F317066CL9T-6GBPS REPORT - problems:insane voltages needed to run way lower than spec

[moebius]

I have seen there are similar combination already tested.
FOr example Asus P6T deluxe V2 tested with Perfect Storm F3-16000CL9T.
Given that both motherboard and memory are inferior i expected better results...
ANyway ,your answer means that you cannot yet answer whether this memory kit is defective or not?

[GSKILL TECH]

I do not believe it is an issue with the memory, I believe it is the BIOS. I will test it out in a couple minutes.

Thank you
GSKILL SUPPORT

[moebius]

Well i hoped so in the beginning as well,but i tried them all ,as you can read in the first post.
I tried your settings too,but they were worst(if possible )than the ones i already tried. It seems they simply won't do 2133.
2000 is not doable as well.165V on cpuvtt is not what i'd call "doable".
And with such voltages 2000 is the fastest AND stable thing obtainable with these sticks.With less voltage they keep on having memtest errors or freeze or don't post.
Then another thing: shouldn't it work out of the box when using XMP profile? Or at least with minor tweakings
ANd if it gives errors or don't post in such conditions,how can it be a bios problem?
Again i hope it is like you say,but until now it simply doesn't seem the case.
I read in this same forum of an user that with the same kit managed 2000 stable(or so it seemed from the tests he made)with 1.45 on cpuvtt.Cpu was an i7 920 with bclk at 200 and cpuspeed at 4.0 Ghz.I perfectly know that cpu are not all the same but these differences seem a bit exaggerated.
With that voltage to me it doesn't even post.
Btw , the last advice you gave was one week ago.5 days ago you told you would have tested this combination.Did you ?
regards

[GSKILL TECH]

Yes I did get to testing a little. Seems like there is a block at DDR3-2000. At this and below I was able to pass all tests and run error free, but anything above such as DDR3-2133 was not completely stable.

In order to operate high frequencies, high voltages are needed, which is why the Intel and JEDEC standard is DDR3-1333. For the XMP Profile, it depends whether the motherboard manufacturer has set the XMP Profile for the specific memory package. In this case, it is able to pick up the correct memory settings, but just like the problem we are having, it is unable to operate DDR3-2133 possibly due to the high manual voltages necessary.

The reason why I know it is not the memory is because I can plug these into any 2133 certified board and not have any issues.

Thank you
GSKILL SUPPORT

[moebius]

I think that "rampage extreme II" and "ddr3 2000 wall" cannot really stay in the same sentence.I can address you to heaps of reviews/threads that contraddict what you deducted from your test.It is more likely an incompatibility between Perfect Storm kit and the mobo.
Anyway here the problem is that even to operate at ddr3 2000 THIS PARTICULAR KIT THAT I BOUGHT needs 1.65 cpuvtt and 1.75 dramv. SO ,even in the case of an unlikely rampage extreme II wall there is anyway something going wrong.If i didn't use these voltages(i am talking about 2000) memory would keep on giving me memtest errors or freezes or no post.Anyway i appreciate you have simulated my same test conditions and you are giving me feedback.
Regards

p.s. :this is what i managed to find with 1 min search on bing .Trident kit tested on Rampage extreme II :
http://www.hardwarecanucks.com/forum...it-review.html
As you can see 2100 is attainable with reasonable voltages(1.5 on water is reasonable to me): 1.5 vtt and 1.65 dramV
Your adviced settings in this thread http://gskill.us/forum/showthread.php?t=1166 seems to suggest similar settings. You advice for 1.55 (that is little more than 1.5 with multimeter with Rampage Extreme II) and they use 1.5. How is it possible that 2133 PS kit needs 1.65V(cpuvtt ) and 1.75(dramV)?more than 2000 Trident kit on the same mobo?

[GSKILL TECH]

Well I did not mean maxing out the motherboard, I meant within the proper specifications, 1.65V and without going 1.70V+ on the VTT.

Here are the settings I used to run DDR3-2000 stable. 2133 passed some tests, but during torture tests I was given odd errors.

AI overclock - manual
OC from CPU level up - auto
OC from Memory level up - auto
CPU Ratio setting - 20
CPU Turbo Power Limit - enable

CPU configuration
CPU Ratio Setting - 20
C1E Support - disable
Hardware prefetcher - enable
Adjacent Cache line prefetcher - enable
Intel Virtualization tech - disabled
CPU TM Function - disable
Execute Disabled bit - disabled
Intel HT Technology - disable
Active Processor Cores - all
A20M - disabled
Intel C-STATE tech - disabled

BCLK frequency - 143
PCIE frequency - 100
DRAM frequency - 2006 MHz
UCLK frequency - AUTO
QPI frequency - auto

Dram Timing Control
DRAM CAS# Latency - 9
DRAM RAS# to CAS# Delay - 9
DRAM RAS# PRE Time - 9
DRAM RAS# ACT Time - 24
DRAM Timing Mode - 2N
All others can be on AUTO

EPU II Phase Control - Full Phase
Load-line calibration - enable
CPU differential amplitude - auto
Extreme OV - disabled


CPU voltage - AUTO
CPU PLL voltage - 1.90V
QPI-DRAM voltage - 1.575V

IOH voltage - 1.35V
IOH PCIE voltage - 1.60V
ICH voltage- 1.20V
ICH PCIE voltage - 1.60V

