NF7-S v2: mixing RAM CAS types okay ?

I have an NF7-S v2 system, with an Athlon XP 2600 (Barton) and 512 Mb
Corsair ValueSelect CAS 3-3-3-8 RAM. It all works perfectly well (no
overclocking) but I'd like to add another 512 Mb RAM.

The closest match in the stock list of my usual supplier is a 512 MB
stick of Corsair ValueSelect CAS 2.5 RAM - and I'm wondering whether
there's any problem with mixing the latencies like that.

The first 512Mb is 2x256Mb installed in the first two slots with Dual
Channel mode enabled, and I'd plan to leave them there and put the new
DIMM in the third slot.

Does anyone know of any reason why this marriage should not take place
:-) ?

TIA
Nick Boyce

Nick Boyce wrote:
> I have an NF7-S v2 system, with an Athlon XP 2600 (Barton) and 512 Mb
> Corsair ValueSelect CAS 3-3-3-8 RAM. It all works perfectly well (no
> overclocking) but I'd like to add another 512 Mb RAM.
>
> The closest match in the stock list of my usual supplier is a 512 MB
> stick of Corsair ValueSelect CAS 2.5 RAM - and I'm wondering whether
> there's any problem with mixing the latencies like that.
>
> The first 512Mb is 2x256Mb installed in the first two slots with Dual
> Channel mode enabled, and I'd plan to leave them there and put the new
> DIMM in the third slot.
>
> Does anyone know of any reason why this marriage should not take place
> :-) ?
>
> TIA
> Nick Boyce
shouldnt be a problem....just that both sticks will run at the slower speed.

-Dylan C

"Nick Boyce" <> wrote in message
news:...
>I have an NF7-S v2 system, with an Athlon XP 2600 (Barton) and 512 Mb
> Corsair ValueSelect CAS 3-3-3-8 RAM. It all works perfectly well (no
> overclocking) but I'd like to add another 512 Mb RAM.
>
> The closest match in the stock list of my usual supplier is a 512 MB
> stick of Corsair ValueSelect CAS 2.5 RAM - and I'm wondering whether
> there's any problem with mixing the latencies like that.
>
> The first 512Mb is 2x256Mb installed in the first two slots with Dual
> Channel mode enabled, and I'd plan to leave them there and put the new
> DIMM in the third slot.
>
> Does anyone know of any reason why this marriage should not take place
> :-) ?
>
> TIA
> Nick Boyce


You don't mention speed. If you are currently running at 200 MHz FSB
(400 MHz DDR2) to run PC-3200 memory sticks in dual-channel mode in the
2 slots, you'll have to reduce FSB down to 166 MHz (333 MHz DDR2) to use
all 3 slots. That's what you will lose when you go to 3 slots from 2
slots. That's why I removed the 2 sticks and got 2 bigger ones so that
I did not have to reduce my FSB which far more greatly affects memory
performance and a different in CAS from 2.5 to 3.

Read the specs section in the manual, which says:

Supports 3 DIMMs DDR 200/266/333 (Max 3GB)
Supports 2 DIMMs DDR 400

Those are mutually exclusive. With 2 sticks, you get dual-channel mode
AND 400 MHz (200 FSB). With 3 sticks, FSB reduces to 333 MHz (166 FSB).
Your memory performance will get hit hard. Other than benchmarks,
you'll be hard pressed to notice any real-world difference between CAS
2.5 and 3, so run with a CAS setting of the slowest stick (i.e., highest
CAS value).

Also, with 3 sticks, you still get dual-channel mode for 2 slots but not
for the 3rd slot which also impacts memory performance. Check how much
virtual memory (pagefile) you are using. If you use lots of it, adding
more memory even if it has to run at a slower FSB may show a boost in
responsiveness in the system because less of the programs then have to
be on the slower hard drive, but then simply deciding what you really
need running and disabling the rest might accomplish a reduction in
pagefile usage, too. Also, in Windows, you can define different
hardware profiles and under each you can define which NT services get
loaded. So you could have a "game" hardware profile with many NT
services disabled so they don't consume memory.

On Tue, 20 Feb 2007 22:10:21 -0600, "Vanguard" <>
wrote:

>"Nick Boyce" <> wrote in message
>news:.. .
>>
>> The first 512Mb is 2x256Mb installed in the first two slots with Dual
>> Channel mode enabled, and I'd plan to leave them there and put the new
>> DIMM in the third slot.
>>
>> Does anyone know of any reason why this marriage should not take place
[...]
>You don't mention speed.

