Nforce 4 memory question

[miso]

Specifically, this is for use with a Gigabyte GA-k8NF-9 mobo. Not
having built a computer in a few years, I'm a bit baffled by the term
dual channel. My understanding is dual channel refers to how the memory
controller handles the memory, not the memory itself. Yet some vendors
are marketing dual channel memory. Is this marketing or is the memory
different?

I'd like to stuff the computer with 4 512M modules. Having played the
"cas" game in the past, I'm not too concerned with speed of the memory,
but rather have a stable system.

The manual calls for DDR 400/333/266/200 DIMM. Would PC2700 suffice?

 

[David Maynard]

Specifically, this is for use with a Gigabyte GA-k8NF-9 mobo. Not
having built a computer in a few years, I'm a bit baffled by the term
dual channel. My understanding is dual channel refers to how the memory
controller handles the memory, not the memory itself.

Correct. "Dual channel" means there are two memory channels operating in
tandem for twice the bandwidth of one.

Yet some vendors
are marketing dual channel memory. Is this marketing or is the memory
different?

The technically inclined can consider it 'marketing' as they're simply two
'matched' sticks, meaning the same speed and type. For the not so
technically inclined, however, it simplifies things and provides some
assurance since not all memory sticks of even the same size are of the same
chip configuration (I.E. 'high density', 'standard density', 'low density',
etc.) much less the same timing parameters.

I'd like to stuff the computer with 4 512M modules. Having played the
"cas" game in the past, I'm not too concerned with speed of the memory,
but rather have a stable system.

The manual calls for DDR 400/333/266/200 DIMM. Would PC2700 suffice?

It'll work but DDR400 would be better.

 

[miso]

Obviously faster is better, but generally there are diminishing
returns. I'm assuming the dual channel should help make up for the
slower speed. In any event, I'm going to avoid generic memory, so I
hope that should minimize the problems.

How do you convert DDR400 to those inflated PC speeds?

 

[David Maynard]

Obviously faster is better, but generally there are diminishing
returns. I'm assuming the dual channel should help make up for the
slower speed. In any event, I'm going to avoid generic memory, so I
hope that should minimize the problems.

How do you convert DDR400 to those inflated PC speeds?

I don't understand the question. What 'inflated PC speeds'? 'Convert' what
to what?

 

[Michael Hawes]

I don't understand the question. What 'inflated PC speeds'? 'Convert' what
to what?

DDR<n> = PC<nx8>, so DDR400 = PC3200 If you use PC2700 memory your
system will run a bit slower than if you use PC3200.
Mike.

 

[Richard Dower]

Obviously faster is better, but generally there are diminishing
returns. I'm assuming the dual channel should help make up for the
slower speed. In any event, I'm going to avoid generic memory, so I
hope that should minimize the problems.

How do you convert DDR400 to those inflated PC speeds?

It's DDR memory, Double Data Rate. So it actually runs at 200Mhz but in DDR
terms it's 400Mhz.

Try Patroit memory, they have good value 1GB kits with excellent timings.

 

[David Maynard]

It's DDR memory, Double Data Rate. So it actually runs at 200Mhz but in DDR
terms it's 400Mhz.

The confusion here is "clock" vs "data rate" and using the word "actually."
The clock is 200 MHz but the data rate is 400 (by transferring data on both
clock edges). They're both "actually."

By comparison, an XP 3200+ CPU external 'clock' is 200 MHz but it runs
internally at 2200 MHz. Does anyone claim the "actual" CPU speed is 200
MHz, because that's the external clock, or do we recognize that the CPU
operates at the faster speed?

 

[Richard Dower]

The confusion here is "clock" vs "data rate" and using the word
"actually." The clock is 200 MHz but the data rate is 400 (by transferring
data on both clock edges). They're both "actually."

