Single/Dual Channel Confusion

[Jody]

Got a question on single/dual channel memory.
I'm being told different things by different suppliers..

DDR400...say goes for about 100 dollars.

I'm being told that if i get two pieces of ddr400, it
automatically runs in dual channel mode(of course, if the board
supports it).

Which leads me to wonder why they offer 'dual channel ddr memory'
from kingston that comes in two pieces at roughly twice the
price..

So, i guess i'm asking, is there 2 types of ddr/400 i should be
choosing from?
Thanks!
jody

 

[Ben Pope]

Got a question on single/dual channel memory.
I'm being told different things by different suppliers..

DDR400...say goes for about 100 dollars.

I'm being told that if i get two pieces of ddr400, it
automatically runs in dual channel mode(of course, if the board
supports it).

Depends on the chipset/board and where you out the DIMMs.

Which leads me to wonder why they offer 'dual channel ddr memory'
from kingston that comes in two pieces at roughly twice the
price..

'cos it sounds cool.

So, i guess i'm asking, is there 2 types of ddr/400 i should be
choosing from?

No. DDR400 = PC3200. Thats it. There is no "Dual Channel Memory" only
chipsets that support Dual Channel, which requires (at least) 2 DIMMs.

So don't spend extra on a Dual Channel Kit. a 1024MB Dual Channel Kit is
just 2x512MB of the same memory.

Ben

 

[Philip Callan]

So don't spend extra on a Dual Channel Kit. a 1024MB Dual Channel Kit is
just 2x512MB of the same memory.

But guaranteed to be same speed and normally same batch.

You may be right for the non-oc'er or value concious, but I bought a 1G Kit,
for the
same reason I'd buy matched CPU's (model *and* stepping) for a SMP system.

It may work for most applications or such, but kit's are normally modules
that have
been tested and stressed together as a team, and their performance
complements each
other.

 

[Ben Pope]

But guaranteed to be same speed and normally same batch.

What and 2 PC3200 modules aren't "the same speed"?

You may be right for the non-oc'er or value concious, but I bought a 1G
Kit, for the
same reason I'd buy matched CPU's (model *and* stepping) for a SMP system.

You've got exactly the same chance of 2 random PC3200 from the same
manufacturer reaching say, 220MHz with the same timings as you have with a
kit.

It may work for most applications or such, but kit's are normally modules
that have
been tested and stressed together as a team,

OK, fine - some manufacturers are supposedly testing them together. I don't
see the point though. A module that can do 200MHz can do 200MHz, since when
does a module sitting next to it have any influence (one thats capable of
200MHz, obviously)?

(ok, on some systems the address bus is shared between slots and you'll find
that sticking two modules in that could individually reach maybe 220MHz,
results in the pair only reaching maybe 215MHz, but if the modules
capacitance is that high, or their tolerances so close to the edge that only
the "best" ones will work as a pair at the rated speed, then you're buying
crap RAM and overclocking is out of the window anyway) - actually, after
re-reading that, I'm wrong. The address buses are unlikely to be shared
across channels. So there is no benefit in testing them as pairs.

and their performance complements each other.

How do they "complement each other"? I really don't see the difference,
it's not like there's some magic bond between them that allows them to
perform better because the packet has got "Dual Channel" written on it.
Look up the word "synergy" thats used by marketeers trying to sell something
to somebody too, but is a bit passe now

You're welcome to buy into the marketing BS, but it's exactly that - smelly,
stodgy and distinctly brown, with flies hanging over it

Incidentally I have a Dual Channel Kit too, but thats 'cos it was cheaper
than 2 seperate modules.

Ben

 

[Ed]

What and 2 PC3200 modules aren't "the same speed"?


You've got exactly the same chance of 2 random PC3200 from the same
manufacturer reaching say, 220MHz with the same timings as you have with a
kit.


OK, fine - some manufacturers are supposedly testing them together. I don't
see the point though. A module that can do 200MHz can do 200MHz, since when
does a module sitting next to it have any influence (one thats capable of
200MHz, obviously)?

(ok, on some systems the address bus is shared between slots and you'll find
that sticking two modules in that could individually reach maybe 220MHz,
results in the pair only reaching maybe 215MHz, but if the modules
capacitance is that high, or their tolerances so close to the edge that only
the "best" ones will work as a pair at the rated speed, then you're buying
crap RAM and overclocking is out of the window anyway) - actually, after
re-reading that, I'm wrong. The address buses are unlikely to be shared
across channels. So there is no benefit in testing them as pairs.


