not* the processor.  I.e., the processor manufacturer expects
your "board" to provide power "per this specification" and you
(the MB manufacturer) are free to implement that however seems
best for your goals (e.g., it need not be a *removable* module if
your reliability/service criteria are such that putting the
components directly on the MB would be sufficient)

> As I vaguely recall... there were 2 formats of VRMs: one was a dual-inline pin
> (about 2x20 pins), the other was a slot connector, and hence not
> interchangeable between formats. There were also a number of different
> versions, in increasing revision number. Generally, the higher the revision
> number the greater the number of voltage output options (i.e. an early P3 may
> need a 8.1 spec, a later P3 needing a different voltage may need a 8.2 spec).
> Generally, revisions were backward compatible.

So, moving "too far" up the performance curve may push you out
of the capabilities of a particular MB+VRM implementation.
E.g., it may not be possible to "upgrade" a system as much as
one might wish...

> To make matters even more complicated, the VRM spec assigned 5V and 12V input
> rails on the connector, but neither was mandatory. So there were 5V input
> VRMs, 12V input VRMs, 5V/12V either-or "universal" VRMs (where the VRM would
> run off either rail if present), and 5V/12V required VRMs (where the VRM
> required both rails to operate). The motherboard manufactures also choose to
> not always implement both rails... so there were some motherboards with only
> 5V or only 12V supplies to the VRM.
>
> Therefore, you had to match the VRM to the type of motherboard supply you had.
> A 5V VRM would not work in a 12V-only board, or vice-versa, but a
> 5V/12V "universal" would work in either motherboard.

OK, this just reinforces the conception that the VRM is effectively
part of the MB (and not the CPU itself).

> I did a writeup on differences between a bunch of different VXI-brand VRMs
> some 10 years back (when I was selling a bunch of refurbs), but I can;t seem
> to find it now. Intel had a number of very information design spec documents
> at the time. Search for the Intel 8.1, 8.2, 8.3, and 9.0 VRMs design pdfs.

Thanks for the insights!  Bottom line:  blindly swapping components
that *seem* to "fit" is in no way a guarantee that they will work
(nor be "safe" for the other bits on the MB).

Thx,
--don