A8N-E memory in slots A1/B1 single channel

[Howard Delman]

A8N-E, Athlon64 3800, Samsung 512M DDR PC3200 (M368L6523CUS-CCC)

I have a 512M Dimm in sockets A1 & B1. At boot, the bios reports:

Memory testing: 1048576K OK
Memory information: DDR 400, single channel, 64-bit

The memory should be dual channel. Does anyone have any ideas about why
it is showing as single channel?

Howard Delman

 

[Paul]

A8N-E, Athlon64 3800, Samsung 512M DDR PC3200 (M368L6523CUS-CCC)

I have a 512M Dimm in sockets A1 & B1. At boot, the bios reports:

Memory testing: 1048576K OK
Memory information: DDR 400, single channel, 64-bit

The memory should be dual channel. Does anyone have any ideas about why
it is showing as single channel?

Howard Delman

Mod64BitMux control bit was added to Rev.E processors. Previous
revisions of Athlon64 processors had a fixation about B1 and B2
slots, when operating in 64 bit mode. Now, it seems, either An or
Bn slots can be occupied and the processor can operate in 64 bit
mode.

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF


So, at least the Mod64BitMux bit, might explain how the BIOS is able
to find a total of 1GB of memory, when you have the DIMMs in
A1 & B1. What I cannot explain for you, is why the BIOS has
not been able to detect the fact that the DIMMs are matched,
and it is possible to go into 128 bit mode (popularly misnamed
as dual channel mode).

To start, get a copy of CPUZ from www.cpuid.com, and see if
the declaration of the DIMMs matches. I think there is an
option there, to dump some part of the SPD EEPROM, and you
might see if everything looks OK there. If one of the SPD
chips is broken (is unreadable, returns all 1's or the like),
that might explain what you are seeing. The SPD chips will
not be entirely identical, as the serial number of the DIMMs
will be different, and perhaps the manufacture date, but the
timing and other physical params should match.

There have been the odd unexplained behavior with Athlon64 and
some of the Asus BIOS. For example, I recall someone having a
matched pair of brand_x DIMMs, and a matched pair of brand_y
DIMMs, which were likely to have matching topology, and
would have been nice candidates for interleaved operation
in dual channel mode. Instead, the BIOS insisted on ignoring
one of the two pairs of DIMMs entirely. You may want to
experiment with several releases of BIOS files for your
motherboard - maybe you'll find a release of BIOS that is a
bit more cooperative. Don't be afraid to try out a beta release.

Also, you may want to try running a memory benchmark, and see
if at nominal settings, your two DIMMs are yielding 3.2GB/sec
or 6.4GB/sec. It could be just the text message on the BIOS
screen is wrong

Paul

 

[Tim]

Try installing the DIMMS into A1 and A2.
The manual says on page 1-15 to use A1 and B1 then A2 and B2, but later
implies otherwise.

Let us know how you get on.
- Tim

 

[Mark A]

I have an AN8-E and here is the scoop:

Two sticks of dual channel memory go in slots A1 & B1 (1st and 3rd slots),
or A2 & B2 (2nd and 4th slots).

I use A2 & B2 to keep the sticks a little further away from the heatpipes on
my Thermalright XP-120 HSF on the CPU. It would fit in A1 & B1, but would be
a tight fit.

One thing I notice is the user manual lists M368L6523BTM as tested and
approved memory for the A8N-E, but not the M368L6523CUS. This may, or may
not, be significant.

Also, you should always buy memory that is sold by vendors in matched pairs
for dual channel use. The memory must be exactly the same in both sticks to
enable dual channel, and the better vendors (like Corsair) test them in
pairs to make sure of that.

 

[Howard Delman]

I tried the modules in A1/B1, A1/A2, and A2/B2, with no effect.

Howard

 

[Mark A]

I tried the modules in A1/B1, A1/A2, and A2/B2, with no effect.

Howard

You need memory that is certified by the vendor as a matched pair.

 

[Tim]

You need memory that is certified by the vendor as a matched pair.

That is not strictly true. Asus gives a fair summary of what is required in
the manual. Paul has also done so previously in his usual meticulous manner
with more detail than Asus.

- Tim

 

[Howard Delman]

Paul wrote:

To start, get a copy of CPUZ from www.cpuid.com, and see if the
declaration of the DIMMs matches. I think there is an option there,
to dump some part of the SPD EEPROM, and you might see if everything
looks OK there. If one of the SPD chips is broken (is unreadable,
returns all 1's or the like), that might explain what you are seeing.
The SPD chips will not be entirely identical, as the serial number
of the DIMMs will be different, and perhaps the manufacture date, but
the timing and other physical params should match.

Paul

Paul,

You hit the nail on the head. I got cpuz, and discovered that the two
dimms were not identical. Electrically, they were. But Samsung revised
the PCB, and thus changed part of the model number. Thus, from the
bios' perspective, the two dimms were different. The store exchanged
one for a second of the other, and now I have dual channels, 128 bit
memory. (BTW - The serial numbers of the two dimms are only two units
apart. That's about as matched as one could hope for.)

Thanks again.

Howard

 

[Mark A]

That is not strictly true. Asus gives a fair summary of what is required
in the manual. Paul has also done so previously in his usual meticulous
manner with more detail than Asus.

- Tim

You are correct that it is not Strictly true that two sticks of memory must
be certified by the vendor as matched pairs for dual channel access.

What I meant (and which was obvious) is that if you don't want to get
burned, and don't want to have to worry about sending memory back to the
retailer for a refund if the sticks are not exactly the same, then you
"need" memory certified and sold as matched pairs.

 

[Paul]

You are correct that it is not Strictly true that two sticks of memory must
be certified by the vendor as matched pairs for dual channel access.

What I meant (and which was obvious) is that if you don't want to get
burned, and don't want to have to worry about sending memory back to the
retailer for a refund if the sticks are not exactly the same, then you
"need" memory certified and sold as matched pairs.

But what Howard has discovered in this case is interesting. Based
on some of the crazy things the Asus BIOS have been doing with
other Athlon64 boards, it seems the BIOS has a pretty simplistic
(and incorrect) design. In the past, you could rely on a BIOS
to compare rows, columns, banks, and ranks as the only criterion
for a topology match. Timing should not matter, and you should be
able to mix a CAS3 module with a CAS2 module (the BIOS then selects
the slowest timing that will make all modules work - it is not that
hard to figure out). Based on Howard's observation, and a couple of
other failure cases I've read about, it almost seems the BIOS is just
comparing the entire contents of the SPD on the DIMMs, rather than
following the more traditional decision tree. I'd say some BIOS writer
got lazy...

Even a company certifying "dual channel" RAM, might have mixed
different versions of PCBs. The reason I say that, is most DIMM
manufacturers would be interested in matching rows, columns,
banks, ranks, _and_ timing, but they would never suspect that
the PCB release number has to match. If the BIOS writers and
the DIMM manufacturers don't agree on what constitutes a match,
there is still room for a customer to be inconvenienced.

Paul

 

[Mercury]

What I meant (and which was obvious) is that if you don't want to get

burned, and don't want to have to worry about sending memory back to the
retailer for a refund if the sticks are not exactly the same, then you
"need" memory certified and sold as matched pairs.

I agree - good advice although sometimes expensive to follow and as we see
here, sometimes expensive to ignore...

 

[Mark A]

I agree - good advice although sometimes expensive to follow and as we see
here, sometimes expensive to ignore...

I don't recall the difference in price, but I don't think it is significant,
if any.