Odd divisor correct and what it does.

[RyderOCZ]

I added the comment to the Dailytech article that you already implemented the detection Cpuz

http://www.dailytech.com/article.aspx?newsid=1772

[bachus_anonym]

Edit : I added ODC correction in this version :
http://www.cpuid.org/download/cpu-z-133-1.zip

Awesome Franck! Thanks for replying to my PM and fixing it sooo fast You're da man

[CapFTP]

Awesome Franck! Thanks for replying to my PM and fixing it sooo fast You're da man

Really thnx by my side too..

[Slay0r]

Very interesting ! The ODC option is mentionned in AMD's documentations, but I had no idea about its purpose. I also did not know about the minium 5x memory divider. I'll fix all of this.

Just to be sure I understood well : the ODC changes the memory multiplier as long as the memory divider is odd, is it correct ? (whatever the memory ratio is ?)

Thanks for the info

Edit : I added ODC correction in this version :
http://www.cpuid.org/download/cpu-z-133-1.zip

Speedy update there

[cpuz]

no problem

I also noticed that ODC only concerns rev E CPUs, but I'm wondering about rev F (I guess yes ?). Anyway, if it works, I put this version in replacement of 1.33.

Concerning the validation process, as soon as a new version is released, the previous versions are not able to validate.

[Deathspawner]

Kudos on getting out a new CPU-z revision out so quickly!

[lopri]

Well... It's sort of like using half multies.. in some cases where you can't exactly match your memory max and CPU max. And if my understanding is correct, just like this "Odd Divisor Correct", using half multiplier has absolutely no harm to the performance as long as you know what your mem speed is going to be. Yah you won't get the "theoretically correct" memory speed, but if that's the best speed you could get with what you have - CPU multies, HTT limit, Mem limit, Divider choices (especially with non-DFI NF4 boards) - I doubt that there is any harm in running half multi and have been saying that all along.

[SoF]

hey thx for the cpuz update - it's finally over that some stupid ones thinking they hit 330 MHZ with their rams because of the bug
cheers

[ozzimark]

it's finally over that some stupid ones thinking they hit 330 MHZ with their rams because of the bug

ouch.

also, it's a "feature" not a "bug"

[CapFTP]

ouch.

also, it's a "feature" not a "bug"

lol....yes...seems that unless one don't understand the real meaning says it's a bug....

[yasoumalaka]

I'm such a newb. I thought by running 366 htt with 3/4 my mem was at 274. I'm in total disbeliefe. My comp just newbed me.

[zungkono]

Friend, I show you what I found this morning :

Opteron 148 ~ multi x11 ( 264x11=2904MHz )
DFI nf4 expert R.A02 ~ bios E704-2BT divider180 (cpu/13)
Memory Geil 512MB single ch (dimm2/orange)

A64info shows cpu/13=223.39mhz
cpu-z 1.33.1 shows cpu/12=242mhz



is it a bug of cpu-z?

[cpuz]

The "180MHz" setting is a struggle. I can't figure out if it is actually 180MHz (ratio 10/9) or 183MHz (12/11). cpuz assumes it is 183MHz.
In this case, 11x12/11 = 12 -> CPU/12
If it is 10/9 : 11x10/9 = 12,2 -> CPU/13

Only a memory bandwidth test would tell.

[zungkono]

The "180MHz" setting is a struggle. I can't figure out if it is actually 180MHz (ratio 10/9) or 183MHz (12/11). cpuz assumes it is 183MHz.
In this case, 11x12/11 = 12 -> CPU/12
If it is 10/9 : 11x10/9 = 12,2 -> CPU/13

Only a memory bandwidth test would tell.

I set divider 180 on bios, so I think my mem run 223mhz, that save to test my cpu max speed at 1.52v

[cpuz]

I changed this recently. Check this thread.
http://forum.x86-secret.com/showthread.php?t=4505

[bachus_anonym]

Yes, it is CPU/12 and we have couple threads here that show bandwidth and SuperPi results to prove that. I did some of those, also funkflix and one more person that I remember. I will try to dig that out... Also Everest reports CPU/12.

