| 1 |
Isidore Ducasse <ducasse.isidore@×××××.com> posted |
| 2 |
20070514113637.3a4ba97d@Bazaar, excerpted below, on Mon, 14 May 2007 |
| 3 |
11:36:37 +0200: |
| 4 |
|
| 5 |
> Very interesting post! |
| 6 |
> Could you explain what "mobo" means? |
| 7 |
|
| 8 |
The other answers are correct, mo(ther)bo(ard). Sorry for the |
| 9 |
unexplained confusing shorthand. |
| 10 |
|
| 11 |
> And BTW (_almost_ off-topic...) I've heard that RAM sticks should be |
| 12 |
> identical when plugged on the same motherboard, but it was some "good |
| 13 |
> vendor advice" so I'd rather rely on some experienced user's answer. So |
| 14 |
> is there an issue if two RAM sticks of different brands are plugged on |
| 15 |
> the same motherboard? What if, whilst of the same brand, they don't have |
| 16 |
> the same capacity? Could Peter's issue be related to this kind of |
| 17 |
> problem? |
| 18 |
|
| 19 |
I agree with the others here, especially Florian, too. Same |
| 20 |
manufacturer's lot is a good idea when the memory is interleaved (should |
| 21 |
be a BIOS setting controlling that). Common interleaving is dual- |
| 22 |
channel, but with dual Opterons as here, quad channel is an option, |
| 23 |
interleaving the nodes as well as the sticks on the same node. However I |
| 24 |
stick with dual-channel and independent nodes, as I prefer NUMA mode |
| 25 |
(each CPU/core gets its own memory space to work with, on its physically |
| 26 |
local memory, if possible) with its independent memory access, the apps |
| 27 |
on each CPU interfering less with the other's memory, over the additional |
| 28 |
bandwidth. If it's not interleaved, it doesn't matter (or shouldn't, |
| 29 |
anyway), tho as Florian mentioned, the lower clock rate is used if the |
| 30 |
differ. |
| 31 |
|
| 32 |
The same thing applies to capacity. If one is interleaving, same |
| 33 |
capacity is required across the interleaf but not necessarily between two |
| 34 |
separate interleaves. |
| 35 |
|
| 36 |
The problem could indeed be unmatched timing related (such unmatched |
| 37 |
timing being the reason different lots, even of the same manufacturer, |
| 38 |
often causes problems when interleaving), especially if interleaving is |
| 39 |
turned on. Again, however, trying it with a stick at a time would of |
| 40 |
course eliminate such interleaving timing issues, altho if that's the |
| 41 |
issue, the memory sticks would test out fine individually. |
| 42 |
|
| 43 |
-- |
| 44 |
Duncan - List replies preferred. No HTML msgs. |
| 45 |
"Every nonfree program has a lord, a master -- |
| 46 |
and if you use the program, he is your master." Richard Stallman |
| 47 |
|
| 48 |
-- |
| 49 |
gentoo-amd64@g.o mailing list |
Archives