Core i7 Vdimm/VTT Graveyard :D

[Chicken Patty]

The EVGA X58 does raise the voltages under load.

i agree.

idle for me right now is 1.107v load is 1.141v

[Utnorris]

I agree on the EVGA and also so does the Blood Rage. This is when Vdroop compensation is enabled.

[Chicken Patty]

I agree on the EVGA and also so does the Blood Rage. This is when Vdroop compensation is enabled.

i have mine disabled. Still does it. Normal on x58 maybe?

[BenchZowner]

Don't know if it's Vdimm or Vtt ( Vqpi ) related but I think I've lost 1 channel of the DDR3 controller in my first i7 920 C0 retail CPU
I say I think because I'm not entirely sure that the CPU is damaged, it could be the board, and it could also be a bent socket pin ( planning to check it out tomorrow ).
I shall update this post once I have done my research and have a conclusion and autopsy report ready

[sofos1990]

Bill, today is not your lucky day man.

[BenchZowner]

I think I'm gonna quit ( for today ).
I better do. I'm afraid that if I leave any of my PCs running tonight I'll wake up and find another problem as well.

[systemviper]

I've been running around 1.5-1.57v lately for some stability, I wouldn't have expected this UD4P to be so good but it just took to this i920. Got to love the Bclock, I have a feeling that if i juice it, i can get the elusive bclock 225 = 4.7Ghz





.





.

[eva2000]

Nice gotta set up my Gigabyte X58-UD4P soon... is that the real memory timings shown in cpuz validation 10-12-12-31 ? or bug in cpuz ?

[systemviper]

I think that is the real ones, at that point i was running them on auto...

[BenchZowner]

Don't know if it's Vdimm or Vtt ( Vqpi ) related but I think I've lost 1 channel of the DDR3 controller in my first i7 920 C0 retail CPU
I say I think because I'm not entirely sure that the CPU is damaged, it could be the board, and it could also be a bent socket pin ( planning to check it out tomorrow ).
I shall update this post once I have done my research and have a conclusion and autopsy report ready

Looks like it was caused by 1 or 2 "possibly" bent pins in the socket.
Fixed it anyways

[sofos1990]

My i7 965 also damaged on my R2E.
I'll try the same chip on my BloodRage maybe it plays, but i don't think so...
VTT:1.6V
Vdimm:1.88V
Vcore:1.4V

[Bandit_44]

1,475 VTT and 1.78 vdimm for now. CPU needs a LOT of volt. Mem 8-8-8-18-1T.

Hey Loco,,who makes that blue test bench in your post?? Thanks in Advance,I Love it!

[saaya]

run your chips a week/month long with ~1.58vcore/1.65qpi/1.96vdimm on air bro

i actually did, more than a month... less vcore, but around 1.7v vtt and 2v vdimm

Who the hell is mark?
Anyway I did run up to 2V Vdimm and 1,55V VTT for short periods of time (like 10 mins) but that shouldn't have caused it to die... as the temps were subzero. I couldn't detect any signs of corrosion on the CPU pins either.

lol, my bad, i confused you with somebody else

My i7 965 also damaged on my R2E.
I'll try the same chip on my BloodRage maybe it plays, but i don't think so...
VTT:1.6V
Vdimm:1.88V
Vcore:1.4V

any updates? is the cpu really dead? how long did you run those voltages? 24/7? for how many weeks/months? did the cpu die when pushing it ie changing the voltages? if yes, then it might have been a frealk accident, otherwise it sounds like the cpu degraded or died from high vtt or vdimm hmmmm strange since you got vtt and vdimm so close to each other...

Bandit_44, that test bench is made by dimas
good stuff! we have a bunch here in our QuantumForce lab...

[sofos1990]

The cpu is dead. No, I didn't use those voltages for 24/7, only for a few hours. The cpu died when I was trying to reboot my system from bios(pressing F10).

