All Haswell degradation info in here please.

[K404]

I might have just had a chip completely fall over on me. Yesterday, I was benching 5.625GHz, 1.70-1.75vCore, vUncore 1.15v or less, vDIMM 1.80v or less.

 

Today, It very briefly gave me 5.625GHz, then crashed time after time and eventually wouldn't even boot using my basic 4.5GHz profile. I'll test again once it's defrosted. It's possibly corrupted 3 windows installs whilst doing it *sigh*

[Vivi]

From my experience,

 

Unstable uncore = corrupted windowses

 

Good ration between vrin and vcore = good clocks (1.61v 2.73vrin, 1.7v 2.82 vrin, 1.74v - 2.88vrin)

DRAM i normally keep at 2v, uncore volts i normally do 1.5 on ln2 it seems to help for stability about 100mhz, but my mobo ads .05v to my uncore on LN2 so just watch your DMM guys

 

Not sure if i degraded a chip yet still so random clocking them

[pro]

Don't think anything degraded mate. It was probably moisture or bad settings. Just dry iy off real well and tryagain. II've noticed on this platform things go pair shaped faster if theres moisture in first dimm slot. Start using vas

[K404]

I'll add some more insulation to the first DIMM slot, it can't hurt Thanks.

[Gamer]

One cpu here just goes better and better..

But might be just my tweaking skills

 

Retail Haswell is not good for XOC, for now.

[Massman]

From my experience,

 

Unstable uncore = corrupted windowses

 

Good ration between vrin and vcore = good clocks (1.61v 2.73vrin, 1.7v 2.82 vrin, 1.74v - 2.88vrin)

DRAM i normally keep at 2v, uncore volts i normally do 1.5 on ln2 it seems to help for stability about 100mhz, but my mobo ads .05v to my uncore on LN2 so just watch your DMM guys

 

Not sure if i degraded a chip yet still so random clocking them

 

That's a very high Vccin you're using. The official Intel guideline is "Vcore +~ 0.5v = Vccin". So for 1.65V Vcore, you would go for about 2.2V Vccin.

 

You should think about the Vccin-Vcore ratio on Haswell like the Vuncore-Vdimm ratio on Gulftown. Sure, you could run 2.2V on the memory, but a lot of Gulftown chips would degrade or die because of it.

[K404]

I heard vRIN should be 2x vCore for as long as possible? That's the info I had to hand for the launch event and i've been using that rule ever since. I was following the thought that it was the input voltage to the mobo PWM and was quite disconnected from the CPU-side..... similar idea to the 12v rail of a PSU. High input voltage to clean the signal and minimise losses, low voltage out.

 

....I didn't have any whitepapers to hand though.....

Edited June 28, 2013 by K404

[Vivi]

That's a very high Vccin you're using. The official Intel guideline is "Vcore +~ 0.5v = Vccin". So for 1.65V Vcore, you would go for about 2.2V Vccin.

 

You should think about the Vccin-Vcore ratio on Haswell like the Vuncore-Vdimm ratio on Gulftown. Sure, you could run 2.2V on the memory, but a lot of Gulftown chips would degrade or die because of it.

 

argh but it helps so much if it is run high, i'll try different combinations maybe i was wrong

[K404]

Hang on a sec..... what's default vCore on Haswell? Isn't stock vRIN 1.80v? If so, that's a difference of > 0.5v immediately.

[Xtreme Addict]

I think death also qualifies as degradation?

 

Officially I surrender. My good looking Haswell died during testing CB on low volts (1.5v), going from -80*C to -105*C. I got BSOD and that's all.

 

It was QEH6 ES reviewer sample. 1.275v 4/8 @ 5 GHz SPI 32M.

 

Bye bye chip.

 

I love you Intel :*

[chispy]

Don't think anything degraded mate. It was probably moisture or bad settings. Just dry iy off real well and tryagain. II've noticed on this platform things go pair shaped faster if theres moisture in first dimm slot. Start using vas

 

Thank you for the heads up James , that is exactly what happened to me , ram slots getting moisture ,water or condensation makes the cpu crazy.

