"Marc Guyott" said:
Greetings all!
I am wondering how much of a performance boost you see if you use PC2700
instead of PC2100?
My motherboard, an ASUS P4V8X-X, supports either 1GB of PC2700 or 1.5GB of
PC2100 and I can't decide which is the better way to go. The PC is used for
online gaming (I play DAoC (-: ), some light software development and then
lots of Word and Excel as well as Outlook/email and web browsing. So I
don't think that it is overly important to have the 1.5GB of memory but if
the added memory generally offsets the slower speed then I might go that
way. Right now I am leaning towards the 1GB of PC2700. Does anyone have
any experience or insights into which setup would generally perform better?
Thanks for any help!
Marc
Don't forget that memory is backward compatible.
You can run a stick of PC3200 rated memory, at PC2700 or PC2100.
That means you could buy two sticks of PC3200 today, plug them
in and run them as 1GB at the PC2700 rate (DDR333 transfer speed),
and then, if you check the percent memory being used and find
you are actually running out, add a third stick of PC3200, and
turn the memory speed down another notch to DDR266.
Buying actual PC2100 memory at this point in time, would
reduce the ability to resell the memory later. With the PC3200,
you could even move the memory to another motherboard later, that
could use the full PC3200 rate. (PC3200 is the fastest JEDEC
approved DDR transfer rate - faster memories are used for
overclocking applications.)
The only time it pays to buy exactly the rated speed of memory,
is if you have a motherboard with absolutely no BIOS support
for setting memory speed options. There are a few -mx or -vm
boards, where the motherboard relies on the info stored in the
SPD chip on the DIMM, and for those, buying a faster memory
can sometimes cause a problem. Your motherboard has a few
adjustments, so that won't be a problem.
If you already have the board, it pays to check the allowed
relationships between the CPU clock (and resultant FSB speed)
setting, and what speed the memory controller can be run at.
For example, it could be that at FSB533, the RAM options are
DDR333 and DDR266. And at FSB400, the RAM options are
DDR266 and DDR200. If you have an FSB400 processor, you might
be worried about nothing, if in fact the memory controller
cannot be run faster than DDR266 anyway. Some manuals have
an explicit table containing the information, while with others
you have to search far and wide to get the information before
you buy the motherboard.
There are few scenarios, where you'll see a difference between
1GB and 1.5GB of memory. The extra memory can be used as a file
cache by the OS, and I suppose that might speed up a compile
or something.
If you use a Linux OS, it is possible you might find ways to
use the extra memory. For example, when I boot Knoppix (the
"read-only" Linux), and use the "toram" option, the entire
Knoppix CD is stored in RAM, to speed access for searches
and the like. The Knoppix CD can then be removed from the
drive, and then the OS has no rotating media at all.
If I have a machine with 1.5GB of memory, and the CD takes
up 700MB, then the remainder of the memory is then just
enough for a comfortable environment.
So, you can buy two sticks now, and monitor the percentage
of memory used, to figure out whether a third stick can
be profitably used. If you buy PC3200, you'll have a
"maximum flexibility" RAM in your hand (runs at DDR400,
DDR333, DDR266, DDR200).
To estimate performance, remember that your processor has
an L1 and L2 cache. For non-memory intensive applications,
most of the memory references will be answered by the cache,
and in those cases, the speed difference won't be that
noticable. If, on the other hand, you do stuff where long
bursts of memory are accessed in a row (Photoshop), then
the memory speed difference will be noticable. I use a
1/3 rule for estimation - a 30% difference in memory bandwidth
will cause a 10% difference in application performance, for
whatever value there is in silly rules. Using the ratio of
DDR333 to DDR266 gives 1.25 or a 25% difference in raw clock
speed. The CAS of the memory will improve by a notch, so the
bandwidth difference will be less than 25%. Taking a third
of that, means an 8% estimated application difference, where
again, in a non-memory intensive application, you might not
notice that at all.
On motherboards with an integrated graphics controller, the
rules are a bit different, as the graphics controller also
relies on the memory bandwidth. I've noticed a bit of a
difference on an integrated graphics machine, in terms of
how snappy the desktop performance is, with differences in
memory bandwidth. With your AGP card and its self contained
frame buffer, that is a non-issue.
HTH,
Paul