[saaya]

The cpu is dead. No, I didn't use those voltages for 24/7, only for a few hours. The cpu died when I was trying to reboot my system from bios(pressing F10).

hmmmmm what was the last change you made, do you remember?
if it was vtt or vdimm or vcore, then its possible that the voltages fluctuated and that killed the cpu... it shouldnt happen, but its possible depending on how the bios and board apply the voltages during post.

i saw that on some boards vtt starts at default and then jumps to the bios setting, this can result in vtt fluctuating and that can kill the cpu. actually fluctuating vtt is the fastest and easiest way to kill an i7 cpu, vcore is very tough and vdimm is ok, but vtt...
if the vtt is missing or present at the wrong time, or its fluctuating then the cpu will die immediatly. during the early engineering stages of building x58 boards almost all the cpus that were damaged or died in our labs were a result of vtt not working properly. it was only half a dozen or maybe a dozen cpus overall during a period of a couple of months, but out of those i think 80% died from vtt related problems and another 10% from vcore (2.4v heheh) and the rest from some other problems like heat or qpi or vdimm.

i think this was a freak accident... it could be a board problem, but i havent heard of more cpus dieing in that board... so i assume it was some bios glitch or hardware glitch... hope you could get it replaced with an equal or better clocking chip

[sofos1990]

Yes, the last change I made, was the vdimm if I remember well. I had too many problems with that mobo, especially with memory overclocking.
I hope my next chip being better!
Thanks for the information.

[Bandit_44]

Thanks a Bunch Saaya,I really like that,the next thing is to find one,lol!

[cadupradoo]

I know it's not very aggressive, but i'll post anyway :P

BCLK = 200, HT OFF Turbo ON

vcore: 1.46V
CPU VTT: 1.5
vdimm: 1.71

For a month now, but I think I'm going back to 4.1, with 1.38vcore and 1.4vtt.. it's not worth..

[Amurtigress]

Hi people,

my i7 920 system has been working rather well for four weeks now. I am using a ASUS P6T Deluxe V2 with V1 BIOS...

VDIMM has been at or below 1.65V 2/3 of the time running 1200 MHz, and 1/3 of it at 1.8V/1600 MHz. My machine is running roughly 14h a day, so, do the math.

I haven't had any signs of degradation or other issues due to the 1.8V. However my system won't be all stable at DDR3-1600 mode and 4x1GB modules. I suspect the IMC is having it's limits there.

As for occasionally not detected memory modules, in my case they have always been cases where Vdimm has been insufficient (1.65V at 1600, for example) or very tigh timings that the IMC apparently couldn't cope with.

[saaya]

thx for sharing your experience
what vtt are you running? i think you just need more vtt, i dont think any imc craps out at 1600 with 4 sticks already

[Amurtigress]

Hi Saaya,

by Vtt I assume you mean the uncore voltage. My uncore is set to either 1.35V or 1.375V. I found that setting the NB and CPU clock skews to something else than Auto, in the range of 100-400 ps, would increase stability on the system and reduce sudden reboots/freezes to maybe every 6-12 hours at 1600 DDR3 clock. There is a higher chance of provoking a freeze or reset when *starting* 64 bit prime 95 on all eight threads, and after having run other heavy stuff for a while like LinX or Secondlife.

I still had those clock skews in mind from the GTL bus tweaking thread, I tried to avoid the two clock signals to crosstalk by delaying both by a different amount of time.

[Mescalamba]

Pretty low uncore.. yes VTT is uncore. At 1600 you need around +100mV VTT and I guess it will be another +100 on 1800MHz. Also its very very important value when trying to reach higher bclk.

I used clock skew only when I needed BCLK to reach 210+. Shouldn't be needed, but I have Bloodrage, its probably different on ASUS.. Kinda don't get that 4x1GB modules.. ehm, why? You don't like tri-channel?

[Amurtigress]

Mesca:

No, the reason is rather simple. I bought my Corsair XMS3 modules together with a P5Q3 Del. board about seven months ago, as a first step towards the forseeable i7 transition. I tried to avoid having to buy THREE components at a time...That's also why they are 1.8V modules.

