Distorted said:
Hi Paul,
Just like you said it worked with 4 sticks (two matching pairs) and
just like you said the bandwidth dropped and the command rate dropped
to 2T
The benchmark results with SiSoftware Sandra are:
2Gb (2x 1Gb)
RAM Bandwidth Int Buff'd iSSE2: 5155 Mb/s
RAM Bandwidth Float Buff'd iSSE2: 5141 Mb/s
3Gb (2x 1Gb & 2x 512Mb)
RAM Bandwidth Int Buff'd iSSE2: 4192 Mb/s
RAM Bandwidth Float Buff'd iSSE2: 4238 Mb/s
I don't understand why the bandwidth drops when using four sticks. (I
know you said it would). Given these results does the bandwidth
increase for 2Gb outweigh the benefits of having an extra 1Gb?
Only you can answer that question. It depends on your OS. For example,
with Vista, that new OS finds uses for available RAM, and has all
sorts of cacheing and accelerating concepts built in. Other
Windows OSes might use the extra RAM for a flushable file cache,
but otherwise might not get any real usage from the memory.
What I suggest to people, is use Task Manager (control-alt-delete)
and look at the memory performance in there. For example, running
my favorite game, if I alt-tab out of the game, and look at the
Task Manager, I can see 600MB of my 1GB of memory in use. If I
bumped my memory to 3GB, I'd just be increasing a resource I'm
not running out of.
On my current machine, in two years, I only had heavy swapping
just once. I tried to print the world's longest web page (it
takes about 10 minutes to download and is full of pictures),
and the disk just went crazy. Now, if I considered that to
be an important operation to optimize, then I'd stick 3GB in
the machine. As it is, I'm quite happy with the 2x512MB
I'm currently using.
In terms of the above penalties, your readings are
raw memory bandwidth. My rule of thumb, is a "1/3rd rule".
Say, take 4192/5155 = 0.81. That is a 19% drop in bandwidth.
If I divide the 19% number by 3, I get about 6% or so. I
would expect to see some "average" application run 6% slower
than normal. Applications like Photoshop would be more
affected. Applications that run within the cache of the CPU,
would be less affected.
You can lessen the impact of using four sticks, by tweaking
the settings. It all depends on your processor, as to whether
it could take a higher memory clock setting. To be quite honest
with you, I wouldn't use your 3GB setup - I'd go with 2x1GB
unless there was a damn good reason to do otherwise. If you
do stuff that causes swapping to the disk all the time, then
go with the 3GB configuration. For just about everything else,
a 2GB configuration should work great. Try it with 2GB for
a while, and see for yourself.
In terms of things that affect bandwidth:
1) Command rate setting. 2T is perhaps 20% slower than 1T.
It means the memory bus has fewer opportunities to send
new commands to the memory. The result is more "idle bubbles"
in the memory timing diagram.
2) DDR333 is slower than DDR400 (no surprise there).
3) CAS plays a small part. Maybe in the 5% range or so.
Other memory parameters might be less than that.
2x1GB could run DDR400, Command Rate 1T.
2x1GB+ 2x512MB could either run DDR400 Command Rate 2T or
could run at DDR333 Command Rate 1T. The
first of those options is better (as reported
by a poster here a while back).
You can go into the BIOS, and tweak the settings. Check the
current settings by using CPUZ from cpuid.com . Then try the
alternatives, by going to "Manual" in the BIOS, and only
changing the settings you understand. Then run memtest86+.
If memtest86+ is clean after a couple passes, boot into
Windows and try Prime95. If your tweaks were wrong, Prime95
Torture Test option will error out in seconds. Good settings
should withstand Prime05 Torture Test in an overnight test.
But really, for the time you'll spend doing that kind of
testing, you could be enjoying the 2x1GB config instead.
And the 2x512MB might come in handy in an emergency (like
a RAM failure at some future date).
Paul