Thanks for changing that in CPU-Z as well, Franck

[zungkono]

Yes, it is CPU/12 and we have couple threads here that show bandwidth and SuperPi results to prove that. I did some of those, also funkflix and one more person that I remember. I will try to dig that out... Also Everest reports CPU/12.

Thanks for changing that in CPU-Z as well, Franck

so the A64info give error freq ? I try clockgen, that also give cpu/13=223mhz
I will try using superpi

[filmbot]

Where exactly in the BIOS would this option be found to see if its enabled or disabled?
Genie section?

EDIT:
Also, where on CPU-Z does it show if its enabled or not?
"Correction" under the SPD tab?

[Fission]

I don't see what the big deal is all about. You guys explained what ODC does perfectly and people are complaining about MORE flexibility with RAM dividers? gfg people, i don't see why you people are even overclocking in the first place if you can't appreciate a feature like this.

I agree 100%. I am probably misusing the word "ironic" here, but it's kinda ironic reading that and looking at your sig, since you're claiming 265 htt & memCLK with a 1/2 multi. GG indeed :p

(From Sig)
A64 3500+ Venice (265x8.5 @ 1.375*113%V)
DFI NF4 Ultra-D (704-2BT)
2x1GB Team Xtreem Cronus Micron PC4000 (265 @ 2.8V)

[eva2000]

Bachus

All the ATI/Sapphire boards have ODC, most of the ATI chipset DFI boards have ODC, Abit have just added it to a SLI32 board also so i can't see what your issue is with 406 being a bad bios.

ODC does have its uses as i just helped all the RS482 infinity users get the 150 4/3 ratio where the 3/2 or 133 ratio is not working, they just need to set 6/5 and ODC enabled and they have the magic 150 everyone on a low multi cpu loves.

Its just another tweak that when used correctly will help end users to get the speed we need, regarding software all that needs to happen is ODC detection needs to be built in that compensates for the option.

Im sure the guys at X-86 etc can work this into CPU-Z etc real easy.

Guess this is related, I swapped out my 146 CAB2E 0540FPMW cpu for my 148 CABYE 0543FPMW for the higher 11x cpu multiplier and came across a weird thing not sure if it's related to the Odd Divisor Correct (i don't even have it enabled) outlined in this thread

Further info on Odd Divisor Correct:

This function always rounds your Ram divider to an even number, so if you are running 1:1 with an Odd CPU multi, your ram will run slower than the FSB (very similar to the 0.5 multiplier's)

Example: 250 FSB x 11 CPU multi at 1:1 = 2750 CPU speed / 11 = 250 Ram speed (DDR500)
If you enabled Odd Divisor Correct, your Ram divider would be 12 which would put your ram at 229.167 in this example.


Example #2: 330 x 9 at 5:6 (166 memclock) = 2970 CPU speed / 11 = 270 Ram speed.
Enable Odd Divisor Correct and your divider will go to 12 and the ram speed would be 247.5 MHz.

Software - CPU-z, Memtest, Everest, etc will not identify that your ram is running slower. The only way that I have been able to demonstrate it, is with memtest and comparing the measured bandwidth (I will try to confirm in Sandra as well and post some screen shots shortly.)

Now the exception (sort of) to this rule is when you have an odd CPU multi and your divider already comes out to an even number, Odd Divisor Correct will have no effect because there is nothing to "correct".

This can be very beneficial when you are trying to push the CPU that little bit farther, but your ram is holding you back.
We have also noticed that when you are running the "Upl****" ram ratio's with the "E" Die CPU's it will give you "something in between" the given ratio's

When i set speeds to 11x 274HTT with 180 divider A64 Mem divider helper says my ram should be at 231mhz. But in memtest86+ v1.65 bandwidth reports 2943MB/s which i thought was way higher than usual.

