Dual-Channel & FSB: Calculating the bottlenecks

[Dave C.]

I've searched through the faq and did not find anything specific to
this.

There are several cases I've wanted to install memory in pairs to
achieve dual-channel memory speeds, but I've come across a caveat or
two about whether or not this is worth it.

Quick questions:
1. How do I calculate whether or not the FSB is fast enough to take
advantage of the memory speeds. I'll see cpuz refer to things like
FSB:memory as "1:2", which I'm not sure refers to the memory
controller
taking advantage of the up and down side of the clock waveform, or if
it's saying "give up, we're half the speed we need".

2. How do I calculate dual core beasts? Do they share the FSB, or do
they each have their own resulting in double access speed [attempts]
to dual ported memory (note, not dual channel).

Forgive me please, my questions may be tripping all over themselves.
It's been a longgggg time since I've calculated any of this out during
my stint as a DSP software engineer.

I'll be glad to rephrase this in any way you all like.

THANKS!

You're thinking too hard. RAM is like a spreadsheet. If you dual channel or triple channel or whatever, you are multiplying the width of the spreadsheet, so that data can be accessed quicker. However, you're only going to see a significant difference if the data being transferred is insanely huge. A regular size data file can be transferred to/from a 1GB stick of RAM pretty quickly. Your typical home system used for office apps. and web browsing and even Gaming is not going to see a bottleneck related to RAM, unless it is starving for RAM.

Unless you are regularly (regularly) accessing data files that are at least 10% of the size of any one stick of RAM, I don't think dual-channel is going to help you.

And the few experts who have run the benchmarks conclude that dual channel is not significantly faster. So there you have it. Do it, or not. Just stop fretting about it. -Dave

 

[Thomas G. Marshall]

I've searched through the faq and did not find anything specific to
this.

There are several cases I've wanted to install memory in pairs to
achieve dual-channel memory speeds, but I've come across a caveat or
two about whether or not this is worth it.

Quick questions:
1. How do I calculate whether or not the FSB is fast enough to take
advantage of the memory speeds. I'll see cpuz refer to things like
FSB:memory as "1:2", which I'm not sure refers to the memory
controller
taking advantage of the up and down side of the clock waveform, or if
it's saying "give up, we're half the speed we need".

2. How do I calculate dual core beasts? Do they share the FSB, or do
they each have their own resulting in double access speed [attempts]
to dual ported memory (note, not dual channel).

Forgive me please, my questions may be tripping all over themselves.
It's been a longgggg time since I've calculated any of this out during
my stint as a DSP software engineer.

I'll be glad to rephrase this in any way you all like.

THANKS!

 

[Thomas G. Marshall]

I've searched through the faq and did not find anything specific to
this.

There are several cases I've wanted to install memory in pairs to
achieve dual-channel memory speeds, but I've come across a caveat or
two about whether or not this is worth it.

Quick questions:
1. How do I calculate whether or not the FSB is fast enough to take
advantage of the memory speeds. I'll see cpuz refer to things like
FSB:memory as "1:2", which I'm not sure refers to the memory
controller
taking advantage of the up and down side of the clock waveform, or if
it's saying "give up, we're half the speed we need".

2. How do I calculate dual core beasts? Do they share the FSB, or do
they each have their own resulting in double access speed [attempts]
to dual ported memory (note, not dual channel).

Forgive me please, my questions may be tripping all over themselves.
It's been a longgggg time since I've calculated any of this out during
my stint as a DSP software engineer.

I'll be glad to rephrase this in any way you all like.

THANKS!

You're thinking too hard.  RAM is like a spreadsheet.  If you dual channel or triple channel or whatever, you are multiplying the width of the spreadsheet, so that data can be accessed quicker.

Ah, not if the FSB (the way the cpu gets to the memory) is not fast
enough, no it won't. All your following comments make sense, except
for the 10% part of a stick of memory<---that part I don't quite
understand.

If there is a memory bound application, whether or not it is grabbing
10% of the stick of memory, as long as it is exceeding data cache it
is going to need to read from something external.

So let's say for yucks that you're right in any case just so I
understand: is the FSB bandwidth *shared* among dual core chips or
does each have its own (say 800MHz each) accessing dual ported memory?

Note: yep, dual ported is not dual channel as a notion, I know that.