Yes, sorry - I should have mentioned more details: the existing RAM is
PC3200 but I'm more interested in running cool (=quiet) than fast so I
left it running at the default 333 MHz (even tho the V2 mobo can do
400MHz). It's all so much faster than my older machines that I really
don't mind

>If you are currently running at 200 MHz FSB
>(400 MHz DDR2) to run PC-3200 memory sticks in dual-channel mode in the
>2 slots, you'll have to reduce FSB down to 166 MHz (333 MHz DDR2) to use
>all 3 slots.

So I guess I won't actually lose any FSB ...

>Also, with 3 sticks, you still get dual-channel mode for 2 slots but not
>for the 3rd slot which also impacts memory performance.

.... but will have to accept the hit of one stick not operating in
dual-channel mode - I think that'll be ok (better than staying as is).

>Check how much virtual memory (pagefile) you are using ... simply deciding
>what you really need running and disabling the rest might accomplish a
>reduction in pagefile usage, too.

Um, that's very variable, depending on what I'm doing (sometimes just
surfing, sometimes doing multimedia stuff). Usually I have spare
physical memory ... but I'm starting to make more use so I think an
increase will be useful.

>Also, in Windows ...

Ah .. I multiboot, and am in Linux a lot :-)

Thanks for your very helpful post (and thanks to Dylan too) - it's
good to know I'm not about to do anything completely stupid :-}
I may curdle my spread-spectrum - whatever that is :-) - but the
machine will still run fine.

Cheers,
Nick Boyce

"Nick Boyce" <> wrote in message
news:...
> On Tue, 20 Feb 2007 22:10:21 -0600, "Vanguard" <>
> wrote:
>
>>"Nick Boyce" <> wrote in message
>>news:. ..
>>>
>>> The first 512Mb is 2x256Mb installed in the first two slots with Dual
>>> Channel mode enabled, and I'd plan to leave them there and put the new
>>> DIMM in the third slot.
>>>
>>> Does anyone know of any reason why this marriage should not take place
> [...]
>>You don't mention speed.
>
> Yes, sorry - I should have mentioned more details: the existing RAM is
> PC3200 but I'm more interested in running cool (=quiet) than fast so I
> left it running at the default 333 MHz (even tho the V2 mobo can do
> 400MHz). It's all so much faster than my older machines that I really
> don't mind
>
>>If you are currently running at 200 MHz FSB
>>(400 MHz DDR2) to run PC-3200 memory sticks in dual-channel mode in the
>>2 slots, you'll have to reduce FSB down to 166 MHz (333 MHz DDR2) to use
>>all 3 slots.
>

Nonsense. That's nvidia's GUARANTEED minimum spec. Both my Abit NF7x boards
could run 200+ with all 3 DIMM slots populated, with tight memory timings.


> So I guess I won't actually lose any FSB ...
>

>>Also, with 3 sticks, you still get dual-channel mode for 2 slots but not
>>for the 3rd slot which also impacts memory performance.

Depends on how the memory is arranged

> ... but will have to accept the hit of one stick not operating in
> dual-channel mode - I think that'll be ok (better than staying as is).

Your description of the memory on the board doesn't match Abit's labelling
of the DIMM slots or the generic manual for nforce2 (nf7x) boards. I would
have thought it highly unlikely that Abit altered this for your v2 board.

In any case, the DIMM slot near the edge of the board (DIMM3) makes 1 memory
channel, the pair closest to the CPU make up the other. If you put the 512MB
module in DIMM3 and the 2 256MB in DIMM 1 and 2 you should have dual channel
operation for ALL of your memory.

<snip>

On Thu, 22 Feb 2007 07:36:32 GMT, "ned ludd" <>
wrote:

>> On Tue, 20 Feb 2007 22:10:21 -0600, "Vanguard" <>
>> wrote:
>>
>>>If you are currently running at 200 MHz FSB
>>>(400 MHz DDR2) to run PC-3200 memory sticks in dual-channel mode in the
>>>2 slots, you'll have to reduce FSB down to 166 MHz (333 MHz DDR2) to use
>>>all 3 slots.
>
>Nonsense. That's nvidia's GUARANTEED minimum spec. Both my Abit NF7x boards
>could run 200+ with all 3 DIMM slots populated, with tight memory timings.

Hmm ... I just dug out the manual ... on page 1-1 ("Features &
Specifications") it says
Supports 3 DIMMs DDR 200/266/333 (Max 3Gb)
Supports 2 DIMMs DDR 400 (Max 2Gb)
and on page 2-3 ("Memory Installation") it just says you can't exceed
2Gb if you run it at 400Mhz, but can have up to 3Gb at 333MHz.

