"Russ" said:
Can anyone tell me if my synoptis is correct?
DDR Ram: In order to achieve dual channel, your choice of ram must be 2
stcks = to the CPU Bus Speed, ie. DDR 266(PC 2100) x2 = P4 533 FSB.
So in order to get it, take your processor FSB and divide that by 2 to get
the minimum clock frequency for the type of DDR Ram to use.
Can you figure the math??
533 FSB/2 = 266
So if you are running a motherboard with a P4 400 FSB processor with 2x DDR
333 (PC 2700) Ram, you will not for the life of you get dual channel. The
only way to accomplish this is to purchase a processor with an FSB of 533 or
800. If your DDR Ram is PC 2100 (DDR 266) and you are running a 400 FSB
processor (core frequency of 200 MHz (100 clock frequency x2), you will not
be able take advantage of the dual channel capabilities.
I hope you can understand that and please correct me if i am wrong.
Russ
"A mind is a terrible thing especially if it is ignorant"
The memory subsystem is demand based - it only carries out a transaction
when commanded by one of the interfaces on the Northbridge. The Northbridge
has an AGP interface, a processor interface (quad pumped on P4), and
some kind of interface to the Southbridge (PCI bus speed or faster). Any
one of these three interfaces can "stand in line" at the memory controller
queue, to read or write its data.
The processor front side bus likes to gulp data in bursts. But, like
any good silicon, it is patient. If it is starved for data, due to the
memory bus running too slow, it simply stalls and inserts wait states.
You lose performance. If the memory bus were to run too fast, then
the data words could be queued inside the Northbridge (at least one
burst's worth), until the processor eats them. Suffice it to say,
that there are techniques to ensure that nothing bad happens, whether
the memory is faster or slower than the processor.
So, if you put a dual channel bus on a design which doesn't use all
the bandwidth, no harm is done. You've just paid extra money for a
feature that isn't helping you.
Extra bandwidth doesn't have to go to waste - it can be used to support
the AGP interface, on those rare occasions when textures are being
fetched. (I'm not even going to mention the PCI bus, because at the
best of times it can only use a couple percent of memory bandwidth
during a DMA transfer.)
There is nothing wrong with doing the calculations, to see if the memory
subsystem is balanced with the demands of the processor. I'm only pointing
out that nothing catastrophic happens if they don't match. So, your
dual channel design can say "dual channel" all it wants, and the design
still works reliably, no matter what processor is used.
Another tiny detail about memory, is the memory has a control bus and
a data bus. The data bus doesn't run continuously - there are gaps
when no data is being transferred (perhaps because the control bus didn't
send some commands in time, or there was a resource conflict inside the
memory chips that required waiting a couple of cycles). So, while your
dual channel DDR400 memory bus might have a 6.4GB/sec theoretical
bandwidth, the practical average will be a lot less than this. The
practical average will be influenced by whether the memory is CAS2 or
CAS3 etc. So, when you do your arithmetic, it still pays to err on
the "generous" side, to compensate for the efficiency of the memory
interface. If you can afford it, a little extra bandwidth never hurts.
You should also keep this "memory fixation" in perspective. Processors
have L1 and L2 caches, which help ensure that the processor has the
info it needs, when it needs it, most of the time. This is why, if you
look at Sandra memory benchmarks and 3DMark or PCMark benchmarks, that
the 3DMark/PCMark improves so little compared to the memory benchmark.
In other words, a 10% memory improvement, might only buy you a 3%
processing improvement. This is because the processor has satisfied
most of its hunger via the cache, so it doesn't get to use the "new
and improved" bandwidth that much. To feed your appetite for numbers,
run your calculator over the data in "matrix.jpg". So, when you cannot
afford the most expensive memory, like PC4000 that has just been
released, don't feel bad. (Some people just gotta have the last
few percentage points of performance - if they are involved in
a business, they can afford to invest in good memory.)
http://www.systemcooling.com/modules.php?name=News&file=article&sid=426
http://www.systemcooling.com/images/reviews/Misc/Overclocking_Guide/matrix.jpg
The most important factors in your computer, are the speed of
the processor core and the size of the cache. While it is fun to spend
an extra couple hundred bucks buying expensive memory, the sad reality
is that the money might be better spent on a better video card (note -
you Photoshop, video editing, 3D rendering people can unplug your
ears now - you people are a special breed...).
To see how much people are willing to pay for extra cache inside their
processor, check out the Intel wholesale price list
http://www.intel.com/intel/finance/pricelist/
HTH,
Paul