So basically you're saying I have to set uncore to at least 1.45V? Going to try that.

Makes sense however. My experiences with the P45 and X38 chipset were such that whenever extra load arises (ex. four modules instead of two, quadcore instead of dual), the northbridge required extra voltage. For example the X38 would only be stable at FSB 460 and roughly between 1.55V and 1.61. 1.36-1.44 for the die-shrinked P45.

[saaya]

Hi Saaya,

by Vtt I assume you mean the uncore voltage. My uncore is set to either 1.35V or 1.375V. I found that setting the NB and CPU clock skews to something else than Auto, in the range of 100-400 ps, would increase stability on the system and reduce sudden reboots/freezes to maybe every 6-12 hours at 1600 DDR3 clock. There is a higher chance of provoking a freeze or reset when *starting* 64 bit prime 95 on all eight threads, and after having run other heavy stuff for a while like LinX or Secondlife.

I still had those clock skews in mind from the GTL bus tweaking thread, I tried to avoid the two clock signals to crosstalk by delaying both by a different amount of time.

i dont think its a problem to have them at the same clock skew setting since the clock skew delay should match the trace lengths, and cpu to qpi and cpu to mem traces are not equally long/mem clock?
1.4v vtt almost i guess... and you still cant get a higher bclock?
try a bit more, and try to raise ioh, it helps to get high bclocks/qpi clocks stable under load.
dont know how asus calls it, i think qpi pll?

p35/x38/x48 reached higher mem clocks or the same mem clocks with less vdimm with increased nb volts. it varied from chipset to chipset, but most stopped scaling around 1.8v nb voltage. most cpus seem to stop scaling at around 1.7v. and one more thing, more doesnt always mean better, just like with northbridge chipsets i heard too much vtt can mean worse clocks as well, so the finer you can adjust vtt the better.

thats why we increased the vtt setting granularity in the latest beta bios for BR btw. it seems the high and low points for vtt are set automatically by intels mrc code in the bios when the system boots, if vtt is too high then the ref points will be set different which might cause worse stability.

oh and increasing vtt too much during post causes problems with intels mrc code, we kinda worked around that in the bios by applying vtt in steps during the post process, but if you want to push for suicide shots i recommend increasing vtt while your in windows. but you should boot with a higher vtt already so the ref points are set accordingly by the intel mrc code.
experiment, im not sure whats better, booting with low vtt and then bumping it up in windows or booting with as high vtt as possible and then bumping it further in windows. maybe even booting with huge vtt and reducing it in windows will help... maybe the mrc code then sets a higher tolerance values for vtt ref points, and while that hiogh vtt might be unstable, then reducing it while keeping those ref points may result in a higher stable bclock/mem clock.

i find that tweaking bclocks is not really worth the effort though, its more a fun thing then really useful for 24/7 as even the best boards with the most tweaking and super cooling cant run a notably higher bclock than 222 stable, its just a few mhz which makes a microsocopic diference if any even in synthetic benchmarks...

[Amurtigress]

Saaya,

I can only subscribe to your point of view on tweaking the HELL out of my board, like bclock. I am not event rying to experiment much in that regard, I just tried to go towards 220 to get a feel for the new hardware initially. I can see that around 4 GHz the Intel 45nm CPUs are going into a progression curve where the amount of voltage that's needed to get higher is getting ridiculous as compared to the actual gain in performance. P=U*I comes to mind here, the heat produced by the extra voltage multiplied by the caused extra current really isn't worth it.

I am running at 19x200, but what I do is only changing the bclock to memory clock ratio to have 1600 instead of 1200, and certainly the VDIMM voltage and timings accordingly.

To clarify that: I am not getting any memory erorrs there. No stopping Prime95 threads, no memtest errors, no LinX errors. Just plain reboots.

Edit: I am currently running at 1.4V IOH and 1.45V uncore. 1.26 IOH and 1.45 uncore gave me a sudden reboot two hours ago.