P4B FSB:DRAM Ratio - Does the BIOS lie or CPU-Z ?

[arifi]

Hello Everyone,

I installed a Celeron 2.0 MHz (Northwood) into my P4B and trying to get
the most out of it.

Since this CPU is multiplier-locked (or?), I said I would increase the
FSB as much as i can, and keep the Memory frequency at 1:1 to FSB. I do
not plan to go over FSB=133, hence PC-133 memory would be operating
below or at but bot above its limits.

I went ahead and set the FSB-RAM ratio to 1:1 in the BIOS, and started
playing with FSB.

The system runs fine at 20x120 (did not try higher), but when I check
the memory status with CPU-Z, it reports that it is running at approx.
160 MHz with FSBRAM ratio 3:4. Oops ! Now, what is going on? Is the
BIOS setting not effective? Is CPU-Z wrong ? What am I missing here ?

I tried three different BIOS revisions: 1012, 1013-003, 1013-004 all
with similar result.

Any pointers DEEPLY appreciated
TIA
-arifi

 

[Arifi Koseoglu]

Following up to my original posting:

Hello Everyone,

I installed a Celeron 2.0 MHz (Northwood) into my P4B and trying to get
the most out of it.

Since this CPU is multiplier-locked (or?), I said I would increase the
FSB as much as i can, and keep the Memory frequency at 1:1 to FSB. I do
not plan to go over FSB=133, hence PC-133 memory would be operating
below or at but bot above its limits.

I went ahead and set the FSB-RAM ratio to 1:1 in the BIOS, and started
playing with FSB.

The system runs fine at 20x120 (did not try higher), but when I check
the memory status with CPU-Z, it reports that it is running at approx.
160 MHz with FSBRAM ratio 3:4. Oops ! Now, what is going on? Is the
BIOS setting not effective? Is CPU-Z wrong ? What am I missing here ?

I tried three different BIOS revisions: 1012, 1013-003, 1013-004 all
with similar result.

Even though CPU-Z always (no matter what the setting in BIOS is) reports
the ram running at 160 MHz with 3:4 ratio, WCPUID reports the ram running
at 120 MHz (matches the setting in BIOS). Now I am even more confused...
Which one is correct ??

Hasn't any P4B owner tried / experienced such a thing?

-arifi

 

[Paul]

Following up to my original posting:



Even though CPU-Z always (no matter what the setting in BIOS is) reports
the ram running at 160 MHz with 3:4 ratio, WCPUID reports the ram running
at 120 MHz (matches the setting in BIOS). Now I am even more confused...
Which one is correct ??

Hasn't any P4B owner tried / experienced such a thing?

-arifi

The CPU/AGP/PCI clocks are controlled by the clock generator
chip. As far as I know, the AGP clock and the Hub clock are
the same frequency. The Northbridge multiplies the Hub clock
up, to make signals suitable for driving the memory. In
the Northbridge, there is a register called MCHCFG and
bit 11 selects 0=100MHz and 1=133MHz. See page 74 (the 100MHz
setting is classed as "reserved", but the value is explained
in the 845E datasheet, and I am proceeding by analogy here):

ftp://download.intel.com/design/chipsets/datashts/29072502.pdf

To figure out what speed the memory is running at, you need
to know the clock settings (i.e. 120/80/40) and the value
of MCHCFG bit 11. Some samples:

120/80/40 MCHCFG bit 11 = 1 80*2.0 = 160MHz memory
120/80/40 MCHCFG bit 11 = 0 80*1.5 = 120MHz memory
120/60/30 MCHCFG bit 11 = 1 60*2.0 = 120MHz memory
120/60/30 MCHCFG bit 11 = 0 60*1.5 = 90MHZ memory

(The ICS clock generator chip is 950209 on that board, but the
datasheet is not available. This one is similar:
http://www.icst.com/datasheets/ics950208.pdf)

Now, I don't know if H-Oda's WPCREDIT gives access to the
value of MCHCFG or not. You might investigate whether that
tool can observe the bit in question.

One way to verify the memory clock, would be to try to
run three sticks of PC133 memory in the computer. Three
double sided sticks should be able to run at up to PC133
speeds, but I doubt they would work at 160MHz. If you have
three sticks of memory in the computer, and the memory is
error free, then chances are the frequency really is 120MHz.

I've never tried overclocking my P4B with 1.8GHz/FSB400
processor, as it was my work machine at the time.

Paul

 

[arifi]

Dear Paul,

I thank you very very much for the detailed and informative response.

Please accept my apologies for the following - probably too dumb wrt
your knowledge level - questions:

My memory configuration is: 1x512 (chips on both sides)+ 2x256 PC133
SDRAM (chips on one side) - would the practical test you mention be
still valid? (and, if I may ask at this point, why do double-sided and
single-sided sticks behave differently?)

Also, would installing whatever comes with the Intel(R) Chipset
Software Installation Utility downloadable for this MoBo from Asus'
support site make any difference in this issue?

