BOINC: Performance Question (Memory)

[mondobyte]

I made a similar post over in the Processors forum but my question here is relative to BOINC:

[*]What is the effect on BOINC, assuming everything else is constant, of memory @ 400-1T vs 333-2T in the scenarios listed near the end of this post?

I have two identical motherboards, one with an Opteron 165 and another with an Opteron 175. I noticed on boot that the 175 always runs 333 2T and the 165 runs 400 1T with identical memory (the 175 has 4 512M sticks, the 165 has 2 512M sticks).

I wondered why this was happening. (It turns out that for this situation, the difference in the processors can be ignored since they are both Socket 939) Here is some very interesting information about memory configurations that I discovered!
Please examine Page 178 Tables 45 and 46 in this document. It was most enlightening! (Page 175 has information relative for Socket 754 processors)

It turns out that AMD has set some maximum upper limits on memory performance depending on your memory configuration! These constraints are embedded within processor and are probably not subject to modification.

[*]Assuming 4 total memory slots, what, if any, is the advantage of populating 2 slots with, say, 1GB memory modules @ 400-1T compared to populating 4 slots with 512MB memory modules @ 333-2T?

[*]Are all socket 939 motherboards capable of 128 bit memory access or are some limited to 64 bits ?



mondo

 

[Insidious]

When I was doing my builds on MSI MoBos, the manual showed this performance limitation in it's section about memory configurations. I just assumed it was a motherboard thing.... didn't realize it was a CPU thing.

As to memory, I had an experience where I took my DDR400 memory out to put in my son's machine until I could get more, I put 1G (2x512) of DDR333 in my machine. There was no significant dip in my Rosetta production?! :shocked:

I think that the dual core processors are much more sensitive to CPU speed than memory bandwidth.

(I didn't do any science... just noticed it and was surprised)

One thing to note, usually the timings of 1G sticks are slower than with 512MB sticks (especially CL). IMO (and it's not a well qualified opinion at that) I think you will see very little difference in performance with a pair of slower 1G sticks at 1T than you will see with 2 pairs of faster sticks at 2T.

FWIW

-Sid

 

[Rattledagger]

Originally posted by: mondobyte
[*]What is the effect on BOINC, assuming everything else is constant, of memory @ 400-1T vs 333-2T in the scenarios listed near the end of this post?

This depends on the individual projects, in some projects there's likely no difference in speed, while in others there can be a significant difference...