sillyputty said:
I'm looking at the Asus P6T mobo for the i7 940 CPU. The video dude on
tigerdirect [
http://www.tigerdirect.com/applications/SearchTools/item-
details.asp?EdpNo=4268650&Sku=A455-2854]
says the CPU and memory voltages are linked and the max memory voltage
is 1.6v. However, I called Frys and asked about that particular MB he
said the power supplies for the CPU and memory are separate. Who's
right? Thx, as usual.
If I look at a picture of the P6T Deluxe...
http://images17.newegg.com/is/image/newegg/13-131-351-S03?$S640W$
and look at the right hand end of the DIMM slots, I see a three
phase regulator of some sort. If I look near the CPU socket, I
see the usual multiple phase Vcore supply. It is even possible
that the stuff around the CPU socket, is split into VCC and
VTTx rails.
So there is a good possibility of independent power sources.
If I look at the i7 datasheet...
VCC Vcore for main portion of CPU 1.55V
VTTA,VTTD (Memory Controller,QPI,Cache) 1.35V
VDDQ DDR3 I/O Pad Voltage 1.875V
VCCPLL Processor PLL 1.89V
So there are potentially four rails on the CPU. VTTA and VTTD are
kept separate for noise considerations, but should be joined
on the motherboard copper.
While the Intel datasheet rates the rails, as if they are independent
of one another, there can be some hidden assumptions. For example,
inside some chips, there are unintentional (part of the technology)
parasitic paths, between power rings. On some of the chips we designed
in house, if one rail was more than a diode drop above its neighbour,
a large current could flow inside the chip, between rails. Sometimes it
is hard to get an answer from the fab, about what the allowed rail
relationships are, between rails.
It could be, that the Vcore and VDDQ, cannot be more than X volts
apart. If you set Vcore to 1.0V and VDDQ to 1.875V, maybe something
nasty happens.
That seemed to be true on a previous generation of AMD processors.
People had a few FX55's die, before it was figured out by the
users, that if Vdimm was boosted, Vcore had to be boosted as well.
As long as the differential between them was less than some magic
value, the chip would survive.
I'd keep my eyes peeled, on a site like xtremesystems.org/forums,
to see what rules the users have discovered so far. As far as I know,
at the moment 1.65V seems to be one possible value for Vdimm. I don't
know if they've started torturing the things yet or not.
The above specification table is "Absolute Maximum" and is not
intended as settings for your first boot. The nominal (default)
settings will be somewhat less. For some firms, the specs in a
table like that, would be for an expected 100,000 hour lifetime,
implying the voltage could in fact be high enough to cause long
term degradation. You have to know the specification practices
of the company providing the table, to understand whether
Absolute Max is instant destruction or long term wearout.
The distance between the real "Instant Destruction" value and
values in those tables, has been shrinking over the years, so
there is likely not a lot of headroom above those values.
Paul
