CPU-Z lies with memory speed ?

Printable View

12-09-2004, 12:29 AM
XYZPawel

CPU-Z lies with memory speed ?

Hi All (i'ts my first post here :)

I have problem with memory speeds shows by CPU-Z and when memory dividers are in use:

FSB 242 x 9 = CPU 2178, divider 166.

BIOS, memtest86, aida32 - shows - 201 MHz (memtest 201 (DDR403)

CPU-Z, GemClock - 198 Mhz

I think that CPU-Z shows wrong speed :(

With math: divider 166 = 5/6 FSB = 201,7 (DDR403)

What do you think ?

Edit
===========================================
It looks like CPU-Z tells truth :)
evidence is in AMD doc 31411 pdf 939 func data, there is a table - page 15. Memory clock isn't divide 5/6 FSB.
it's a shame that BIOS, memtest86, aida32 can't show real memory clock :mad:
I sent email to memtest bugs report...
12-09-2004, 12:50 AM
STEvil

198 according to A64 MemFreq 1.1
12-09-2004, 02:08 AM
Formann

A64 does not use 6:5, 3:2, 5:4 dividers. The mem divider depends on which multi you are running since it divides the cpu speed and not the bus speed.

Example:

242*9 = 2178Mhz

With mem @ 166Mhz i bios, and running 9x multi you get cpu/11 as your actual mem divider.

2178/11 = 198Mhz

Most programs use the bus speed to calculate thus making the numbers wrong. Like 242/6*5 = 201.6Mhz <-- wrong!
12-09-2004, 02:52 AM
XYZPawel

Formann ->

BIOS, memtest86, Aida32 are wrong ?!?

How to check it for sure ?
12-09-2004, 03:01 AM
OC Detective

Formann is 100% correct that is exactly how the divider is worked out. Example multiplier of 11 running with HTT:RAM at 6:5 then it is 11 x 6/5 = 13.2 however the memory divider is ALWAYS the next HIGHEST integer i.e 14. Then its a case of dividing the HTT frequency by 14 to get memory frequency.
12-09-2004, 04:46 AM
arctic-k20

i think cpu-z rounds the divider number up or down to the closest whole number - so whne you use the 166 divider on a 9 multiplier it says 11 but really the divider is 10.8

divide your numbers using 10.8 and you well get your actual ram speed !!!
12-09-2004, 08:42 AM
arctic-k20

also when using the 8 multi and the 166 divider the true ram divider is 9.6 - not 10 !!
12-09-2004, 10:03 AM
Hvidgaard

I think that CPU-Z is right... It also makes sence since the ,5 multis lower ram speed at 1:1 is explained...

Mem=FSB*Multiplier/roundup(Multi*ratio)

Multipliers lower than 5 is not supported, and will be treated as 5 when calculating memfrequence (Multi equals 5 if multiplier is 3-4,5).

Mem=250*9,5/roundup(9,5*1/1)=237,5Mhz

I have done some math...

ex.

FSB 200, mem 166(6:5)

Multi cpu mem

15 3000 166,67 =200*15 /roundup(15 *6/5)=3000/18
14,5 2900 161,11 =200*14,5/roundup(14,5*6/5)=2900/18
14 2800 164,71 =200*14 /roundup(14 *6/5)=2800/17
13,5 2700 158,82 =200*13,5/roundup(13,5*6/5)=2700/17
13 2600 162,5 =200*13 /roundup(13 *6/5)=2600/16
12,5 2500 166,67 =200*12,5/roundup(12,5*6/5)=2500/15
11 2400 160 =200*12 /roundup(12 *6/5)=2400/15
11,5 2300 164,29 =200*11,5/roundup(11,5*6/5)=2300/14
11 2200 157,14 =200*11 /roundup(11 *6/5)=2200/14
10,5 2100 161,54 =200*10,5/roundup(10,5*6/5)=2100/13
10 2000 166,67 =200*10 /roundup(10 *6/5)=2000/12
9,5 1900 158,33 =200*9,5 /roundup(9,5 *6/5)=1900/12
9 1800 163,63 =200*9 /roundup(9 *6/5)=1800/11
8,5 1700 154,55 =200*8,5 /roundup(8,5 *6/5)=1700/11
8 1600 160 =200*8 /roundup(8 *6/5)=1600/10
7,5 1500 166,67 =200*7,5 /roundup(7,5 *6/5)=1500/9

I think the pattern will be different if another divider is used.