Anyway, I'm already running the RAM at 333MHz. I won't be exceeding
2Gb so I guess I could try ramping up the speed to 400MHz ... if it
refuses then it's not the end of the world.

>>>Also, with 3 sticks, you still get dual-channel mode for 2 slots but not
>>>for the 3rd slot which also impacts memory performance.
>
>Depends on how the memory is arranged
>
>Your description of the memory on the board doesn't match Abit's labelling
>of the DIMM slots or the generic manual for nforce2 (nf7x) boards. I would
>have thought it highly unlikely that Abit altered this for your v2 board.

Sorry - I was a bit hazy there .. "it's all a long time ago" :-)
I just opened the case: I have the two existing DIMMs in slots 3 & 2
(furthest two from the CPU). I seem to recall reading somewhere at
the time that was necessary to enable dual-channel, though I can't
find that instruction now.

The slots are positioned like this appalling ASCII art :
1 2 3
+-----+
| CPU | I I I
+-----+

I can say for certain that the POST messages always include a message
saying the speed is 333MHz, and that (in bold font) "Dual Channel is
enabled".

>In any case, the DIMM slot near the edge of the board (DIMM3) makes 1 memory
>channel, the pair closest to the CPU make up the other.

I'm puzzled now: that would mean I should currently be using slots 1 &
2 for dual-channel, but I seem to have it working with slots 2 & 3.
Weirdly tho, slots 1 & 2 *are* much closer together than slots 2 & 3,
which would tend to support your statement.

>If you put the 512MB module in DIMM3 and the 2 x 256MB in DIMM 1 and 2
>you should have dual channel operation for ALL of your memory.

That would be *wonderful* ...
I've ordered the RAM but it hasn't arrived yet - I guess all I can do
is wait till it gets here and then try.

Thanks for the input.
Nick Boyce

"ned ludd" <> wrote in message
news:4KbDh.2095$...
>
>>>If you are currently running at 200 MHz FSB
>>>(400 MHz DDR2) to run PC-3200 memory sticks in dual-channel mode in
>>>the
>>>2 slots, you'll have to reduce FSB down to 166 MHz (333 MHz DDR2) to
>>>use
>>>all 3 slots.
>>
>
> Nonsense. That's nvidia's GUARANTEED minimum spec. Both my Abit NF7x
> boards
> could run 200+ with all 3 DIMM slots populated, with tight memory
> timings.

I saw no inference that the OP was interested in overclocking nor in
testing over prolonged periods to determine maximum usable specs that
still provide *stable* operation.

>> ... but will have to accept the hit of one stick not operating in
>> dual-channel mode - I think that'll be ok (better than staying as
>> is).
>
> Your description of the memory on the board doesn't match Abit's
> labelling
> of the DIMM slots or the generic manual for nforce2 (nf7x) boards. I
> would
> have thought it highly unlikely that Abit altered this for your v2
> board.
>
> In any case, the DIMM slot near the edge of the board (DIMM3) makes 1
> memory
> channel, the pair closest to the CPU make up the other. If you put the
> 512MB
> module in DIMM3 and the 2 256MB in DIMM 1 and 2 you should have dual
> channel
> operation for ALL of your memory.

Yep, there are 2 controllers (channels MCO and MC1), not 3: one for
slots 1/2, another for slot 3. One controller handles slot 3. The
other controller handles BOTH slots 1 and 2. From nVidia's own
whitepaper:

"The nForce IGP’s/SPP’s TwinBank Memory Architecture eliminates system
memory as a bottleneck by providing a 128-bit wide DDR 266MHz memory
access path. This is implemented through dual-independent, 64-bit memory
controllers, backed by a single master arbiter. The end result: 4.2GB/s
peak memory bandwidth with minimum system latency. TwinBank’s radical
crossbar memory controller enables the CPU and GPU to access the two
64-bit memory banks concurrently, fully utilizing memory bandwidth."

You can read more at http://www.nvidia.com/object/LO_20010528_5545.html.
There are only 2 memory banks. Each bank gets its own memory controller
rather than sharing one. There are 3 slots. Obviously 2 of the slots
are sharing a bank and its independent controller.

