Memory distribution question

[octessence]

I have two systems, one is the prime machine, the other runs linux and
just takes all of the bits left over from my main PC when I upgrade. I
have the following memory to be distributed across two machines (I want
2GB on the prime, 1GB on the secondary):

1x 1GB DDR400 PC3200U 3.0-3-3-8@200MHz
2x 512MB DDR400 PC3200 (?U) 3.0-3-3-8@200MHz
2x 512MB DDR400 PC3200 (?U) 2.5-3-3-8@200MHz

I'm not sure what the U means in the above or whether it's relevent.

The options are as follows:

1) Put all the CL3 memory in the prime machine because they are all the
same timings. However will the two 512MB modules be able to run in dual
mode with the odd 1GB module inserted? If not, will I take much of a
performance hit?

2) Put all 512MB modules in the prime and run them all at 2.5-3-3-8. I
would run memtest86 over a weekend to verify that it's stable. I've
done some experimenting with this but my motherboard (ASUS P4P800 SE)
seems to set the memory to run at 320MHz instead of 400MHz, but maybe I
could get this to work after more fiddling.

3) As (2) above but run them at 3.0-3-3-8.

Also, I will be doing some experimenting with overclocking once I and
convinced that my system is stable. With option (2), my memory would
already be pushing it so maybe this would limit and further
overclocking such as raising the FSB.

 

[Paul]

I have two systems, one is the prime machine, the other runs linux and
just takes all of the bits left over from my main PC when I upgrade. I
have the following memory to be distributed across two machines (I want
2GB on the prime, 1GB on the secondary):

1x 1GB DDR400 PC3200U 3.0-3-3-8@200MHz
2x 512MB DDR400 PC3200 (?U) 3.0-3-3-8@200MHz
2x 512MB DDR400 PC3200 (?U) 2.5-3-3-8@200MHz

I'm not sure what the U means in the above or whether it's relevent.

The options are as follows:

1) Put all the CL3 memory in the prime machine because they are all the
same timings. However will the two 512MB modules be able to run in dual
mode with the odd 1GB module inserted? If not, will I take much of a
performance hit?

2) Put all 512MB modules in the prime and run them all at 2.5-3-3-8. I
would run memtest86 over a weekend to verify that it's stable. I've
done some experimenting with this but my motherboard (ASUS P4P800 SE)
seems to set the memory to run at 320MHz instead of 400MHz, but maybe I
could get this to work after more fiddling.

3) As (2) above but run them at 3.0-3-3-8.

Also, I will be doing some experimenting with overclocking once I and
convinced that my system is stable. With option (2), my memory would
already be pushing it so maybe this would limit and further
overclocking such as raising the FSB.

If the Prime machine is the P4P800 SE, it takes four sticks
without penalty. When I tested my machine, with 2x512MB and
a second test with 4x512MB matched sticks, memtest86+ reported
exactly the same memory bandwidth. Due to supporting more open
pages of memory, it should have reported that the 4x512MB config
was better, but I guess the addressing pattern of memtest86 doesn't
benefit from open pages. If the access pattern of Prime95 is
different than memtest86+, perhaps it will benefit from using
the four sticks.

"Intel 865 Chipset Memory Configuration Guide"
ftp://download.intel.com/design/chipsets/applnots/25303601.pdf

So anyway, if the P4P800 SE is the Prime machine, set it to
CAS3 (the slower of the two types of RAM used), install your
2x512MB CAS2.5 and 2x512MB CAS3. The diff between CAS2.5 and
CAS3 is not worth fighting over.

There is a table in your manual like this. If the P4P800 SE
has an FSB533 processor, then it will be limited to the "DDR333"
setting. If the processor is FSB800, then the memory can go
to the full DDR400. This limitation is determined by the
dividers inside the Northbridge. If that leaves you feeling
bummed, you can always overclock your FSB533 processor a bit.

    CPU FSB      DDR DIMM Type           Memory Frequency
    800 MHz      PC3200/ PC2700*/PC2100  400/333 */266 MHz
    533 MHz      PC2700/PC2100           333/266 MHz
    400 MHz      PC2100                  266 MHz

    *When using 800MHz CPU FSB, PC2700 DDR DIMMs may run only at
    320MHz (not 333MHz) due to chipset limitation.

That leaves you with the 1x1GB for the other machine, and as long
as that DIMM uses 64Mx8 chips, it should be happy in just about
any motherboard that claims to support 1GB sticks. Even a picky
AMD S754 board should be happy with the 1x1GB. (Would've helped
if you said what the second motherboard was.)

If you own an 875/865 motherboard, and you also own either a S754
or S939 motherboard for the other one, the massive numbers of
sticks go into the Intel board. The Athlon64 works best with
one stick per channel (2 sticks S939, or 1 stick S754).