So i boot into windows and see that cpuz reports my ram @251mhz 2-2-2-5 which would be consistent with 2943MB/s memtest86+ bandwidth reported.

Click on image for full screenie which includes A64info reporting ram should be 231mhz as well.


I see i do have an even number memory divisor of 12 as pointed out in RyderOCZ's quoted post.

Exception to the rule:
So this is the exception to the rule right ? when the 11x cpu multiplier when using 180 divider the memory is probably indeed running with 12 divisor meaning what cpuz v1.33.1 or higher is reporting for mem speed is indeed the correct mem speed ?

http://www.xtremesystems.org/forums/...8&postcount=87 and http://www.xtremesystems.org/forums/...6&postcount=88 and http://forum.x86-secret.com/showthread.php?t=4505

[bachus_anonym]

It's not exception actually. It's something I've been trying to convince Franck (CPU-Z creator) about since summer last year. I'm glad CPU-Z 1.33 finally shows that now... It has nothing to do with Odd Divisor Correct.
At multi x11 and div180, memory speed is CPU/12 and NOT CPU/13 as it would look like from Oscar Wu's table or some monitoring software. Everest has been showing CPU/12 for a long time now. A64Info does not and I think Clockgen has not been updated either.

[eva2000]

Ah i see... doing 32M run and yes 11x274htt 180 with 251mhz 2-2-2-5 is on track to have similar ~24min flat time as with 10x300HTT 166 with 250mhz 2-2-2-5 ram heh

[lukija]

At multi x11 and div180, memory speed is CPU/12 and NOT CPU/13 as it would look like from Oscar Wu's table or some monitoring software. Everest has been showing CPU/12 for a long time now. A64Info does not and I think Clockgen has not been updated either.

A64Info math is based on Oscar's table. I'm assuming that 180 is actualy 183.00MHz. Same goes for 140 -> 143.00MHz. Only 11 multi is problematic with 180 divider since it is on the edge (12 if you assume 183.333... or 12.02, if you assume 183.00, thus making it 13). I have proposed test to solve this dilemma but no one responded:

http://www.xtremesystems.org/forums/...8&postcount=57

This post also convinced me that divider is 13:

http://xtremesystems.org/forums/show...6&postcount=40

My firiend managed to measure bandwidths, but it is only one case:

http://www.xtremesystems.org/forums/...9&postcount=59

It is importain to maintain the same memory timings in both tests. Unfortunately, I do not have CPU with 11 multi to test it.

[Repoman]

Hmm, little ot but since CPU-Z is here, figured I'll bring this up-



140MHz ram divider isn't detected correctly. I'm not sure if this is new in 1.33, I don't remember ever having the problem before

[eva2000]

A64Info math is based on Oscar's table. I'm assuming that 180 is actualy 183.00MHz. Same goes for 140 -> 143.00MHz. Only 11 multi is problematic with 180 divider since it is on the edge (12 if you assume 183.333... or 12.02, if you assume 183.00, thus making it 13). I have proposed test to solve this dilemma but no one responded:

http://www.xtremesystems.org/forums/...8&postcount=57

This post also convinced me that divider is 13:

http://xtremesystems.org/forums/show...6&postcount=40

My firiend managed to measure bandwidths, but it is only one case:

http://www.xtremesystems.org/forums/...9&postcount=59

It is importain to maintain the same memory timings in both tests. Unfortunately, I do not have CPU with 11 multi to test it.

Ran 32M and definitely the ram is running 251mhz at 11x274HTT with 180 divider the time of 23min 54s is spot on for the ram it seems.



