AMD64 - RAMHz vs VCore - Please Share

[Rabbi_NZ]

After messing round with my system for the past week I am starting to see a pattern... it seems like, to a certain extent at least, we still need to treat the memory controller and "cpu core" as two different entities.

I say this because after seeing both my results and those of others it is becoming more and more evident that once a certain 1:1 HTT limit is reached it is often the VCore that needs to be raised and not the VDimm (EG, MemController gives up before RAM).... even if the CPU is at a lowly 1500MHz.

To me, that points to the memory controller requiring more juice to handle that memory speed... even though the "core" and cache are still skipping along merrily without a care at that voltage.

So... I want some results from everybody to see if maybe we can STILL treat FSB chases as a combination between vChipset & vDimm... all be it, now we are forced to have the same value for both seeing as they're on the same die.

I will dedicate the next post for all the results... PLEASE POST YOUR EXPERIENCES.

Cheers

[Rabbi_NZ]

RESERVED FOR RESULTS

Rabbi: 1.4v Mobile 3400+ 0437 CG CH

10x256 (2560MHz) , 5/6, 213 RAMHz @ 3-2-2-10: VCore required 1.55v - Buffalo
10x260 (2600MHz), 5/6, 217 RAMHz @ 3-2-2-10: VCore required 1.61v - Buffalo
11x225 (2475MHz), 1/1, 225 RAMHz @ 3-2-2-10: VCore required 1.55v (maybe less) - Buffalo
10x256 (2560MHz), 1/1, 256 RAMHz @ 2-2-2-10: VCore required 1.65v - OCZ VX
10x260 (2600MHz), 1/1, 260 RAMHz @ 2-2-2-10: VCore required 1.75v - OCZ VX

1st stick 7x250 3.3v 2-2-2-10 1T MemTest #5 10 loops no errors
2nd stick 7x250 3.3v 2-2-2-10 1T MemTest #5 10 loops no errors
2x512 7x240 3.2v 2-2-2-10 1T MemTest #5 40 loops no errors
2x512 7x245 3.2v 2-2-2-10 1T MemTest #5 20 loops no errors
2x512 7x250 3.2v 2-2-2-10 1T MemTest #5 20 loops no errors
2x512 7x250 3.2vdimm 1.55vcore 2-2-2-10 1T Prime Blend 1 hour & SuperPI 32MB no errors
2x512 10x250 3.2vdimm 1.61vcore 2-2-2-10 1T Prime Blend 8 hour no errors
2x512 9x260 3.2vdimm 1.55vcore 2-2-2-10 1T errors in MemTest #5 straight away
2x512 9x259 3.2vdimm 1.55vcore 2-2-2-10 1T errors in MemTest #5 straight away
2x512 9x260 3.3vdimm 1.55vcore 2-2-2-10 1T MemTest #5 3 loops no errors
2x512 9x258 3.2vdimm 1.55vcore 2-2-2-10 1T MemTest #5 6000-7000 errors in 400-500 loops
2x512 10x256 3.2vdimm 1.61vcore 2-2-2-10 1T SuperPI 32MB pass, Prime-Blend fail in 30mins
2x512 10x256 3.2vdimm 1.65vcore 2-2-2-10 1T Prime-Blend 11hours no errors
2x512 7x257 3.2vdimm 1.55vcore 2-2-2-10 1T MemTest #5 8 errors in 10 hrs
2x512 10x260 3.3vdimm 1.65vcore 2-2-2-10 1T Prime errors within 30min / SuperPI 32MB fails
2x512 10x260 3.4vdimm 1.65vcore 2-2-2-10 1T Prime errors within 30min / SuperPI 32MB fails
2x512 10x260 3.4vdimm 1.71vcore 2-2-2-10 1T Prime Large FFT no errors in 1hour / SuperPI 32MB fails
2x512 10x260 3.4vdimm 1.75vcore 2-2-2-10 1T SuperPI 32MB Passed
2x512 10x260 3.3vdimm 1.75vcore 2-2-2-10 1T SuperPI 32MB Passed / Prime 10hrs+ / Sandra Mem / MemTest

[cuddles]

I'll try it out next time I'm home. Seems logical, looks promising.