DRAM Bus voltage 1.65V

CPU spread spectrum - disabled
PCIE spectrum - disabled
CPU clock skew - auto
IOH clock skew - auto


Thank you
GSKILL SUPPORT

[moebius]

I have checked your results.
DId you try mine?
AI overclock - manual
OC from CPU level up - auto
OC from Memory level up - auto
CPU Ratio setting - 21

CPU configuration
CPU Ratio Setting - 21
C1E Support - disable
Hardware prefetcher - enable
Adjacent Cache line prefetcher - enable
Intel Virtualization tech - enable
CPU TM Function - disable
Execute Disabled bit - enable
Intel HT Technology - enable
Active Processor Cores - all
A20M - disabled
Intel Speedstep tech - disable
Intel C-STATE tech - disabled

BCLK frequency - 200
PCIE frequency - 100
DRAM frequency - DDR3 2005MHz
UCLK frequency - 4009
QPI frequency - 7218 MT/s

Dram timing control
1st iformation - Insert Timings Here in 9(tCL)-9(tRCD)-9(tRP)-24(tRAS) -98(tRFC)-
2nd information - Insert Timings Here in 2N(CR)-
3rd information - Insert Timings Here in
EPU II phase control - full phase
Load-line calibration - enable
CPU differential amplitude - auto
Extreme OV - disabled


Cpu voltage - 1.375
CPU PLL voltage - 1.88
QPI-Dram voltage - 1.64

IOH voltage - 1.41
IOH PCIE voltage - 1.51
ICH voltage- 1.113
ICH PCIE voltage - 1.51


DRAM Bus voltage - 1.73
DRAM REF voltage - auto

Debug mode - string
Keyboard TweakIt control - enable

CPU spread spectrum - disabled
PCIE spectrum - disabled
CPU clock skew - normal
IOH clock skew - normal


It is true that i used higher cpuvtt(1.64 from bios) and higher dramV(1.736 that anyway is not critical),but i managed 4.2 Ghz with ddr3 2000.
I noticed one thing very strange.It doesn't seem that lowering cpu overclocks influences that much voltages required to attain 2000.
Anyway i can tell one thing for sure:this memory kit gave to me (and to you from what i see) very odd results. You needed almost 1.6 V on cpuvtt to reach 143 bclk with ddr3 2000.
You will agree that this is is indeed a very poor result .On my side i needed 1.64 (cpuvtt )and 1.74(ddr3V) to reach 200*21 rock stable ddr3 2000.
Even this is a very poor result,not for the clocks themselves but for voltages needed.
Did you check the links i sent you.Trident Kit seems to behave much better.That's real weird!
I'll tell you that who buys this ram,expects to see stable results with >2100 ddr3 and around 210 bclk with cpu that can keep these speeds OR AT LEAST tight timings and relatively low voltages with ddr3 2000 for a daily use.
What i saw until now are the performance of a cheap ddr3 2000 kit,but i better say of an average ddr3 1866 kit not those that who buys 400$(here in europe) memory kit expects on such high end mobo.ANyway when i talked about the fact that rampage extreme II has ddr3 wall way higher than 2000 i was not talking about insane voltages but about voltages much lower than the ones we used.

[moebius]

With further tweakings i managed to make ddr3 2000 CR=1N stable with ioh=1.55 , dramv=1.73 , cpuvtt=1.6625 and cpu 200bclk*21.
I managed to tighten timings to 7-9-6-20.
Stability test used was Intel Burn Test (xtreme stress mode(right click on start buttonin the main window),32 threads,20 pass,maximum memory.Anything less is NOT stable.People that say that 10 pass of linx with 8 threads is a stable system are soooo wrong.
What stability test did you use for your tests?

[GSKILL TECH]

We run a series of tests with all of the various burn tests at maximum intensity. If they all pass, it is considered stable.

Thank you
GSKILL SUPPORT

[moebius]

and which programs and tests do you run?

[GSKILL TECH]

LinX, Intel, PassMark, Prime, there are many programs available to use.

Thank you
GSKILL SUPPORT

[moebius]

The heaviest and more reliable is IBT in the way i test it.In what way do you test IBT?With what parameters?I am curious to know what is that you call "stable".
regards

[mcf.beyer]

Who can tell?

Hi moebius,
since you seem to be experieced with memory, I am seeking an answer to who is the manufacturer of these 1.35 low voltage chips used on DDR3 ECO CL7?
Do you have any idea on that?
The heatspreader looks like it is not detachable.

[moebius]

I can't tell for sure,didn't find anyone that opened them,what i can tell is this:
In end of october were announced new Samsung 40 nm ICs that were rated at 1.35 V
Eldpida Hyper have 1.35 jedec spec.
It can be the new Hynix H5TC2G43AFR, it operates at 1,35V. also.
So these are possibiliities...I know that i listed all main ICs manufacterer.
I advice you this thread where there are benchmarks and overclocks numbers,but i would give an eye to what they call stable...People like to post screens of Everest or memtest but it is of no use if you can have stability issues in programs...
http://www.xtremesystems.org/forums/...d.php?t=239426
YOu will find a couple of links in it and there is a G.SKill tech posting in it.

Anyway if you are interested open a new thread.This is about Perfect Storm 2133 c9

[moebius]

With what parameters do you run IBT? Why do you use LinX that is a lighter version of IBT and gives much less warranties of stability?