How do they "complement each other"? I really don't see the difference,
it's not like there's some magic bond between them that allows them to
perform better because the packet has got "Dual Channel" written on it.
Look up the word "synergy" thats used by marketeers trying to sell something
to somebody too, but is a bit passe now

You're welcome to buy into the marketing BS, but it's exactly that - smelly,
stodgy and distinctly brown, with flies hanging over it

Incidentally I have a Dual Channel Kit too, but thats 'cos it was cheaper
than 2 seperate modules.

Ben

Can't agree with you more Ben, I have 2 kits and 1 stick from each batch
fails at their spec rating, I think if Corsair actually tested these
sticks they wouldn't even have passed to begin with, let alone they
claim they were tested in pairs!

Using 1 stick from each kit works just fine at rated spec, how could
Corsair even distinguish which ones are matched?, all numbers on all 4
sticks are identical, 2 kits - same numbers, sounds fishy to me! ;p

Cheers,
Ed

 

[Ben Pope]

Can't agree with you more Ben, I have 2 kits and 1 stick from each batch
fails at their spec rating,

Proof enough that two sticks in a "matched pair" or "dual channel kit" are
no more likely to be similar than any other two sticks of the same type.

I think if Corsair actually tested these
sticks they wouldn't even have passed to begin with, let alone they
claim they were tested in pairs!

Yes, that is a bit odd... Usually Corsair modules are rated well within
spec... are they 3200LL modules? They got re-rated from 6-2-2-2 to 6-2-3-2,
IIRC. Oh, and I'm under the impression that Corsair warranty them to 2.8V,
and recommend 2.7V.

Using 1 stick from each kit works just fine at rated spec, how could
Corsair even distinguish which ones are matched?

Presumably either batch or the order they come off of the line.

all numbers on all 4
sticks are identical, 2 kits - same numbers, sounds fishy to me! ;p

Marketing BS. They do say they test them all on an A7N8X, in Dual
Channel... that would be the most laborious, time consuming and simply
rediculous testing procedure a manufacturer of a mass produced product has
ever gone to. Much more likely they're tested outside of a motherboard, on
some kind of custom testing rig as part of the production line.

Ben

 

[Ed]

Proof enough that two sticks in a "matched pair" or "dual channel kit" are
no more likely to be similar than any other two sticks of the same type.


Yes, that is a bit odd... Usually Corsair modules are rated well within
spec... are they 3200LL modules? They got re-rated from 6-2-2-2 to 6-2-3-2,
IIRC. Oh, and I'm under the impression that Corsair warranty them to 2.8V,
and recommend 2.7V.

Ya 3200LL, I tried 2.7V it didn't help any. Yes I see they changed the
specs, if I set them to 6-3-2-2 in the BIOS they work just fine, so in
order to run 3 sticks that's the settings I've been using for months now
without a single error. 3Dmark01 and memory speed tests dropped a hair,
not even enough to worry about. ;p

Presumably either batch or the order they come off of the line.


Marketing BS. They do say they test them all on an A7N8X, in Dual
Channel... that would be the most laborious, time consuming and simply
rediculous testing procedure a manufacturer of a mass produced product has
ever gone to. Much more likely they're tested outside of a motherboard, on
some kind of custom testing rig as part of the production line.

Ben

yeah, I bet they don't even have an A7N8X anywhere in site! ;p
Cheers,
Ed

 

[Jody]

Marketing BS. They do say they test them all on an A7N8X, in Dual
Channel... that would be the most laborious, time consuming and simply
rediculous testing procedure a manufacturer of a mass produced product has
ever gone to. Much more likely they're tested outside of a motherboard, on
some kind of custom testing rig as part of the production line.

Ben

Thanks guys. I was pretty sure it was something marketting wise!
jody

p.s. Ben, like your sig

 

[DaveW]

The two sticks of RAM should be MATCHED by the vendor so you don't get
timing errors. Buy the more expensive pre-matched pair.

 

[Ben Pope]

The two sticks of RAM should be MATCHED by the vendor so you don't get
timing errors. Buy the more expensive pre-matched pair.

Rubbish.

We've already demonstrated in this thread that matched pairs are no more
matched than any other two modules of the same type.

With the nForce2 the docs specifically say you can use different modules. I
don't know about Intel chipsets.

Ben

 

[Ed Medlin]

The two sticks of RAM should be MATCHED by the vendor so you don't get
timing errors. Buy the more expensive pre-matched pair.

Not necessary. Just buy 2 sticks of the same part# and you should be fine.
If you find them at the same price, then go with it. If they charge a
premium, it is just BS like Ben said. Why in the hell would they even have
part#s if they aren't the same memory with the same timings. If they
weren't, then the companies' quality control is crap.