Can someone confirm this?

EDIT: Great, my nice notepad layout got screwed up :mad:
12-09-2004, 10:45 AM
Pimp!

i think if you use

Code:

text here

tags then it wont mess up your layout,

EDIT: forgot it would do the code bit :p

use the [ code ] [ / code ] tags
12-09-2004, 01:09 PM
Hvidgaard

Code:

FSB 200, mem 166(6:5) Multi cpu mem 15 3000 166,67 =200*15 /roundup(15 *6/5)=3000/18 14,5 2900 161,11 =200*14,5/roundup(14,5*6/5)=2900/18 14 2800 164,71 =200*14 /roundup(14 *6/5)=2800/17 13,5 2700 158,82 =200*13,5/roundup(13,5*6/5)=2700/17 13 2600 162,5 =200*13 /roundup(13 *6/5)=2600/16 12,5 2500 166,67 =200*12,5/roundup(12,5*6/5)=2500/15 11 2400 160 =200*12 /roundup(12 *6/5)=2400/15 11,5 2300 164,29 =200*11,5/roundup(11,5*6/5)=2300/14 11 2200 157,14 =200*11 /roundup(11 *6/5)=2200/14 10,5 2100 161,54 =200*10,5/roundup(10,5*6/5)=2100/13 10 2000 166,67 =200*10 /roundup(10 *6/5)=2000/12 9,5 1900 158,33 =200*9,5 /roundup(9,5 *6/5)=1900/12 9 1800 163,63 =200*9 /roundup(9 *6/5)=1800/11 8,5 1700 154,55 =200*8,5 /roundup(8,5 *6/5)=1700/11 8 1600 160 =200*8 /roundup(8 *6/5)=1600/10 7,5 1500 166,67 =200*7,5 /roundup(7,5 *6/5)=1500/9 I think the pattern will be different if another divider is used.
12-09-2004, 01:16 PM
bias_hjorth

http://xtremesystems.org/forums/showthread.php?t=41595 <-- complete memory table there.
12-09-2004, 03:10 PM
Hvidgaard

#11 I've seen it, but I just wanna give people the formula, because we're not exactly useing 200MHz fsb ;)
12-09-2004, 08:55 PM
OC Detective

Quote:

Originally Posted by arctic-k20

also when using the 8 multi and the 166 divider the true ram divider is 9.6 - not 10 !!

All dividers must be whole numbers not a fraction of a number.
12-09-2004, 09:01 PM
Zanr Zij

sad 2 say that truth... On mine testing V-DATA 128MB DDR333@DDr460 no change timing but CPUZ report DDR553 :rolleyes: :D :D :D

http://img.photobucket.com/albums/v5...DATA333553.jpg
12-10-2004, 01:44 AM
bias_hjorth

Quote:

Originally Posted by Hvidgaard

#11 I've seen it, but I just wanna give people the formula, because we're not exactly useing 200MHz fsb ;)

Formula is there too
12-10-2004, 02:15 AM
Formann

If you take a look here its easy to calculate the actuall memory speed

Notice the mem DIV? Its ALWAYS an Integer.

