Paul Lacey said:
I am experiencing major instability problems when running my system with
my RAM (512 MB Samsung PC2700)in DIMM slot 1. Within 2 minutes of booting
the system will blue screen (usually a win32k.sys or
IRQL_LESS_THAN_OR_EQUAL errors). Extremely reproduceable. I have had the
board for about a year and have been running my RAM in slot 2 (no
particular reason) and have received regular lockups (1 or 2 per week). A
few weeks ago I moved the RAM to slot 3 and it hasn't crashed yet.
Is this by design or a faulty board? I have read in some groups and have
been told by a friend that it is recommended to put your first stick of
RAM in slot 3 (furthest from the CPU) and the 2nd (if used) in slot 1.
Has anyone seen anything from ASUS on this? I couldn't find anything in
the manual or on their site that mentions this.
These recommendations, of placing the first stick furthest from the
processor and two sticks in slot one and three, is based on some
transmission line theory. Placing the stick in the last slot is like
building a series damped transmission line, where the reflections only
help to build the signal to its final amplitude. Using slots 1 and 3
helps spread out the "depression" of the equivalent line impedance,
which means that the size of the reflections is smoother. It is a
slight effect, and using slots 2 and 3 wouldn't be that much worse.
When you place a single stick close to the processor, it leaves a
piece of wire (i.e. slots 2 and 3) wiggling off the end of the bus,
which can be worse for reflections.
All of this hand waving is moot, because you have to run a time domain
simulation to see in each case what is actually best. The exact
results are dependent on the lengths of the copper wires used (i.e.
how the layout is done), how well the motherboard stackup matches
the DIMM and so on. But as you have experienced first hand, the
hand waving theory is working out pretty well in your case.
There are also reports from people who have no trouble at all. This
simply means that their combination of components was well matched
enough, that the repositioning of the sticks is not enough to cause
memory errors. (It still screws up the signals, but not enough to
cause visible memory errors.)
The multidrop memory bus is a terrible design compromise, but it was
invented to save the customer money. If each memory stick had its
own bus, the PC motherboard would be bulletproof. But a Northbridge
with a couple thousand balls on the bottom would cost a fortune. I
think the current record for connections on the bottom of a chip
is somewhere around 6000 or so.
The motherboard designer has to balance the operation with one stick
to the operation with three sticks. If the designer could force you
to always use three sticks for example, then he could build a better
solution. It is the fact that so many memory configurations are
possible, that makes it difficult to get all configurations to work
well.
Also, if the company that made your motherboard, also made memory
sticks for it, they could be tuned for one another. Of course, the
price charged by the marketing department for this service would
be double for the memory sticks.
This is also a speed dependent thing. At DDR200 or DDR266, you won't
be able to detect any of these problems. It is the fact that the
memory bus design got pushed past its practical limits (i.e. DDR400)
by the marketing department, that makes it so crappy. I'm not
expecting DDR2 to give users any better experience.
You could try RMAing the board, but the next one could be just as
bad. It is the luck of the draw.
HTH,
Paul