If both motherboards are dual channel, then find another
1GB stick, to match the one you've got. Otherwise, one of
the motherboards is going to suffer by being run in a
single channel mode.

Paul

 

[octessence]

Thanks for the quick response. The CPU runs at 800MHz FSB so I'll try
running all the memory modules at 400MHz with CL3.0. I think I need to
fiddle a bit but it was confusing that my motherboard decided to run
them at 320MHz when this doesn't match any of my hardware.

I'm not too worried about the second machine running in single channel
mode. It's just a minimal linux box running my web server, ftp,
downloads etc so it never gets very stressed.

I suppose the third option is to hunt around all the PCs in work and
find more CL2.5 memory to swap my CL3.0 stuff with. Nobody would ever
know but it's probably not worth my job .

I hadn't heared of Prime95 so I'll give that a try as well when I'm
running my burn in tests.

 

[paulmd]

Thanks for the quick response. The CPU runs at 800MHz FSB so I'll try
running all the memory modules at 400MHz with CL3.0. I think I need to
fiddle a bit but it was confusing that my motherboard decided to run
them at 320MHz when this doesn't match any of my hardware.

I'm not too worried about the second machine running in single channel
mode. It's just a minimal linux box running my web server, ftp,
downloads etc so it never gets very stressed.

I suppose the third option is to hunt around all the PCs in work and
find more CL2.5 memory to swap my CL3.0 stuff with. Nobody would ever
know but it's probably not worth my job .

I hadn't heared of Prime95 so I'll give that a try as well when I'm
running my burn in tests.

Prime95 is a prime number generator. It's not designed to test
hardware. It just happens to be good at it, more or less by accident.
It won't actually tell you much, if if fails, just "hardware failure."

 

[Paul]

Thanks for the quick response. The CPU runs at 800MHz FSB so I'll try
running all the memory modules at 400MHz with CL3.0. I think I need to
fiddle a bit but it was confusing that my motherboard decided to run
them at 320MHz when this doesn't match any of my hardware.

I'm not too worried about the second machine running in single channel
mode. It's just a minimal linux box running my web server, ftp,
downloads etc so it never gets very stressed.

I suppose the third option is to hunt around all the PCs in work and
find more CL2.5 memory to swap my CL3.0 stuff with. Nobody would ever
know but it's probably not worth my job .

I hadn't heared of Prime95 so I'll give that a try as well when I'm
running my burn in tests.

For estimates of the effects of CAS, there is this article.
Winrar seems to be the most impacted by the CAS setting. Many
other programs don't even notice the difference. It shouldn't
take you long, to use your CAS2.5 pair of sticks, and bench
Prime while running CAS2.5 and then set the BIOS to CAS3
and try a second run.

http://www.tomshardware.com/2004/01/19/ups_and_downs/page10.html

If you are using the Asus built-in overclocking function,
where it says "10% overclock" or the like, that BIOS function
can turn down the memory clock. Basically, the Asus version
of built-in overclock, tries to not stress the memory, so
the memory ends up turned down more than necessary. Most manual
overclockers, would adjust both their memory conditions and core
conditions, for as much as they could take

If you avoid the Asus automatic function, and set the CPU clock
manually, you'll have more control. Use CPUZ (cpuid.com) to
monitor what is happening to both the core and the memory
frequency, as you make your adjustments. Take small steps
in frequency at first, so you can figure out what is happening.

On my P4C800-E, the memory speed is scaled by the CPU clock
setting. If I were to change my CPU clock from 200MHz (FSB800)
to 250MHz (FSB1000), the memory speed would go from DDR400 to
DDR500 (even though the BIOS would still display "DDR400" as
the speed). If I turned the memory down, and set it to "DDR320",
then the scaling would make that an actual value of DDR400. So you
can use the memory setting, to compensate for the increase caused
by changing the CPU clock. (The P4P800 SE may not work like that.
I don't know if the BIOS memory speed setting is "true" on that
motherboard, or is bogus like it is on my motherboard. CPUZ will
tell you whether the BIOS setting is a "true" one or not. I don't
believe the BIOS behavior is the same across all Asus 875/865
boards, so it pays to test it.)

HTH,
Paul

 

[octessence]

I was playing around last night and I've got all the memory running at
CL3.0. There was another option hidden away from the other memory
settings that was automatically setting the speed. I'm going to try
CL2.5 when I have more time to see if that is stable and how much
performance I gain.

As for the "Automatic Overclocking" functionailty of the motherboard, I
think that I'd prefer to overclock manually so that I know exactly
what's going on. I'm not really up to speed with overclocking yet but
I'm learning. I'm not sure how much I can push the FSB speed before
things start failing or how to tell whether it's the CPU, Memory or
PCI/AGP slots that fail first. I suppose one way would be to underclock
the CPU and memory and see how far I can increase the FSB speed, but I
want to understand everything a bit more before getting my hands dirty.