The slot combinations are:

dual-channel (keeping separate banks 1 & 2):
slots 1 & 3 or 2 & 3
(i.e., slot 3 and either 1 or 2 but not both 1 and 2)

single-channel mode (sharing on bank 1):
1 & 2 & 3 (all 3 populated, 1 & 2 on same bank) <-- ?
1 & 2 (same bank)
1, 2, or 3 (1 stick, 1 bank)

Asymmetric dual-channel mode is supported. That means that the sticks
do not have to be equal in capacity; e.g., you could have a 512MB stick
in slot 3 and a 256MB stick in slot 2. However, the "dual channel" mode
is limited to 2 times the size of the smallest DIMM. In this example,
you would get 2x256MB in dual-channel mode and the last half of the
512MB stick is accessed in single-channel mode. Populating both slots 1
and 2 with 256MB sticks so they had the same capacity as the 512MB stick
in slot 3 *might* get dual-channel mode to work across the 512MB range
of both banks. I don't know how Abit implemented a bank that consists
of multiple modules. So it may be possible for the OP to move his
memory so slot 1 & 2 have the 256MB sticks and slot 3 has the 512MB
stick to still get dual-channel over the whole 512MB range of each bank.

Other than memory benchmarks, like 2%, I haven't seen any real-world
performance gain from dual-channel mode. I have heard that it helps
when you use onboard video (i.e., you get a mobo with the IGP
southbridge instead of SPP). I've never bothered with any mobos with
onboard video (too limiting in performance, features, compatibility, and
upgradability). I wouldn't worry [much] about losing dual-channel mode
anymore than using CAS-2.5 versus CAS-3 but I would be concerned about
having to run [reliably] at a slower clock speed (FSB).

"Nick Boyce" <> wrote in message
news:...
> On Thu, 22 Feb 2007 07:36:32 GMT, "ned ludd" <>
> wrote:
>
>>> On Tue, 20 Feb 2007 22:10:21 -0600, "Vanguard" <>
>>> wrote:
>>>
>>>>If you are currently running at 200 MHz FSB
>>>>(400 MHz DDR2) to run PC-3200 memory sticks in dual-channel mode in the
>>>>2 slots, you'll have to reduce FSB down to 166 MHz (333 MHz DDR2) to use
>>>>all 3 slots.
>>
>>Nonsense. That's nvidia's GUARANTEED minimum spec. Both my Abit NF7x
>>boards
>>could run 200+ with all 3 DIMM slots populated, with tight memory timings.
>
> Hmm ... I just dug out the manual ... on page 1-1 ("Features &
> Specifications") it says
> Supports 3 DIMMs DDR 200/266/333 (Max 3Gb)
> Supports 2 DIMMs DDR 400 (Max 2Gb)
> and on page 2-3 ("Memory Installation") it just says you can't exceed
> 2Gb if you run it at 400Mhz, but can have up to 3Gb at 333MHz.
>
> Anyway, I'm already running the RAM at 333MHz. I won't be exceeding
> 2Gb so I guess I could try ramping up the speed to 400MHz ... if it
> refuses then it's not the end of the world.
>

I shouldn't have mentioned this since it only applies if you intend to
overclock both your memory and CPU. One of the quircks of the nf2 chipset
was that although asyncronous memory operation was supported eg. running FSB
at 166 and memory at 200, memory performance reputedly suffered greatly.
It's too long ago to remember by how much but I suspect you'll see little or
no gain by running memory at 200 and with the testing needed to ensure
stable operation, it hardly seems worthwhile for these ancient boards. Mine
has long ago been relegated to internet and WP duties.

>>>>Also, with 3 sticks, you still get dual-channel mode for 2 slots but not
>>>>for the 3rd slot which also impacts memory performance.
>>
>>Depends on how the memory is arranged
>>
>>Your description of the memory on the board doesn't match Abit's labelling
>>of the DIMM slots or the generic manual for nforce2 (nf7x) boards. I would
>>have thought it highly unlikely that Abit altered this for your v2 board.
>
> Sorry - I was a bit hazy there .. "it's all a long time ago" :-)
> I just opened the case: I have the two existing DIMMs in slots 3 & 2
> (furthest two from the CPU). I seem to recall reading somewhere at
> the time that was necessary to enable dual-channel, though I can't
> find that instruction now.
>
> The slots are positioned like this appalling ASCII art :
> 1 2 3
> +-----+
> | CPU | I I I
> +-----+
>
> I can say for certain that the POST messages always include a message
> saying the speed is 333MHz, and that (in bold font) "Dual Channel is
> enabled".
>
>>In any case, the DIMM slot near the edge of the board (DIMM3) makes 1
>>memory
>>channel, the pair closest to the CPU make up the other.
>
> I'm puzzled now: that would mean I should currently be using slots 1 &
> 2 for dual-channel, but I seem to have it working with slots 2 & 3.
> Weirdly tho, slots 1 & 2 *are* much closer together than slots 2 & 3,
> which would tend to support your statement.

