P4P800 and dual channel mem

[Driekes]

On the Asus site I found the list of certified DDR400 memory for the Asus
P4P800. So far so good, but what I don't read (or don't get ) is what mem
I should use in dual-channel mode? Or can I just use two sticks of mem.
listed in there list?

If else can someone give me some advise?

Thanks!

Driekes

 

[Dorph]

On the Asus site I found the list of certified DDR400 memory for the Asus

P4P800. So far so good, but what I don't read (or don't get ) is what mem
I should use in dual-channel mode? Or can I just use two sticks of mem.
listed in there list?

Certified my ass - just 2 DDR400 of same brand - same size - same
everything - and you are god to go for dual-channel mode.

Dorph

 

[Tim Payne]

The list gives columns A, B & C to show what combinations work with each
memory type: single, one pair or two pair. So long as the memory meets the
'B' column (works in a pair), just buy two sticks of that memory and you're
good to go.

What I don't get is why the memory I'm looking to buy (Kingston
KVR400X64C3A/256) won't work in two pairs (4 modules). Why would this be?

Tim.

 

[Linux Lover]

What I don't get is why the memory I'm looking to buy (Kingston
KVR400X64C3A/256) won't work in two pairs (4 modules). Why would this be?

Perhaps a special signal necessary to allow that is not implemented on
that module?

Seems strange, considering the fact that the KVR400X64C3A/512 (notice,
exactly the same number, except for the 512 vs. 256 to denote a 512MB
module), uses the same exact hynix chips (HY5DU56822BT-D43) and _will_
work in two pairs.

Then again, maybe the 5th column in the table provides a hint (SS vs.
DS). Don't know what that acronym means for sure, but I am guessing its
Single-Sided vs. Double-Sided.

But then, I see some other modules (from different brands) that are SS
_and_ work in two pairs.

Strange.

 

[Paul]

Perhaps a special signal necessary to allow that is not implemented on
that module?

Seems strange, considering the fact that the KVR400X64C3A/512 (notice,
exactly the same number, except for the 512 vs. 256 to denote a 512MB
module), uses the same exact hynix chips (HY5DU56822BT-D43) and _will_
work in two pairs.

Then again, maybe the 5th column in the table provides a hint (SS vs.
DS). Don't know what that acronym means for sure, but I am guessing its
Single-Sided vs. Double-Sided.

But then, I see some other modules (from different brands) that are SS
_and_ work in two pairs.

Strange.

The combinations that work, should be determined by bus loading.
A SS has half the load of a DS, so in theory more of them should run.

I get the impression that not all memory module vendors conform to
the JEDEC specs. One web site mentioned fooling around with the
series damping resistors on their module, and that type of customizing
might mean that those modules might prefer a certain kind of motherboard
over some others. The JEDEC standards were invented to give
broad compatibility between modules and motherboards.

On the 875/865 Northbridge, it looks like Intel has integrated
the SSTL series damping resistors inside the Northbridge. This also
is not "normal", as most motherboards have them external to the chip.
SSTL consists of a series damping resistor next to the Northbridge
plus a parallel termination resistor at the end of the memory bus.
The voltage used to do the termination of the bus is heavily
bypassed with capacitors located near the parallel termination
resistors.

It is differences of this type, plus maybe some small variations
in setup and hold times for certain kinds of chips, that give the
weird compatibility tables.

Asus most likely has constructed their table empirically, using a
small sample size of modules. In the case of Kingston modules, as
many as three different kinds of memory chips can be used to fill
an order for a single KVR code. I've read of at least one case,
where when someone couldn't get their modules to work in their
motherboard, they phoned up Kingston, and were told to return the
modules, so the correct vendor chips could be used. (I think this
was a Hynix versus Infineon thing).

I'm afraid Dorph's observation is entirely correct :-(
Memory is a crapshoot...

HTH,
Paul