Settings used

FSB Bus Frequency: 274
LDT/FSB Frequency Ratio: 4x
CPU/FSB Frequency Ratio: 11x
RD580 HT PLL Speed: High
HT Bus NMOS Drive Strength: 5
HT Bus PMOS Drive Strength: 5
HT Bus Receiver Impendence: 5
CPU HT Bus Drive Strength: Strong
K8 Cool & Quiet Support: Disabled

CPU VID Control: 1.55v
CPU VID Special Control: AUTO
DRAM Voltage Control: 3.4v
SB PCIE Voltage: 1.8v
NB Analogue Voltage: 1.34v
LDT Bus Voltage: 1.34v
NB Core Voltage: 1.34v

DRAM Frequency Set: 180
CPC: Enabled
CAS (tCL): 2
tRCP: 2
tRAS: 5
tRP: 2
tRC: 7
tRFC: 14
tRRD: 2
tRW: 2
tRTW: 1
tWTR: 2
tREF: 4708
Bank Interleaved: Enabled
Errata 94: Disabled
Errata 123: AUTO
Odd Divisor Correct: Disabled

DQS Skew Control: AUTO
DQS Skew Value: N/A
DRAM Drive Strength: 3 (weak)
DRAM Data Drive Strength: 2 (-33%)
DIMM 1/2 Clock Timing Skew: AUTO
DIMM 3/4 Clock Timing Skew: 150ps delay
Max Async Latency: 7ns (tried 6ns but there was a long delay before booting system)
Read Preamble Time: 5ns
IdleCycle Limit: 16clks
Dynamic Counter: Enabled
R/W Queue Bypass: 16x
Bypass Max: 7x
Burst Length: 4

More tests to confirm ram is at 251mhz 2-2-2-5

Everest Home
Read = 7,191 MB/s
Write = 3,300 MB/s
Latency = 36.1 ns

Sandra 2005 Lite SR3
Buffered = 7,508 / 7,445
Unbuffered = 3,574 / 3,777


vs

10x300HTT = 3000mhz
@250mhz 2-2-2-0 7-12-0211 at 3.44v

Memtest86+ v1.65 bandwidth = 3006MB/s

FSB Bus Frequency: 300
LDT/FSB Frequency Ratio: 4x (tried 5x but system didn't boot even with NMOS/PMOS drive strength @31 so dropped it to 4x)
CPU/FSB Frequency Ratio: 10x
RD580 HT PLL Speed: High
HT Bus NMOS Drive Strength: 5
HT Bus PMOS Drive Strength: 5
HT Bus Receiver Impendence: 5
CPU HT Bus Drive Strength: Strong
K8 Cool & Quiet Support: Disabled

CPU VID Control: 1.5v (set higher due to low vcore + high vdimm issue)
CPU VID Special Control: AUTO
DRAM Voltage Control: 3.44v
SB PCIE Voltage: 1.8v
NB Analogue Voltage: 1.34v
LDT Bus Voltage: 1.34v
NB Core Voltage: 1.34v

DRAM Frequency Set: 166
CPC: Enabled
CAS (tCL): 2
tRCP: 2
tRAS: 0
tRP: 2
tRC: 7
tRFC: 12
tRRD: 0
tRW: 2
tRTW: 1
tWTR: 1
tREF: 4708
Bank Interleaved: Enabled
Errata 94: Disabled
Errata 123: AUTO
Odd Divisor Correct: Disabled

DQS Skew Control: AUTO
DQS Skew Value: N/A
DRAM Drive Strength: 3 (weak)
DRAM Data Drive Strength: 2 (-33%)
DIMM 1/2 Clock Timing Skew: AUTO
DIMM 3/4 Clock Timing Skew: 150ps delay
Max Async Latency: 8ns
Read Preamble Time: 4.5ns
IdleCycle Limit: 0clks
Dynamic Counter: Enabled
R/W Queue Bypass: 16x
Bypass Max: 7x
Burst Length: 8

Super Pi v1.5 32M


The above 23min 59.297s time managed to beat my DFI NF4 Ultra-D time of 23min 59.641s.



Everest Home:
Read = 7,142 MB/s
Write = 3,361 MB/s
Latency = 36.2 ns

Sandra 2005 Lite SR3
Buffered = 7,486 / 7,419
Unbuffered = 3,717 / 3,847