Adding Memory to slots 2 & 3 of four slots slows down memory speed

[SeaJack]

I'm puzzled by certain results from Memtest68+ V5.01 (previous version 4.0 gives similar results).

I am helping a small non-profit consolidate older machines into a few good ones to upgrade from WinXp to Win7. In doing so I am putting additional memory into the machines we will upgrade. Each of the machines has four memory slots. I am finding that adding memory in the last two slots shows a much slower memory speed according to memtest86+  ???than when the first two slots are the only ones populated.

Here are some results from memtest86+ testing on an e-Machines T6528 with AMD Athlon 64 3500+ processor. Note that all memory components are
similarly spec'ed PC-3200 400 MHz parts.

1GB x 2 in slots 0 & 1: 1919 MB at 2203 MB/s (best performance)

1GB x 2 in slots 0 & 2: 1919 MB at 1186 MB/s (OK that's to be expected)

1GB x 2 in slots 0 & 1 and 500MB x 2 in slots 2 & 3 2943 MB at 1124 MB/s Speed goes from 2203 to 1124 MB/s (that's dramatic)

I've tried other combinations including 4 x 1GB using identical parts and see similar slowdowns in memory speed when all slots are populated.

I initially blamed the e-Machines architecture but now find that an HPQ DC5000S/P2.8E with a P-4 2.8 GHz processor gives a similar result. Is this result real and if so is it expected. I would expect memory speed to be uniform so long as the components are the same across the board. What surprised me the most is when I had 4 1GB identical memories installed and saw the decline in speed. That would eliminate any variation in specifications for memories as the culprit.

If the results are genuine then I'm torn between leaving the machines at 2GB vs slowing them down by adding more memory. I know that these machine have done a lot of paging under WinXP and adding more memory will alleviate the paging problem but slower memory speed may be worse, or is it?

Any thoughts are appreciated?

Thanks in advance

[Computer_Commando]

...

Here are some results from memtest86+ testing on an e-Machines T6528 with AMD Athlon 64 3500+ processor. Note that all memory components are
similarly spec'ed PC-3200 400 MHz parts.

1.  1GB x 2 in slots 0 & 1: 1919 MB at 2203 MB/s (best performance)

2.  1GB x 2 in slots 0 & 2: 1919 MB at 1186 MB/s (OK that's to be expected)

3.  1GB x 2 in slots 0 & 1 and 500MB x 2 in slots 2 & 3 2943 MB at 1124 MB/s Speed goes from 2203 to 1124 MB/s (that's dramatic)

...

4 slots indicates dual-channel architecture, usually the matched slots are the same color.  The memory must be identical to function in dual-channel mode.  Use HWInfo or CPU-Z to see if the combination you are using is using dual channel mode.

It seems you are not using all 4 slots, only 2 of the 4, correct?
It appears that slots 0 & 1 are the matched pair, are they the same color?  I've seen motherboard manufacturers screw this up.
If the 2-500MB modules are slower than the 2-1GB modules, all the memory will function at the slower speed.

[SeaJack]

Sorry to be so slow to reply to your reply.

I was using two of four slots. Slots 0 & 2 are green and 1 & 3 blue but running two identical 1GB memories in slots 0 & 2 give me the 1186 MB/s results while using 0 & 1 give me 2203 MB/s. I had used the same kind of memories in all four slots and had seen the decline in speed. Unfortunately two of the four went out in another machine and I didn't record the actual numbers when I did the test. I'm now trying to scare up four more identical parts to continue experimenting. The two 500MB parts were rated the same but just didn't give the throughput. I don't know if it helps but the mother board is an MS-7207 Version 2.0 e-machines.

Note I also timed memtest86+ in both configurations and the elapsed time to do the tests is proportional to the speed indicated by memtest.

I'll report further results when I have the required memory to do testing but do appreciate the response.

[patio]

Until you have 4 matched dual-channel stiks to test it's just guesswork.

[Geek-9pm]

Until you have 4 matched dual-channel stiks to test it's just guesswork.

Worth repeating.