Installing DIMMs, order of slots

  • Thread starter Thread starter cyg
  • Start date Start date
C

cyg

There seems to be a common wisdom that one should always populate DIMM slots
starting with the lowest-numbered slot. However, I've never seen an
explanation as to why this is. I have reason to believe that my second slot
may be faulty, but I'm not sure if Memtest will be affected by the fact that
in order to isolate the issue, I can't have a DIMM running in the first
slot.

Actually, I do know that Memtest reports the speed of the chip differently
depending on whether I put it in the first slot or the second slot. Then
again, I am using an AMD processor with an integrated MC, so maybe it is
also something to do with the CPU. Anyone have answers?
 
cyg said:
There seems to be a common wisdom that one should always populate DIMM slots
starting with the lowest-numbered slot. However, I've never seen an
explanation as to why this is. I have reason to believe that my second slot
may be faulty, but I'm not sure if Memtest will be affected by the fact that
in order to isolate the issue, I can't have a DIMM running in the first
slot.

Actually, I do know that Memtest reports the speed of the chip differently
depending on whether I put it in the first slot or the second slot. Then
again, I am using an AMD processor with an integrated MC, so maybe it is
also something to do with the CPU. Anyone have answers?

In digital design, things have "signal integrity" issues. Any time
there is a shared bus concept, there is a possibility that some
hardware configurations give better signal quality than others.
And that can affect the speeds used by the BIOS.

On DDR memory, there are terminator resistors to help absorb
reflections at the end of the bus. With those in place, the
slot position should be less sensitive. (In this picture, Rt
is a terminator at the end of the bus lines. This picture does
not show the RAM slots, and they would be located where that
coaxial symbol "bus" is located. This bus termination scheme
is called SSTL.)

http://media.maxim-ic.com/images/appnotes/1775/1775Fig01.gif

In the Northbridge (integrated or discrete), an address decoder,
takes the address delivered by the processor, and associates it
with a slot. As long as that design is flexible, you should be
able to shove a DIMM, into whatever slot you like. The BIOS
figures out which slot is populated (by seeing the SPD chip
on the DIMM). It then programs the Northbridge, to deliver
chip select signals to that slot, in response to a particular
range of addresses.

On an AMD processor, with its integrated memory controller, if
a chip select signal going to one of the slots is damaged, that
could prevent the slot from being used. If that was the case,
then a different processor would make the slot work again.

If the slot pin is bent, then changing processors would not make
a difference.

For non-AMD processors, like say my old Pentium 4 motherboard,
the memory controller is in a separate Northbridge chip. If slot #2 was
broken there, and the slot connector looked undamaged, it could
be the discrete Northbridge which is broken. A new motherboard might
be required, to make slot #2 work again.

If your motherboard is an AMD S754 board, with three slots,
the bus connections are a little more complicated. Such a board
has one data bus (single channel mode). But the address bus
is split into two pieces. One address bus drives two slots,
the other address bus drives one slot. The address buses carry
the "1's complement" of each other, and only one address bus
carries "valid" information at a time. Chip select signals tell
the DIMMs, when their bus is valid. The purpose of that, is
to make switching noises cancel, at the processor. For S754,
the speed of operation the BIOS is willing to support,
depends on whether one or two sticks are sitting on that one
address bus. For retail motherboards, they may include a
configuration table in the manual, showing what speed is
possible with various RAM slot population patterns.

If the motherboard is S939, people call it a "dual channel" board,
but strictly speaking, it is not. AMD calls the operation of
the board, either "64 bit" or "128 bit" mode. For older S939
processors, there are slot restrictions when using a single
DIMM, due to those 64 bit versus 128 bit modes. Only two
slots work in 64 bit mode, and the user has to know which slots
those are.

If the S939 processor is revision E or higher, the restrictions
are removed, and any single slot can be used that you feel like.

As you can see, when you don't provide details about the exact
computer or motherboard involved, there are lots of choices
and explanations possible.

Memtest86 doesn't care where the memory lives. Memtest86
relies on the BIOS to set things up. All memtest does, is
deliver addresses, and the hardware takes care of delivery,
just like your Post Office.

HTH,
Paul
 
Actually, I do know that Memtest reports the speed of the chip differently
depending on whether I put it in the first slot or the second slot. Then
again, I am using an AMD processor with an integrated MC, so maybe it is
also something to do with the CPU. Anyone have answers?

Memory is addressed in banks. By putting them in the correct slot, they
can be addressed in "dual channel" or "triple channel" mode thus
increasing the throughput.
 
cyg said:
There seems to be a common wisdom that one should always populate DIMM slots
starting with the lowest-numbered slot. However, I've never seen an
explanation as to why this is. I have reason to believe that my second slot
may be faulty, but I'm not sure if Memtest will be affected by the fact that
in order to isolate the issue, I can't have a DIMM running in the first slot.

Actually, I do know that Memtest reports the speed of the chip

Chip??? Don't you mean "module", that thing with all the memory chips
soldered on it?
differently depending on whether I put it in the first slot or the second slot.

The only mobos I've seen that required a DIMM in the first slot were
based on the Intel 810i chipset (815x may be the same). The slot
shouldn't matter, even though the timings for each slot vary a tiny
bit, but with marginal memory that could be enough to prevent proper
operation. So maybe you should go into the BIOS setup and override
the automatic memory timings and substitute the slowest timings
allowed, including for the memory bus speed. Then if the DIMM works
in your second memory slot, it means the slot is good but the memory
isn't. Actually if the DIMM works with less than 100% errors, the
slot is almost sure to be OK.

Lots of retail memory is marginal now because it's made with chips
that were either not fully tested (Kingston buys whole wafers, slices
and packages them in-house, and apparently tests them at lower
standards) or were rejected by the chip manufacturers (those chips are
euphemistically called "untested" -- UTT). Your best bet is to stick
with Crucial modules, which usually have Micron or Samsung chips on
them, or test very thoroughly on your own and keep going back to the
dealer until you get something that passes 100%. BTW, try to use more
than one test program because I've never had an error reported by
MemTest+ or Gold Memory 6.92 but lots reported by MemTest86 ver. 3.xx
(MemTest+ is based on MemTest86) and Gold Memory 5.07 And don't
accept memory that won't work perfectly at the BIOS automatic defaults.
 
Back
Top