(HTT*Multi)/DIV = mem speed

Code:

"166MHz" "5/6" Multi DIV. 12 15 11,5 15 11 14 10,5 13 10 12 9,5 12 9 11 8,5 11 8 10 7,5 10 7 9 6,5 9 6 8 5,5 8 5 6 "150MHz" "3/4" Multi DIV. 12 16 11,5 16 11 15 10,5 15 10 14 9,5 13 9 12 8,5 12 8 11 7,5 10 7 10 6,5 9 6 8 5,5 8 5 7 "133MHz" "2/3" Multi DIV. 12 18 11,5 18 11 17 10,5 16 10 15 9,5 14 9 14 8,5 13 8 12 7,5 11 7 11 6,5 10 6 9 5,5 9 5 8

XYZPawel: Yes. CPU-Z, AIDA etc are wrong. They use HTT to calculate mem speed. The A64 uses CPU speed, and so shall we ;)
12-10-2004, 05:12 AM
arctic-k20

all i am saying is the bios reports memory speeds that can only be obtained by using fraction's of numbers - the difference isnt huge but it can mean the difference of 10~15mhz on the ram - if you are at the very edge of stability on the ram that can make or break your overclock...

my bios says my ram is 250mhz - cpu-z says my ram is 240 - sandra memory bandwidth shows me that my ram is 250 with the scores i obtain....
12-10-2004, 05:47 AM
Formann

Quote:

Originally Posted by arctic-k20

all i am saying is the bios reports memory speeds that can only be obtained by using fraction's of numbers - the difference isnt huge but it can mean the difference of 10~15mhz on the ram - if you are at the very edge of stability on the ram that can make or break your overclock...

my bios says my ram is 250mhz - cpu-z says my ram is 240 - sandra memory bandwidth shows me that my ram is 250 with the scores i obtain....

What are your settings? HTT*multi ram setting in bios?

I say your bios is wrong. The 6:5, 5:4 etc dividers that the BIOS uses are not Intergers, and therefor cant be correct.
12-10-2004, 06:08 AM
arctic-k20

and memtest is wrong too then?

i am using 8x300 on 166 divider

cpuz says 240........memtest/bios say 250..........and my sandra scores would reflect ddr500 - not ddr 480.......i dunno
12-10-2004, 06:11 AM
OC Detective

Quote:

Originally Posted by Formann

What are your settings? HTT*multi ram setting in bios?

I say your bios is wrong. The 6:5, 5:4 etc dividers that the BIOS uses are not Intergers, and therefor cant be correct.

Yes maybe you have an early edition of the bios either that or your mobo manufacturer doesnt know how the memory divider works! Either way its wrong as the memory divider has to be a whole number - this piece of information has been known for a while.....
For anyone that thinks otherwise open this link and click on the AMD Functional Data Sheet, 754 Pin Package go to Page 17 Table 1 and do the maths yourself and you will see that the memory divider is a whole number not a fraction.
http://www.amd.com/us-en/Processors/...9_7203,00.html
Oh and in your case just look at what it says when the cpu is at 1.6Ghz (8 x 200) and DRAM frequency is set to 166 it clearly says the actual interface speed is 160 so in your case of 8 x 300 it will be 160 * (300/200) which is clearly 240 not 250!
12-10-2004, 06:34 AM
arctic-k20

msi neo2
bios 136 (doesnt matter which bios - they all report the same)
0440a2 taiwan
12-10-2004, 06:37 AM
OC Detective

Well arctic-k20 it is wrong - simple as that. Unless you think MSI is right and AMD is wrong?......
12-10-2004, 09:37 AM
Formann

Quote:

Originally Posted by OC Detective

Well arctic-k20 it is wrong - simple as that. Unless you think MSI is right and AMD is wrong?......

Now why should AMD be right about the mem controller in the A64? :stick: :D
12-10-2004, 11:30 AM
arctic-k20

you guys are funny - i wasnt saying one was right or wrong - i was actually asking but either way ....
12-10-2004, 04:35 PM
OC Detective

Quote:

Originally Posted by arctic-k20

you guys are funny - i wasnt saying one was right or wrong - i was actually asking but either way ....

Quote:

Originally Posted by arctic-k20

also when using the 8 multi and the 166 divider the true ram divider is 9.6 - not 10 !!

Seems to me you were claiming it was a fraction rather than a whole number, rather than asking either way.