The pair of slots (DIMM 1 & 2) make a single memory channel, a memory stick
in each will only give you single channel operation. So Abit's
recommendation to fill the slots 3,2,1 in order makes sense. As soon as you
add the second stick you are in dual channel mode.
>
>>If you put the 512MB module in DIMM3 and the 2 x 256MB in DIMM 1 and 2
>>you should have dual channel operation for ALL of your memory.
>
> That would be *wonderful* ...
> I've ordered the RAM but it hasn't arrived yet - I guess all I can do
> is wait till it gets here and then try.
>
> Thanks for the input.
> Nick Boyce

On Thu, 22 Feb 2007 23:43:29 -0600, "Vanguard" <>
wrote:

>"ned ludd" <> wrote in message
>news:4KbDh.2095$...
>>
>> In any case, the DIMM slot near the edge of the board (DIMM3) makes 1
>> memory
>> channel, the pair closest to the CPU make up the other. If you put the
>> 512MB
>> module in DIMM3 and the 2 256MB in DIMM 1 and 2 you should have dual
>> channel
>> operation for ALL of your memory.
>
>Yep, there are 2 controllers (channels MCO and MC1), not 3: one for
>slots 1/2, another for slot 3. One controller handles slot 3. The
>other controller handles BOTH slots 1 and 2. From nVidia's own
>whitepaper:
>
>"The nForce IGP’s/SPP’s TwinBank Memory Architecture eliminates system
>memory as a bottleneck by providing a 128-bit wide DDR 266MHz memory
>access path. This is implemented through dual-independent, 64-bit memory
>controllers, backed by a single master arbiter ... enables the CPU and GPU
>to access the two 64-bit memory banks concurrently ... "
>
>You can read more at http://www.nvidia.com/object/LO_20010528_5545.html.

Interesting paper ...

>The slot combinations are:
>
>dual-channel (keeping separate banks 1 & 2):
> slots 1 & 3 or 2 & 3
> (i.e., slot 3 and either 1 or 2 but not both 1 and 2)
>
>single-channel mode (sharing on bank 1):
> 1 & 2 & 3 (all 3 populated, 1 & 2 on same bank) <-- ?
> 1 & 2 (same bank)
> 1, 2, or 3 (1 stick, 1 bank)

Thanks for this - makes sense now - I evidently misunderstood the
disposition of slots to channels. I'll be finding out soon whether I
still get dual channel operation with all slots filled - the new RAM
arrives tomorrow.

> ... Populating both slots 1
>and 2 with 256MB sticks so they had the same capacity as the 512MB stick
>in slot 3 *might* get dual-channel mode to work across the 512MB range
>of both banks.

That's what I'll try.

>I wouldn't worry [much] about losing dual-channel mode
>anymore than using CAS-2.5 versus CAS-3 but I would be concerned about
>having to run [reliably] at a slower clock speed (FSB).

Okay - thanks again. Just so long as it doesn't run any slower ..
that's all I ask

Cheers
Nick Boyce

On Fri, 23 Feb 2007 06:46:21 GMT, "ned ludd" <>
wrote:

>"Nick Boyce" <> wrote in message
>news:.. .
>> Anyway, I'm already running the RAM at 333MHz. I won't be exceeding
>> 2Gb so I guess I could try ramping up the speed to 400MHz ... if it
>> refuses then it's not the end of the world.
..
>I suspect you'll see little or
>no gain by running memory at 200 and with the testing needed to ensure
>stable operation, it hardly seems worthwhile for these ancient boards.

Okay, thanks for that advice ... I'd certainly be perfectly happy
leaving everything at default .. like I said, I'm happy with the
machine's performance as it is ... just so long as it doesn't slow
down :-}

>> I'm puzzled now: that would mean I should currently be using slots 1 &
>> 2 for dual-channel, but I seem to have it working with slots 2 & 3.
>> Weirdly tho, slots 1 & 2 *are* much closer together than slots 2 & 3,
>> which would tend to support your statement.
>
>The pair of slots (DIMM 1 & 2) make a single memory channel, a memory stick
>in each will only give you single channel operation. So Abit's
>recommendation to fill the slots 3,2,1 in order makes sense. As soon as you
>add the second stick you are in dual channel mode.

Aah ... I evidently was more baffled than I knew :-) Thanks for the
education about how the channels are arranged. I seem to have
understood it when I built the machine and confused myself later

The new RAM arrives tomorrow, and I'm looking forward to the
experiments ...
"will it still say 'Dual Channel Enabled' at boot" ...
"Will it seem faster, or slower"
"will it [gulp] run hotter"
That'll brighten up a dull Monday

Cheers
Nick Boyce