Core i7 Vdimm/VTT Graveyard :D

[Amurtigress]

Bei Fei, Hondacity,

you might want to have a look at post 223 that I just edited. It might shed some light on the whole matter of VDIMM....

[Amurtigress]

Hello, (continued from post #223)

I just realized that it would have another consequence if Intel really designed their CPUs to have the uncore powered by the same voltage as the DIMM sockets.

It would mean that the uncore/QPI voltage would always be the same as the standard JEDEC compliant voltage for normal DDR3 DIMMs. Namely, 1.5V!

As a consequence of this, we should probably do the following:

1) Start considering 1.5V to be the default voltage for the QPI/uncore part of our i7s. And thus stop wondering if any lower voltage than that could kill our precious chip.

2) Reconsider the voltages we use depending on the actual DDR3 clock rates we are torturing our uncores with.

So I am suggesting the following table for uncore/QPI:

DDR3 clock uncore voltage DIMM Voltage

1066 1.5V (JEDEC compliant) 1.5V
1333 1.5V (JEDEC compliant) 1.5V
1600 1.55V See specs of your modules but keep below 2V
1800 1.6V (suggested by Corsair for some of their products)
2000 1.65V

Certainly this has to be tested on any individual setup. And the fact that for example ASUS is defaulting to 1.2V for QPI doesn't make it correct. I rather see what Intel built the CPUs for and stick to that.

It might as well be possible that stable operation of the IMC below 1.5V is a design precaution by Intel to support future die-shrinked memory chips that will stick to upcoming JEDEC specifications, for example Samsung who announced 1.35V DDR3 chips. 1.35V uncore would be supported right away then.

Any thoughts? I hope I am not ridiculing myself here with such a theory.

[saaya]

the imc isnt powered by vdimm afaik, but it needs vdimm since the memory bus is using it, and its interfacing with that bus.
so some of the logic obviously is exposed to vdimm, i dont know how exactly it works, but this logic can work fine with a higher signalling voltage on one side compared to the other, ie higher vdimm than vtt or higher vtt than vdimm, but if the difference is too high, the logic will degrade and at some point burn out.
the ideal situation is where vdimm=vtt i guess, but it really doesnt hurt having higher vdimm than vtt, as long as its not too high.
intels default vtt is around 1.1, not sure on the specs, but they accept lower than this, so lets say 1.05 or 1.00 is still acceptable.
and for vdimm they say more than 1.65v is not ok, again probably 1.75 is still acceptable, so that means a voltage diference of up to 750mv is probably acceptable, which means 1.1 vtt is ok to run 1.85v vdimm, and for 2.3v vdimm 1.55v vtt would actually be enough. then again thats pushing the limits, and more importantly, it really doesnt make sense to run that high vdimm with that low vtt... if you run high vdimm its cause your tweaking your mem to get a tiny bit of extra performance, and if your doing that, your more than likely to be tweaking uncore as well to clock it as high as possible, which means youll be running vtt at 1.55v at least.

if you look at the usual combinations of vdimm and vtt people are running youll notice that nobody is running them more than .5v apart, even if they dont care about the voltage diferential. thats probably why we havent seen more than a handful of i7 chips go bust from this, at least that we know of... and even those might just be freak accidents and not really caused by vdimm and vtt being too far off from each other.

BUT, the newer chips requiring less and less vtt to clock high with might be a reason to worry, but actually the newer memory all needs less vdimm to max out as well, so then again things seem fine... combining d0 or future chips with memory that needs 2.3v to max out might be problematic... but even that is unlikely to cause problems as barely anybody would run that high vdimm 24/7

[Hondacity]

Amurtigress

i just read it, sorry i push my voltages above the values you just wrote. anyways if you want to stay safe and have a good oc on your ram, vtt should be 1.5v-1.65v, intel specs are rather too safe... on the XS overclockers perspective we tend to go beyond the vmax specs. kill the proc or not, it happens

[dopestuff]

hey people i don't know it this is realy offtopic but...
i have to run my qpi @ 1,35 V for 1520 mhz mem and qpi @ 1,45 V for 1866mhz mem (1866mhz rated mem)
howevver i thought my mobo/cpu was faulty (still not sure wich) but according to your theories that kind of voltage is normal?? (still need to learn soooo much about i7 but do wanna learn )

[Jaco]

I thought Intel says 1.35v VTT/QPI max .
1.6v VTT/QPI is just crazy in my opinion.
I don't think any RAM manufacturer would recommend such crazy voltage

Note jcool damaged his cpu with 1.55v QPI in a matter of weeks !!
(this is 24/7, full load)

[saaya]

why did you think your cpu or mobo are faulty dopestuff?