[Orker]

Nice post RabbiNZ :thumbsup:

I can run up to 260 2-2-2-10 3.6v... This VX truly does get better overtime

But ofcourse my cpu cant keep up with this, even with a max vcore of 1.7 which i deem to be unsafe anyway with xp90 cooling

The cpu doesnt like going much pass 250 in sync with ram

so 2500mhz or less for 250+mhz

[Rabbi_NZ]

Thanks for the post Orker :), but could you set out your results like I did please?

Eg,
2x512 10x250 3.2vdimm 1.61vcore 2-2-2-10 1T Prime Blend 8 hour no errors
2x512 10x260 3.3vdimm 1.65vcore 2-2-2-10 1T Prime errors within 30min / SuperPI 32MB fails

That way we can see what settings TRULY are stable and what settings are not... OR, "HOW" stable a certain setting is and what solved it... VCore, Vdd, VDimm, Timings, etc... The more "DETAILED" results we get the better our chance of tracking this down.

[blinky]

there must be a seperate voltage for the mem controller than the cpu, it must be like vht or something.

someone wanna wade through amd white papers?

[STEvil]

Whats the other voltage the DFI board lets you adjust from about 0.8v to 1.55v? Called VID or FID.. the one thats not CPU voltage anyways heh.

[Rabbi_NZ]

CPU voltage on the DFI is a combination of VID & those % options... VCore = VID x %... eg, 1.5v (VID) x 110% = 1.65v (VCore)

[STEvil]

ah.. I always just used the %'s.. enough confusing options on the board already ;)

[ugp]

The board offers alot of adjustablilty to users and is very nice. Although sometimes still hard to tune and dial it in. I have had it since it has been released and I am still trying to dial mine in just right.

[Rabbi_NZ]

wots up eh... i figured this would be the one and only forum where people cared enough to go though the effort of collecting this info.

if we find a "rule" of thumb related to this it should apply to future AMD cpus with on-die controllers too

[STEvil]

Well, its not likely you will find a rule of thumb since temperature and revisions as well as place in the wafer likely all affect this..

Its something everyone will have to find for themselves.

[Rabbi_NZ]

nForce2 varied in revision, Manufacturer of board, RAM used, date of chip, cooling used... but IN GENERAL we saw that more Vdd helped FSB climb when the NB was the limiting factor.

Nothing ventured nothing gained bro

[highspeed]

Nice post Rabbi.

10x260 (2600MHz), 5/6, 217 RAMHz @ 3-2-2-10: VCore required 1.61v
10x260 (2600MHz), 1/1, 260 RAMHz @ 2-2-2-10: VCore required 1.75v

So the CPU must have more Vcore when going 1:1, why´s that?

[computersmsa]

Vcore = 1,40 Volt
http://www.bel-hardware.be/a64moc/CPU/_s.png (http://www.bel-hardware.be/a64moc/CPU/1,4V.png)

Vcore = 1,55 Volt
http://www.bel-hardware.be/a64moc/CPU/_s.png (http://www.bel-hardware.be/a64moc/CPU/1,55V.png)

Vcore = 1,65 Volt
http://www.bel-hardware.be/a64moc/CPU/_s.png (http://www.bel-hardware.be/a64moc/CPU/1,65V.png)

Vcore = 1,75 Volt
http://www.bel-hardware.be/a64moc/CPU/_s.png (http://www.bel-hardware.be/a64moc/CPU/1,75V.png)

2.7 Ghz Syncro 300 Mhz HTT 1.68V
http://www.bel-hardware.be/a64moc/CPU/_s.png (http://www.bel-hardware.be/a64moc/CPU/2700.png)

I thinks that I win anythings in desyncro, but I will test it in near time :)


EDIT : It's with one stick :stick: sorry :toast: :rolleyes:

[Rabbi_NZ]

Thank you computersmsa.

Are those the max speeds you could do SuperPI 4MB at?

Jus want to know how you found the max for each voltage setting... notice how in mine I said Prime, SuperPI, MemTest, etc when I looked for stability... I dont expect everyone to use the same programs, I jus want people to say "HOW" stable the overclocks were.

I will add you results once I have more info :toast:

[computersmsa]

Stability -> OCCT test :)
http://www.ocbase.com/ Dream software (French)
http://www.ocbase.com/base.html -> Data base of stable ocing :) (new)

I test super PI 4m one days :D


For max @ vcore -> I test @ 240, 241 ... :)