"Dragoncarer" said:
OK.
Right now I have 2x512 sticks running in dual channel. (Setup: A1:512;
B1:512)
I'm thinking of getting another GB.
If I bought 1x1GB stick, will it work? I would assume I will have to forgo
dual channel, but I can accept that. (So, I was thinking: A1:1GB; A2:512;
B1:512).
Just a quick answer, thanks.
Many TIA.
"Just a quick answer, thanks." Your question is a tough one!
With a Revision E processor (Venice), up to four sticks can run in
single channel mode. Yes, you can run 2x512MB + 1GB, but your
memory bandwidth will be reduced due to running single channel.
It won't be any worse than running a S754 board, and they
really aren't that bad.
I would buy a pair of 1GB. First, try the 2x1GB by themselves,
at DDR400 command rate 1T. Use memtest86+ or Sandra to benchmark.
Now, reenter the BIOS and set the memory speed to DDR333 and the
command rate to 2T. Save settings, then shut down. Install
2x1GB + 2x512MB, for a total of 3GB. At this point, you may or
may not uncover any BIOS bugs (as there are scattered reports of
the BIOS having trouble deciding what to do with four sticks of
memory, especially it seems, with Rev.E processors). If you
really hate how the four sticks run, the 2x1GB config is best.
If the board will run with 2x1GB + 2x512MB, with DDR333 and
Command Rate 2T, you will then definitely need Memtest86+ to
evaluate whether this config is going to work well. You may
be able to enter the BIOS and set the memory to DDR400 (command
rate has to stay at 2T). Four DIMMs will reduce the memory
bandwidth somewhat, due to Command Rate 2T, but perhaps not
as bad as a single channel mode will.
Now, the thing is, in your single channel configuration, with
three sticks, one channel will have two DIMMs on it. That
channel will mean setting Command Rate to 2T (if the BIOS
didn't do that already). You'll have the penalty of single
channel operation, along with Command Rate 2T setting.
In terms of best to worst configuration:
1) 2x1GB - allows >DDR400 operation, Command Rate 1T (at DDR400 at
least). Command Rate 2T only at high overclocks.
2) 2x1GB + 2x512MB - allows DDR400 operation with Command Rate 2T
- test thoroughly with memtest86+ and adjust
as necessary.
- no significant memory overclock possible
- if unlucky, may have to drop to DDR333.
- 20% drop in memory bandwidth benchmarks
using Command Rate 2T.
- real applications won't be affected that much
3) 1GB + 2x512MB - should do DDR400 with Command Rate 2T
- single channel mode cuts bandwidth in half
- command rate 2T gives more penalty, as above
- memory latency benchmark may actually improve
- you may feel the difference in this config...
If you try a single 1GB stick on one channel, and a single
512MB stick on the other channel, you would run virtual single
channel, but with the potential to use Command Rate 1T. Thus a
1.5GB configuration might give a bit better memory bandwidth
than (3).
Your current config of 2x512MB is the same as (1) above.
Some rough rules:
a) 1 stick per channel is best. Least restriction on memory clock.
Command Rate 2T may only be necessary at extremely high
memory clock.
b) When 2 sticks of double sided memory are installed on a
channel, that pretty well guuarantees Command Rate 2T will
be necessary at DDR400. (I.e. DDR333 Command Rate 1T or
DDR400 Command Rate 2T are likely, with the second option
being better.) Set the BIOS to DDR333 2T before installing the
second stick on the channel, to ensure the system will POST.
Bump up the rate and test with memtest86+, to see how well
your particular processor can do. Don't expect a high
overclock. If the 2 sticks on a channel are single sided,
expect to be able to push closer to (a), but you are unlikely
to own single sided memory.
c) Unmatched stick mixtures, with a Rev.E processor, will work.
Virtual single channel mode is used. Max bandwidth is cut
in half. The same signal integrity rules as in (a) and (b)
are implied. Unmatched sticks, one per channel, is slightly
better than three or four unmatched sticks, where one
channel ends up with two sticks, and behaves like (b) with
respect to memory settings.
HTH,
Paul