I thought Intel says 1.35v VTT/QPI max .
1.6v VTT/QPI is just crazy in my opinion.
I don't think any RAM manufacturer would recommend such crazy voltage

read through this thread and google, there are xmp profiles with 1.7v vtt, and its not programmed by accident...
i know a few people running up to 1.7v vtt on air, for months, without problems...

Note jcool damaged his cpu with 1.55v QPI in a matter of weeks !!
(this is 24/7, full load)

how would he know it was vtt that damaged his cpu?
how is it damaged? what happened?
maybe the newer d0 chips are more sensitive to high vtt...
or maybe his board has some problem with high vtt... you have to know when to increase vtt during the system initialization... and vtt stability during system initialization is very risky, if it fluctuates during the early stages of initializing the cpu, it might get damaged or die pretty quickly.

[dopestuff]

@ saaya: because everyone gets there whit under 1,3 qpi and 0,1V qpi for only 310 mhz i thought was kinda high... my cpu further is pretty good though it does 4 ghz ht on @ 1,376Vcore prime stable (its a C0)

[Amurtigress]

@ saaya: because everyone gets there whit under 1,3 qpi and 0,1V qpi for only 310 mhz i thought was kinda high... my cpu further is pretty good though it does 4 ghz ht on @ 1,376Vcore prime stable (its a C0)

I can tell you why they do, they rely on the board maker's defaults. But the masses aren't always right...

Here's what I found through the link in TheRAMGuy's signature:

http://www.houseofhelp.com/forums/sh...ad.php?t=54052

Quote:

Motherboard make and model : ASUS P6T Premium/Deluxe !
Each set is tested and packaged together using the following settings:
Advanced
Jumper Free
CPU Ratio Control: Depends on CPU
FSB Frequency: AUTO
PCIE Frequency: 100
DRAM Frequency: DDR3-1866 MHz
DRAM Timings Control: Manual
CAS# Latency:___________7
RAS# to CAS# Delay:______8
RAS# Precharge:__________7
RAS# Activate to Precharge:20
DRAM Voltage:_________1.65V
QPI Voltage:___________1.60 V
XMP1.2 settings
All settings not listed should be left at AUTO or BIOS Default
Maximum DDR3 recommended VDIMM: 2.00 Volts

End Quote.

[dopestuff]

so there is nothing wrong whit my cpu/mobo that's nice and according to everything i have learned so far i can also up the dram volts a bit if i want/have to
i have been folowing this thread for some time now and things are getting a bit clearer...
i will keep folowing this thread and do greatly thank you all for sharing your knowledge!!

[Amurtigress]

so there is nothing wrong whit my cpu/mobo that's nice and according to everything i have learned so far i can also up the dram volts a bit if i want/have to
i have been folowing this thread for some time now and things are getting a bit clearer...
i will keep folowing this thread and do greatly thank you all for sharing your knowledge!!

You're welcome, but I wish we were so sure about it. Frankly, all I can do is looking for official things like this on Corsair's support forums, and hope they are right. They're supposed to know it better than I do....

I do have 17 years of experience in making my own computers out of parts and an education in electronics, but I am certainly not an insider who works at Intel...

I can tell that much, my system is stable now at 1.55/1.6V VTT/Uncore, and 1.9V VDIMM.

I hope this means my rig 'feels' good and won't blow. My only basis to judge from.

[saaya]

@ saaya: because everyone gets there whit under 1,3 qpi and 0,1V qpi for only 310 mhz i thought was kinda high... my cpu further is pretty good though it does 4 ghz ht on @ 1,376Vcore prime stable (its a C0)

hmmm yeah ive seen some people run 2000 with 1.3v vtt...
well some imcs are better than others i guess... and maybe it also has to do with what mem and timings and vdimm you use, plus what bclock and mem multiplier and uncore multiplier...

[dopestuff]

hmmm yeah ive seen some people run 2000 with 1.3v vtt...
well some imcs are better than others i guess... and maybe it also has to do with what mem and timings and vdimm you use, plus what bclock and mem multiplier and uncore multiplier...

so in other words part of it is hardware and the other part a LOT of playing around whit settings (or in other words the dude behind it)

now running stable whit 1900 9-9-9-24 t2 mem whit 1,654V mem and 1,47V qpi (maybe in near future some Vdimm overvolting and see if theese corsairs are worth the money )

[Amurtigress]

Hello,

I don't know if this document has been referred to here before. Intel gives some clear numbers on voltages there...

http://download.intel.com/design/pro...hts/320834.pdf

See page 22 and 23.

