I'm running an ASUS P4C800 Deluxe M/B, P-4 3.2 Ghz processor, Windows
XP Pro, and had 1Gb DDRAM in two 512 modules. I installed two more
512 modules for a total of 2Gb and now the machine seems to be
running slower, as in takes a long time for individual windows
to open, or for programs to launch. The original modules are
about a year-and-a-half old, and are PC3200. The two new modules,
which I bought a month or so ago, are also PC3200 and I was told
by Kingston, the mfg, that they should work alright with the older
(non-Kingston) modules. I don't know if I should buy another pair
of Kingston so all four modules are the same, or if I could be
having some other difficulty. Is 2Gb too much RAM? Does Windows
XP have a limit? I do notice physically that although both old and
new modules are 512Mb and PC3200, that the older ones have chips
on both sides of the card, but the newer modules only have chips
on one side. What would this mean? When I added the extra RAM,
should I have made any changes in the BIOS?
Thanks for any info.
The 875P memory guide is here:
ftp://download.intel.com/design/chipsets/applnots/25273001.pdf
What you've got, is two different configuations of memory.
The first pair of DIMMs are (16) x 32M*8, while the second
single sided DIMMs are (8) x 64M*8. That means the addressing
on the DIMMs is different. It has the same effect as if you
ran 2 x 512MB and 2 x 1GB modules at the same time. The memory
controller cannot use its simple interleave pattern, when
the lower and upper pair of DIMMs address differently. Thus
it is forced into "Normal" mode.
Channel 0 Channel 1
512MB SS <--paired--> 512MB SS Dual channel mode, but
512MB DS <--paired--> 512MB DS "Normal" not "Dynamic".
Interleaving disabled.
A small penalty. See
Fig.6 in Mem Guide.
Channel 0 Channel 1
512MB SS not 512MB DS Virtual single channel
512MB DS paired 512MB SS mode. A larger penalty.
Sort of like Fig.3 in
the Memory Guide. You
can "feel" the difference.
Here are some numbers I've measured with memtest86+ bandwidth
indicator. These numbers will be different than the ones you
see in Sandra. I don't have any "Normal" mode numbers, as
my four DIMMs are matched, so I'm in "Dynamic" mode.
Single channel mode - 1681MB/sec DDR400 2-2-2-6
Dual channel mode - 2667MB/sec DDR400 2-2-2-6 PAT disabled, Dynamic
Dual channel mode - 2732MB/sec DDR400 2-2-2-6 PAT enabled, Dynamic
(The last case used BIOS "performance mode" set to "Standard")
The screen has a little extra "snap" to it, when running dual
channel. But I wouldn't describe it as "dog slow" when running
single channel. It is still usable. (I actually noticed this on
a 865G based computer belonging to a relative - I could tell
immediately the store that sold the RAM had not properly
matched it, as soon as I sat dowh at the machine - there
was a four chip and an eight chip DIMM in the computer!
Oh, the horror. )
In your case, you should get maybe a few percent less than my
~2700MB/sec measurement. That is because the interleaving of the
Dynamic mode is disabled, and you would be in Normal mode. Fewer
memory pages remain open, so every once in a while, a few cycles are
spent re-opening a memory page. Obviously, real programs will
see different behavior than a program like memtest86, due to the
more random access, but the difference should still be a low
percentage.
If the memtest86 bandwidth indicator is resting around 1681MB/sec,
then check whether you have paired the DIMMs correctly.
Get a copy of CPUZ from
www.cpuid.com, and see what it says.
You can also examine the SPD on the DIMMs with CPUZ.
As for PAT or MAM, or whatever it is called this week. PAT
also makes a small difference, as the numbers above demonstrate.
With four sticks, I think the deal is, PAT is disabled if the
memory clock strays above DDR400. If you set it to DDR401, I
think the BIOS disables PAT. Of course, at some point, the
extra bandwidth resulting from using a higher memory clock,
will be more than the loss due to no PAT - if the DIMMs could
be run at DDR440, you would be ahead of the game. I think PAT
is only enabled if using CAS2 DIMMs, but I cannot confirm that.
And no, I don't have any Sandra numbers to share with you.
I was testing the memory at the time, with memtest86+, and
that is why I have records of what the bandwidth was. I was
in the process of tuning up my P4C800-E, measuring current
consumption on the rails and the like.
In fact, dual channel mode with two matching sticks was
2732MB/sec and so was four matching sticks. It gave
exactly the same 2732MB/sec figure. In both cases PAT
was enabled. So, that proves there doesn't have to be
any penalty with running four sticks, for what that is
worth. The difference is when you overclock. I think with
a single pair of DIMMs, you can reach up to DDR600, while
with four DIMMs, you are doing good if you get to DDR440.
So, a single pair of DIMMs offers much better overall
bandwidth, due to the ability to overclock. But the other
timing factors can make zero difference by themselves,
when going from two DIMMs to four DIMMs.
The only remaining nagging unknown, is what "Command Rate"
do Intel chipsets use. Nforce2 and Athlon64 are examples
of memory controllers, where Command Rate 1T or 2T makes
a difference. Xbitlabs.com makes the claim, that the
Intel controller operates all the time in Command Rate
2T mode, but the subject is never mentioned in any
Intel datasheets. Since the bandwidth is exactly the
same with two DIMMs and four DIMMs, that would seem to
suggest the Command Rate setting remains the same - we
just cannot observe what that setting might be. There is
no documented register in the memory controller for
Command Rate.
Paul