 

Kind regards: Angelo

[l0ud_sil3nc3]

Well I know for a fact I have degraded a few cpu's some with high voltages and others with what I would consider safe voltages, the most current one was damaged with 2.5 input and 1.74 vcore while simply ramping up the multiplier via software in Windows. All this was at -130c so don't think temp should have been an issue.

 

Vdimm was 2.1v and after rebooting once it hit x59 it could never boot 5Ghz again at any uncore or voltage.

[pro]

Thank you for the heads up James , that is exactly what happened to me , ram slots getting moisture ,water or condensation makes the cpu crazy.

 

Kind regards: Angelo

 

no worries mate

 

dino

[chispy]

no worries mate

 

dino

 

:celebration:

[sin0822]

few dead here: http://www.overclock.net/t/1402493/is-she-dead-4770k/0_20

 

BTW good thread idea to think.

 

Yea my VID is 1.00v and VIN default is 1.8v, but because of turbo and vdroop on the VIN be default it will be closer to 500mv difference. But I say on Ln2 higher VIN helps hurt the FIVR efficiency and make it warmer and help CB and in general higher VIN helps a bit with stability as well. From my testing every 0.1v extra you increase VIN/VCC delta you have about 1-2W more power loss from the FIVR from the decreased duty cycle. there are sweet spots on every CPU tho. But yea VIn can hurt the CPU, and raise its temperature as well on air. All haswell stuff is a tradeoff between voltages and frequencies, that is what makes it so cool.

[Massman]

Hang on a sec..... what's default vCore on Haswell? Isn't stock vRIN 1.80v? If so, that's a difference of > 0.5v immediately.

 

Yeah. Higher voltage, lower margin? 3V Vccin is deadly dangerous (read: could insta-kill), so anything close to that is near deadly dangerous.

 

The Vccin is powering the fiVR (fully integrated VR), which powers the cores, cache, io, imc, ring. You can increase Vccin to have more current when overclocking. Intel spec says Vcore +0.6V is enough to provide all the current needed for regular benching (so not suicide shots). I didn't see much (if any) difference between 2.3V and 2.6V when running 1.8V vcore at load 4c8t. What are you guys seeing?

 

What boards are you guy killing CPUs on?

 

I think death also qualifies as degradation?

 

Officially I surrender. My good looking Haswell died during testing CB on low volts (1.5v), going from -80*C to -105*C. I got BSOD and that's all.

 

It was QEH6 ES reviewer sample. 1.275v 4/8 @ 5 GHz SPI 32M.

 

Bye bye chip.

 

I love you Intel :*

 

So how many CPUs have you killed so far by installing the OS or simply increasing the multiplier at safe voltage?

[Massman]

Heads-up from ROG site posted today. Pretty much says the same I wrote earlier in this thread (even down to the Gulftown analogy )

 

The Key Lesson: VCCIN = +0.4 VCore

 

Intel defines the VCCIN specification (called the ‘Eventual CPU input voltage’ in the ROG BIOS) in relation to CPU Vcore as follows:

• Less than 0.4V – not recommended. Instability is almost guaranteed
  
• 0.4V – ideal value
  
• 0.4-0.6V – general ‘OK’ range
  
• Above 0.6V – not recommend as long-term damage can occur
  
• Generally speaking, higher VCCIN can cause a higher CPU temperature
  

As all 5 internal power rails are pulled from the single VCCIN, below 0.4V difference is not recommended as high loading on the;input voltage will cause a voltage drop that can lead to it being lower than the internal voltages. This will cause the system to lock-up. Above the safe range can cause long-term damage due to a larger than necessary potential difference. This is the same reasoning why DDR3 voltage should not exceed 1.5V, as the CPU Uncore can be damaged.

 

Due to the small voltage difference, the Maximus VI Extreme now has 8 Steps of VCCIN Load-Line Calibration, up from 5 in the previous generation, to more accurately moderate its VCCIN according to detected loading.