Page 22 has a table of absolute maximum voltages, and page 23 gives us a number of nominal voltages on the single power panes.

Typical for VDDQ (VDIMM) would be 1.425-1.575/1.5V, and an absolute maximum of 1.875V on page 22.

The maximum for QPI/VTTA/VTTD should be 1.35V according to this. (there goes my theory! Oh well... )

Maximum core voltage is listed as 1.55V.

Intel's state ment of app. 1.875V on VDIMM fits well to ASUS' BIOS, which allows 1.9V. The extra voltages are only acessible through an overvolt jumper.

Maybe this helps a bit?

To my experiences, these limitations do not mean "More than that kills your processor", but normally "These are the limit that we, Intel, are guaranteeing nominal operation for as stated in this document."

[naokaji]

BUT, the newer chips requiring less and less vtt to clock high with might be a reason to worry, but actually the newer memory all needs less vdimm to max out as well, so then again things seem fine... combining d0 or future chips with memory that needs 2.3v to max out might be problematic... but even that is unlikely to cause problems as barely anybody would run that high vdimm 24/7

For benching it's ok to use the old ram that needs high volts with a D0 I7.
1.6VTT and 2.1Vdimm here and everything is fine, however, I would not try it for 24/7.

[saaya]

so in other words part of it is hardware and the other part a LOT of playing around whit settings (or in other words the dude behind it)

now running stable whit 1900 9-9-9-24 t2 mem whit 1,654V mem and 1,47V qpi (maybe in near future some Vdimm overvolting and see if theese corsairs are worth the money )

i think its 60% hw if not more... good mem plus good cpu = high mem clocks, without much tweaking or high voltages...

check out anandtechs classified review, they were limited by samsung based mem i think, went to elpida based mem and wham, 20mhz higher bclock and higher mem clocks with tighter timings iirc...

Hello,

I don't know if this document has been referred to here before. Intel gives some clear numbers on voltages there...

http://download.intel.com/design/pro...hts/320834.pdf

See page 22 and 23.

Page 22 has a table of absolute maximum voltages, and page 23 gives us a number of nominal voltages on the single power panes.

Typical for VDDQ (VDIMM) would be 1.425-1.575/1.5V, and an absolute maximum of 1.875V on page 22.

The maximum for QPI/VTTA/VTTD should be 1.35V according to this. (there goes my theory! Oh well... )

Maximum core voltage is listed as 1.55V.

Intel's state ment of app. 1.875V on VDIMM fits well to ASUS' BIOS, which allows 1.9V. The extra voltages are only acessible through an overvolt jumper.

Maybe this helps a bit?

To my experiences, these limitations do not mean "More than that kills your processor", but normally "These are the limit that we, Intel, are guaranteeing nominal operation for as stated in this document."

you obviously didnt read this thread... the doc you mention is referred to in the first page of this thread, and the absolute max values are posted there :P

those values are the absolute max values a mainboard should not exceed at any moment. so voltages can vary and fluctuate but should never exceed those values or the chip MIGHT get damaged after some time or even die.
there were cases when cpus really did die if the max voltages were exceeded, but thats rather rare...

those numbers are more of a last line of slight guarantees on intels side as what they guarantee to work... beyond that, if anything goes wrong, your on your own and cant say its intels fault.

naokaji, thx
whats the sweet spot for vtt on your d0?
1.6? or rather below? i heard d0 has rather low sweet spots for vtt while c0 seems to have its sweet spot around 1.7v more or less

[naokaji]

naokaji, thx
whats the sweet spot for vtt on your d0?
1.6? or rather below? i heard d0 has rather low sweet spots for vtt while c0 seems to have its sweet spot around 1.7v more or less

1.6 is the sweetspot for my D0 if I want to tighten the latencies, especially Tras and roundtrip, if I leave those on loose settings though the sweetspot moves down to 1.5 - 1.55.

[saaya]

ok... same as c0, but c0 usually scales until higher than 1.6v from what i heard...

[erickwok]

So, i7 launched and we are all sufficiently scared of using more than 1.65v vdimm alright. Looking at the real world performance boost from running high mem clocks over easily achievable clocks with 1.65v vdimm or even less, it seems high clocks and hence high vdimm isnt needed for gaming and 24/7 at all...
but THIS IS SPA... XtremeSystems!
I know there are quite some people out there pushing for more, a lot more

Please share your results here and i will do my best to update the first post and summarize the experiences. Did you kill an i7 with high vdimm? What vtt and vcore did you use? for how long? Are you running high vdimm without any problems? how high? whats your vtt? whats your vcore?