Ed

 

[Ed Medlin]

Rubbish.

We've already demonstrated in this thread that matched pairs are no more
matched than any other two modules of the same type.

With the nForce2 the docs specifically say you can use different modules. I
don't know about Intel chipsets.

Ben

My previous post is regarding Intel, but all you need is memory with the
same part #. Like I said, if they are not the same specs/timing the company
needs to take a look at their quality control. Whether Intel or AMD the
memory used should be the same brand and part# if possible. I have seen
completely different brands work fine too, but I would at least try and get
the same brand and part # just to be safe. Intel boards tend to be a tad
more picky, but in general, they are not much different in the way they
handle the memory in dual channel.


Ed

 

[Ben Pope]

My previous post is regarding Intel, but all you need is memory with the
same part #. Like I said, if they are not the same specs/timing the
company needs to take a look at their quality control. Whether Intel or
AMD the memory used should be the same brand and part# if possible.

Why? You can use DDR400 at anywhere between 100 and 200MHz, with timings
down to their rated specs. So why when in Dual Channel must you suddenly
match the two DIMMS? As long as you don't attempt to run either DIMM out of
spec in terms of speed and timings, how would they behave differently to two
"matched" pairs?

Admittedly, if you're buying new you might as well buy two matching DIMMs,
but that does not mean you HAVE to have a "Dual Channel kit".

If you have two modules, one PC2100 CAS3 and on PC3200 CAS2, and run the
whole lot at 133MHz CAS3, are you saying you have less chance of them
working than two "matched" PC2100 CAS3 modules?

Please give the technical details if so, as I don't see how that could
possibly be the case.

I have seen completely different brands work fine too,

Well this is hardly magic is it? And argues quite nicely against the
matched pair thing.

Are you really so influenced by marketing that your own experience can be
completely ignored?

Ben

 

[Philip Callan]

A good little link with info on Dual/Single Channel is here:

http://www.neoseeker.com/Hardware/faqs/kb/11,16.html

And like they say 3/4 or 9/10 you COULD use two unmatched sticks,
just going for same manuf, speed, timings. But some boards are
picky, and using 2 random sticks, you run the risk of getting single channel
memory, with no capability of DC.

 

[Ben Pope]

A good little link with info on Dual/Single Channel is here:

http://www.neoseeker.com/Hardware/faqs/kb/11,16.html

And like they say 3/4 or 9/10 you COULD use two unmatched sticks,
just going for same manuf, speed, timings. But some boards are
picky, and using 2 random sticks, you run the risk of getting single
channel memory, with no capability of DC.

"A good estimate would be to assume that 3 out of 4 attempts"

Well thats hardly definitive is it?

You realise that 78.2% of statistics are made up?



Ben

 

[Ed Medlin]

Why? You can use DDR400 at anywhere between 100 and 200MHz, with timings
down to their rated specs. So why when in Dual Channel must you suddenly
match the two DIMMS? As long as you don't attempt to run either DIMM out of
spec in terms of speed and timings, how would they behave differently to two
"matched" pairs?

Admittedly, if you're buying new you might as well buy two matching DIMMs,
but that does not mean you HAVE to have a "Dual Channel kit".

If you have two modules, one PC2100 CAS3 and on PC3200 CAS2, and run the
whole lot at 133MHz CAS3, are you saying you have less chance of them
working than two "matched" PC2100 CAS3 modules?

Please give the technical details if so, as I don't see how that could
possibly be the case.


Well this is hardly magic is it? And argues quite nicely against the
matched pair thing.

Are you really so influenced by marketing that your own experience can be
completely ignored?

Ben

Whoa Ben.................. I actually agree with everything you say, but
if you are buying new, I would try and get the same brand/part#. If they are
the same timings etc, I would think they should work anyway. Intel boards
CAN be more picky about it, but usually will work as long as the ram has the
same specs. You are right on the button about the marketing hype with
"matched" ram.

Ed

 

[Ben Pope]

Whoa Ben.................. I actually agree with everything you say,
but if you are buying new, I would try and get the same brand/part#. If
they are the same timings etc, I would think they should work anyway.
Intel boards CAN be more picky about it, but usually will work as long as
the ram has the same specs. You are right on the button about the
marketing hype with "matched" ram.

Yeah.. if you're buying new it wouldn't really make sense to buy two
different part #s.

I did get a little carried away, but there is so much misinformation out
there and this argument just keeps popping up every week or so.



Ben