 

~ http://rog.asus.com/244672013/maximus-vi/introduction-to-fully-integrated-voltage-regulators-fivr-on-maximus-vi/

 

Not sure how the default 1.0V/1.8V default ratio comes in play, but I assume what they're saying is that Vccin at 1.8V is safe even if you increase the Vcore to 1.4V (air and water cooling). Makes sense to me. Anyway, some of you guys are running double, which would be like 1.3V Vcore and 2.6V Vccin so 1.3V delta ... that can't be good.

 

We all know "extreme overclocking" is about pushing the limits and no one cares about the guidelines, but you always have to take them into account. Especially if you find yourself killing a lot of CPUs ...

[Vivi]

i've found in my cpu i really really do need 2.8 vrin if im running 1.8vcore, if the 1v delta is lost i get instability and lower clocks, im maxing out 1.8vcore because all the board can give me is 2.8vrin, im not a moron tho i have tested ALL possible combos, high vcore low vrin, high vrin low vcore, my sweetspot is 1v offset DURING LN2, not necessarily on air. So yea just saying! ln2 is a mystic creature

[Massman]

What board?

 

Did you test another board to see if it's a similar issue?

[Vivi]

What board?

 

Did you test another board to see if it's a similar issue?

 

GBT Z87X-OC. I'm measuring with DMM, also if your vrin is high it pulls your vcore up regardless of what you set it, its strange lol. Yea hey maybe my cpu is normal will try it in MPower max for MOA. maybe i can do the .5v offset thing will see how it goes.

 

for my 6375 pifast as soon as i get into the High V's like 1.8+ then when the run finished i have to hit LN2 mode when it ends otherwise my PC shutsdown instantly

[sin0822]

yea i think over 1.8v will pull the vcore up a tiny bit. GBT uses the adaptive mode for vcore which works well. Also in the GBT BIOS it says 400mv too lol, but i mean come on Intel, you are the guys who said let the extreme guys test out the effect of Uncore frequency on performance we dont' want to test it lol.

[Leeghoofd]

My retail CPU which I used for reviewing has degraded 100-150MHz, from 4500MHz fully stable at 1.2Vcore till 4350MHz now at 1.15Vcore.

 

4500MHz works fine in Wprime as long as no stress is on the IMC. Thus encoding, Cinebench, 3D11 alll crash with101 or 124 error...

 

Adjusting voltages doesn't help, seems IMC is flakey now at higher speeds ... max it was tested on was 2800MHz with 2 x 4GB dimms with voltages (ring, SA,...) on AUTO. Maybe that's where the culprit is on one particular board.

 

Temps were fine, around 80°C max measured by Realtemp. CPU acts in a similar way on 4 boards...

[Xtreme Addict]

 

 

So how many CPUs have you killed so far by installing the OS or simply increasing the multiplier at safe voltage?

 

ES died during checking CB, died... let's say in 75%. It appears to work and clock with core 0 only With more cores... doesn't enter OS... no matter voltages

 

1x retail died stock @ OS installing

 

1x retail died SS trying to push SuperPI ;] But voltages were 1,65-1.7v @ 5.2 GHz (crappy chip)

[Alex@ro]

I dont' know what you do to the poor cpu's but my experience with haswell is positive.

 

had about 6 retails,all tested with 1.5V+ on air,no problems.

 

Best retail: air 1.5V 2 cores 2001,4c8t for some vantage,fine.

 

Took it on dice for 12+ hours: voltage from 1.6-1.7 3d+spi,-67 idle,-58 huge load(intel xtu).No degradation,absolutely nothing.Cache voltage 1.65,vccin 2.15.

 

Took it back on air,it runs as day 1.It's either me i got a tough bastard accepting punishment with ease,or ...what?

[TerraRaptor]

Best retail: air 1.5V 2 cores 2001,4c8t for some vantage,fine.

not possible i guess.