Thanks!

1.90v vdimm / 1.20v vtt = dead after 1 day (Engineering Sample CPU rev B) - rev B?
1.90v vdimm / 1.20v vtt = dead after 10 days (Retail 965 rev C) - most likely freak accident/bios/board bug
2.25v vdimm / 1.45v vtt = dead after 1 day (Engineering Sample CPU rev C) - buggy board?
1.88v vdimm / 1.50v vtt= dead after 1 week (Retail 965 rev C) - most likely freak accident/bios/board bug

1.65v vdimm / 1.20v vtt = ok forever according to intel
1.80v vdimm / 1.30v vtt = ok for 7 weeks
1.95v vdimm / 1.40v vtt = ok for 5 weeks
2.05v vdimm / 1.45v vtt = ok for 5 weeks
2.05v vdimm / 1.36v vtt = ok for 6 weeks
2.00v vdimm / 1.50v vtt = ok for 8 weeks (cold)
2.00v vdimm / 1.60v vtt = ok for 2 weks
2.00v vdimm / 1.65v vtt = ok for 8 weks
2.25v vdimm / 1.70v vtt = ok for 4 weeks
2.30v vdimm / 1.35v vtt = ok for 2 week (cold)
1.75v vdimm / 1.75v vtt = ok for 1 day (cold)

Conclusion:
Its been 6 months since Intel launched i7 and there have been less than half a dozen reported dead cpus.
Considering all the torturing those chips go through with ln2 cooling and vcores in excess of 1.7v, vdimm in excess of 2.2v and vtt in excess of 1.7v, this is very good news.

1.6v vcore, 1.6v vtt and 2v vdimm
These seem to be 100% safe, as long as the processor is cooled well enough.
I think even higher voltages are fine but there hasnt been a lot of feedback of people using high voltages 24/7 so far.

For extreme cooling there seems to be almost no limit as to what voltage is ok, as long as the cpu still scales, its probably fine

Hi Saaya

if my i7 set dimm 1.8V and vtt at 1.55V
how long will be dead (approx.) under air cooling

[tistou77]

hello

for my Corsair Dominator GT 2000MHz, you need to vQPI 1.49v in bios (Rampage II Extreme) in idle 1.46v and ~1.43v in load
and 1.65v for vDIMM

for a 920D0, is that these voltages are safe for the h24 or is it dangerous (watercooling)?

thank you very much

[devsk]

hello

for my Corsair Dominator GT 2000MHz, you need to vQPI 1.49v in bios (Rampage II Extreme) in idle 1.46v and ~1.43v in load
and 1.65v for vDIMM

for a 920D0, is that these voltages are safe for the h24 or is it dangerous (watercooling)?

thank you very much

I think, and I hope (because I am also running 1.45 VTT), that 1.45vtt is fine for 24/7 use. As long as temps are kept under check.

[Mafio]

vdimm 2.25
vtt 1.75

ok for 3 days at air

[Mescalamba]

Hi Saaya

if my i7 set dimm 1.8V and vtt at 1.55V
how long will be dead (approx.) under air cooling

Depends if your mobo undervolts VTT or not, if you have some chance to test it with multimeter, try it. But even with 0.05V undervolt.. it should be 100% safe like forever..

Btw. what's your mobo?

[infected_]

im using 200*21 on my 920 D0, using multi 10x for the rams, gskill trident pc3-16000 6GB, but it requires 1.63125v qpiv. it this safe for 24/7? my mobo is P6T "vanilla". i have good ventilation, noctua nh-u12p push pull + antec 1200 (7x120,1x200)

thanks,

infected_

ps: btw, running with that qpiv for 10 days now, without probs.

[Leeghoofd]

I thought Intel says 1.35v VTT/QPI max .
1.6v VTT/QPI is just crazy in my opinion.
I don't think any RAM manufacturer would recommend such crazy voltage

Why do they put in their XMP profile then 1.65QPI volts for 2000mhz speeds ? some CPU's like rams etc will die,degrade when being overclocked. Remember though the numbers are very very small compared to the others that run fab... it's a risk you take when you push stuff.

FYI I smoked on of them Kingston PC16000C8 rams, even when it was at only 1900mhz 1.63Vdimm (not even at stock speeds/voltages) These things just happen sadly