By comparison, an XP 3200+ CPU external 'clock' is 200 MHz but it runs
internally at 2200 MHz. Does anyone claim the "actual" CPU speed is 200
MHz, because that's the external clock, or do we recognize that the CPU
operates at the faster speed?

I was trying to explain DDR in simple terms for him, and yes CPU's run at
200 times the multiplier. 200x13 = 2600Mhz = FX-55.

 

[miso]

Thanks for both explanations. I knew this "multiplier" was covered on
Tom's hardware and I believe came about due to the manner in which
RDRAM was specified. However, I just couldn't find the right article.

Incidentally, I used to build systems and only populate half the memory
slots. However, it seems this memory architecture needs identical
memories in all four slots, so it seems you might as well fill all the
slots when building it. Should you want more memory, give the old ones
away I guess.

 

[miso]

The speed differences show up in benchmarks, but in real life PCs are
really so much faster than I need, the exclusion being manipulating
large video files. But in general, for a given dollar expenditure, I
rather have more memory than faster memory. Once you start using
virtual memory, the system is dramatically slower.

 

[David Maynard]

I was trying to explain DDR in simple terms for him,

I understand, and didn't mean to complicate things, but I think terms like
"actually", in a 'vs' context, add rather than subtract from confusion in
general and can lead to misinterpretation. Like, I can imagine someone
wondering that if DDR266 is 'actually' 133 MHz then why not use PC133 that
is 'actually' 133 Mhz?

If you see what I mean.

 

[miso]

I think most people understand the difference between clock rate and
data rate, but you have a point.

There is a similar issue with baud and bits per second, which are not
always interchangable.

 

[David Maynard]

I think most people understand the difference between clock rate and
data rate, but you have a point.

There is a similar issue with baud and bits per second, which are not
always interchangable.

You betcha. And now days they seldom are, especially when speaking of modems.

That's where the erroneous 'phone company tech' statement that they only
'guarantee' 2400 comes from with them alternately saying "2400 baud" and
"2400 bps" not knowing which is which or that there's even a difference.

 

[Ed Medlin]

Specifically, this is for use with a Gigabyte GA-k8NF-9 mobo. Not
having built a computer in a few years, I'm a bit baffled by the term
dual channel. My understanding is dual channel refers to how the memory
controller handles the memory, not the memory itself. Yet some vendors
are marketing dual channel memory. Is this marketing or is the memory
different?

I'd like to stuff the computer with 4 512M modules. Having played the
"cas" game in the past, I'm not too concerned with speed of the memory,
but rather have a stable system.

The manual calls for DDR 400/333/266/200 DIMM. Would PC2700 suffice?

Well, lower CAS ratings should not affect stability at all. IIRC, PC2700 is
DDR333 (could be wrong, but it is rare... . I have 2 512mb modules of
Samsung DDR333 that I can run at 400 but the CAS speeds go from 2 to 3 and I
find that in bandwidth tests 333 at CAS2 is just slightly better than 400 at
CAS3. The difference is so slight that in actual operation, you would not
notice. DDR400 is faster and with lower CAS rating should be more effective
and faster. If you are going to run dual channel, just make sure you buy the
same part# modules from the same manufacturer since all is not the same as
far as running dual channel. Everything has to match up. Don't fall into the
so-called "matched set" that they charge a premium for. I am not familiar
with that particular MB, but I thought that most just ran 2 modules at dual
channel. If that is the case you might think about 2 1gig modules instead if
you need to have 2gigs of memory. If a majority of your time is done
rendering video or doing CAD it might be needed, but 1gig is usually
sufficient for the majority of users.


Ed

 

[miso]

There may be less buss loading using 2 1G versus 4 512M.

If dual channel is really interleaved memory acesses, the PC doesn't
use them in parallel anyway, so I really wonder about the matching
criteria. Ram is a wicked business, so anything to improve the margin
is fair game.

These heat spreaders seem like another gimick. Having design a few
moderately high power ICs in plastic packages, most of the heat goes
out the leads, not the plastic.