Otto said:
Hi,
I have a choice of memory for a Dell Optiplex 745 taking 800MHz PC6400
RAM - 2 timing specs are available: 6-6-6-6 or 4-4-4-12
Any idea if the latter (and a bit more expensive part) is actually any
faster?
Memory has two performance aspects.
Memory bandwidth is measured by "PC2-6400". The maximum rate that
information can come from the memory, is 6400MB/sec. This is useful
when large blocks of memory are being pulled into the processor.
Usually, a "burst" of data is designed to fill a "cache line" inside
the processor, so that is a normal sized transaction for the processor.
The second set of numbers concern latency. That is important for
random access. Not all operations on the computer access large
contiguous blocks of data. Sometimes, the processor effectively picks
just one byte here and one byte there. The CAS Latency defines the delay
from the request, until the data is provided. The lower the number
(your 4-4-4-12 product), the quicker the answer comes back. If you
consider the time duration of each one of those operations, and then
take the inverse of that, you can see how many "queries per second"
the lower latency memory can make. So, for programs which are
referred to as "cache busters", because the data needed never
seems to be in the cache, it is the memory subsystem performance
that determine overall program execution speed. The more
"queries per second" the memory could support, the faster the
"cache buster" program can run. An example of such a program,
is simulator programs, such as logic simulation for the building
of silicon chips.
On some current computers, these numbers don't make a big difference.
If you're mainly an "interactive" computer user, where no program
runs for more than a few seconds, you can fit the crappiest RAM
you want in the computer.
If you're shrinking DVDs for a living, you probably measure the
execution time with a stopwatch. If the 6-6-6-18 RAM took 2 hours 40 minutes,
and the 4-4-4-12 took 2 hours 30 minutes, you'd likely pay the
extra money. For some people, every little bit like that is
important.
In many cases, the only time you can see the difference,
is running benchmark programs, to brag to your friends.
HTH,
Paul