Don said:
There seems an anomaly in documentation for IBM 8113 (M52 ThinkCentre)
concerning RAM sockets,
(1) Numbered (says the manual) from #1 (nearest the CPU) to #4
(2) Coloured in pairs, ## 3 and 4 black, ## 1 and 2 green. This suggests
one pair is what other mfrs. call Bank 0 and the other Bank 1. (The order
is relevant because, when RAM cards of two sizes are present, the
smaller is supposed to go into Bank 0.)
(2) Populated new with one or two 512 Mb cards in what SIW
identifies as slots 1 and 3 i.e. one green and one black.
Has anyone figured this out? How should I install one 512 Mb
and two 1 Gb RAM units for fastest operation?
Simple. Two test cases.
http://support.lenovo.com/en_US/product-and-parts/detail.page?&LegacyDocID=MIGR-60327
+----------------------+
| |
| DIMM4 |
| DIMM3 |
| |
| DIMM2 <--- 1GB |
| DIMM1 <--- 1GB |
| |
+----------------------+
First, there's no size dependency. That existed with much older designs.
It shouldn't exist on a PCI Express based system. RAM installation is
much more flexible. If you want, you could put the DIMMs in the 3/4 slots
if you want. There should *not* be a "put the small DIMM closest to the
processor" type of dependency.
Since you asked for "fastest operation", you'd start by testing that setup.
That's because, it should be the fastest, with the memory you have on hand.
A page here, says the board uses a 945G chipset. Uses DDR2 RAM. You'll
have to confirm it uses a 945G, because I couldn't find a good reference
to back it up.
http://www.memoryupgradecompany.com/memory/IBMThinkCentreM528113MemoryUpgrade.html
Now, get yourself a copy of the 945G datasheet. Look for Figure 10-1.
http://www.intel.com/Assets/PDF/datasheet/307502.pdf
When you install the 2x1GB, that's "dual channel interleaved". Center diagram.
That's the fastest config.
When you add the odd-ball 512MB DDR2 stick, the machine is going to use
"dual channel asymmetric". The chipset doesn't have "Intel Flex Memory",
by the looks of the provided information. So adding the third stick,
like this, should slow it down. You'd need something like a Sisoftware
Sandra memory bandwidth test, to try this slower config out and get
a number. You could also use SuperPI 1.5, because the slowdown is
uniform over the memory space (unlike with Flex memory slowdown, which
is non-uniform).
+----------------------+
| |
| DIMM4 <--- 512MB|
| DIMM3 |
| |
| DIMM2 <--- 1GB |
| DIMM1 <--- 1GB |
| |
+----------------------+
And if you did it like this, the machine should not care. It should
behave exactly the same as the previous diagram. This is asymmetric
as well.
+----------------------+
| |
| DIMM4 <--- 1GB |
| DIMM3 <--- 1GB |
| |
| DIMM2 <--- 512MB|
| DIMM1 |
| |
+----------------------+
So for my purposes, I'd be testing 2x1GB, then 2x1GB+512MB, two
test cases, two bandwidth benchmark runs.
Have fun testing,
Paul