Best,
-arifi

 

[Paul]

Dear Paul,

I thank you very very much for the detailed and informative response.

OK, I tried a few experiments. Memtest86 version 1.4

1.8/CPU=100/MEM=100 WPCREDIT C6-C7 = 0500 (byte swap) v
CPUZ claims memory is 133. WPCREDIT shows bit 11 is zero 0000000000000101
By SPD, 3-2-2-5, Memtest 429MB/sec

1.8/CPU=100/MEM=133 WPCREDIT C6-C7 = 0508 (byte swap) v
CPUZ claims memory is 133. WPCREDIT shows bit 11 is one 0000100000000101
By SPD, 3-3-3-6, Memtest 545MB/sec

1.98/CPU=110/MEM=110 WPCREDIT C6-C7 = 0500 (byte swap) v
CPUZ claims memory is 147 . WPCREDIT shows bit 11 is zero 0000000000000101
By SPD, 3-2-2-5, Memtest 469MB/sec

1.98/CPU=110/MEM=146 WPCREDIT C6-C7 = 0508 (byte swap) v
CPUZ claims memory is 147. WPCREDIT shows bit 11 is one 0000100000000101
By SPD, 3-3-3-6, Memtest 596MB/sec

The conclusion is, CPUZ is not reading C6-C7 properly from the
host interface in the Northbridge. Or, it is reading the register,
but refusing to believe that zero causes 100MHz operation. The
Intel datasheet calls the value reserved, but as the Asus BIOS
allows two values (1:1 or 3:4), the value of zero for the bit
must correspond to the 1:1 condition. WCPUID works correctly.

In the above condensed results, the first three fields are
what I would expect based on the BIOS settings. The WPCREDIT
value is copied off the screen from bytes C6-C7. The two bytes
must be reversed, before writing the bit pattern on the next line.
CPUZ seems to assume the divider is 3:4 all the time, and the
calculated memory clock reflects that result. The "By SPD"
is how I set my memory, and the next four fields are
CAS,tRCD,tRP,ActPrecharge, as listed in the BIOS order.
The memtest bandwidth number, is the value returned for main
memory bandwidth, from the indicator in the upper left corner
of the memtest86 version 1.4 screen (memtest.org). The memtest
results are consistent with the way the Asus BIOS claims to
work. Only CPUZ is wrong.

Another thing that surprised me - my three double sided PC133
memory sticks ran just fine at 146MHz (I haven't tested
them with Prime95 or SuperPI yet.)

As far as possible, the experiment proves that the Asus BIOS is
doing the correct thing, and when set to 1:1, the Northbridge
bit is set to zero, as it should be for that ratio.

Please accept my apologies for the following - probably too dumb wrt
your knowledge level - questions:

My memory configuration is: 1x512 (chips on both sides)+ 2x256 PC133
SDRAM (chips on one side) - would the practical test you mention be
still valid? (and, if I may ask at this point, why do double-sided and
single-sided sticks behave differently?)

My best guess is that configuration won't work at 160MHz. I would
expect to see errors with memtest86. If the bus is really working
well for you, the frequency must be a bit less than 160MHz.

Here is some sample data:
http://abxzone.com/forums/showthread.php?t=38293

The bus consists of data, address, and control signals. The address
and control signals drive the chips in parallel, and this is the
hardest thing to do at speed. A double sided and two single sided
DIMMs is 16+8+8=32 loads on the address bus. And that tends to be
a limiting factor for bus speed. The data signals only have two
loads on a double sided module, and one load on a single sided
module, which means on your bus, a total of four data loads on
the bus. So the address is the part that is hard to drive.

Some Northbridges are better than others at driving a heavily
loaded bus. Mushkin has a 1GB module qualification page, which
makes note of a few motherboards that are good at driving "stacked"
DIMMs - these are DIMMs with double the load of a normal DIMM.
There are also merchants listing stacked RAM modules on Pricewatch,
and they actually list some motherboards known to handle a heavy
load better than others. So, there are some differences in how
well certain chipset vendors chips work.

As I stated in my original post, these are only rough rules of
thumb, and each situation is slightly different. Occasionally,
you'll run into someone here, who has achieved more than should
be possible. I only mention the rules, to prevent people from
buying too many sticks of RAM, and then being pissed off that
their last stick cannot be inserted into the motherboard.

Also, would installing whatever comes with the Intel(R) Chipset
Software Installation Utility downloadable for this MoBo from Asus'
support site make any difference in this issue?

Best,
-arifi

I don't think so. The chipset drivers are an enabler, without
much of a life of their own. The BIOS is doing the "hard work"
here.

HTH,
Paul

 

[arifi]

Dear Paul,

Your responses were probably the most helpful and informative ones I
received for years. I thank you again for all the time and effort you
spent for investigating this situation and explaining me how things
are, and how I can interpret disagnostic output.
Mucho Thanks !
Wishing you the best in life,
arifi