Overclock.net banner
1 - 18 of 18 Posts

RedKnight7

· Registered
Joined
·
61 Posts
Discussion starter · #1 ·
Hi folks,

My 4670K is poor for overclocking; it needs 1.431 VID just to do 4.4 Ghz (at 80 C in the x264 v2 stress test). If I push it any faster, it flirts with 100 C.

There is strong evidence that the Haswell cache (a.k.a. uncore) can be considerably slower than the cores with little ill effect. As a result, many run their cache slow (sometimes even 34/35) because it might need high volts to match OC core speed. The reasoning is, of course, "if cache speed doesn't matter much, why push it?"

But I'd like to explore the flip side of that. Why not push it? What exactly hurts a CPU and/or cache?

My poor CPU hits 80 C when stressed with the cores at 44 and 1.431 VID. At idle, this runs at 50 C if the cache is sidelined (speed 34, Auto volts ~1.094 in my rig in sig). If I match Uncore speed to Core, the cache needs 1.392 V and the rig runs at 53 on idle, but still 80 when stressed. Yes, it makes almost no difference to the X264 v2 stress test - but then, logically I suppose, also doesn't affect temps at all.

The bottom line is that, while fast 1:1 cache may not help much, who's to say it's hurting much? If the only difference is +3 C at idle (but not hotter when stressed), who cares? It's not like I'm worried about extra electricity charges for 1.4 V versus 1.0 V uncore.
thumb.gif


What exactly is the evidence the high voltages damage over time? Or is it actually high temperatures (and "high volts" is just a proxy for that)? 50 versus 53 C is hardly worth writing home about.

Note that I am looking for solid evidence here, if anyone has some or can point to some. Or *lots* of personal experience (you run banks of PCs at work). We all have opinions and guesses.

What solid evidence is there that 1.4 volts will hurt? Looking for something more than "I burned out one CPU", please.

Please don't reply "why push it". The point of this post is, what's the evidence to not push it.

Thanks
 
Voltage is the killer. CPUs and GPUs today reach anywhere from 80 to high 90s at stock clocks and on stock cooling, and will usually start throttling back at around 90-100c (with the exception of Nvidia-gpus which are set not to go higher than 80).

Edit:
I cannot comment on what is "safe" for Haswell 24/7, as I do not have any experience with those. I can tell you though, that I ran 1,38v for quite some time, before I dropped the clock about 200mhz and settled for 1,32v on my 2500k, which has been running like that for two years now.
 
From personal experience, push it until it dies....

I have tried setting my Vid to 1.92v and down clock from 50x to see which one boots then which one will run a stress test.

If it is a clear cut case of beyond the operating limits of the CPU, it shuts down cold. The danger lies where the CPU runs and is stable BUT you are in the range of accelerated damage to CPU from electron degradation. So far I can't find any definitive literature on this part.

My OC method is the non-traditional way. Set everything to max to see how far it goes then down scale to find something stable. Do not try if you do not have adequate cooling measures.
 
Quote:
Originally Posted by RedKnight7 View Post

What solid evidence is there that 1.4 volts will hurt? Looking for something more than "I burned out one CPU", please.
From the Haswell Death/Degradation thread:

CPU: i7-4770k

DELIDED: YES

REPORTED CONDITION: DEAD, R.I.P., MUERTO, KAPUTT

CAUSE OF DEATH: EXCESSIVE VOLTAGE, 1.37-1.4+V VCORE

HOW LONG OVERCLOCKED: Approx one (1) year

OP DESCRIPTION: I had ran a daily overclock of 4.4, but that wasn't the problem. My voltages were off the charts. My vcore was set on adaptive instead of manual and it would range anywhere from 1.375 to 1.4+ My input voltage was 1.9 and my uncore voltage was somewhere up there. I don't remember. I had this chip delided and OC'd for almost a year.

THREAD: http://www.overclock.net/t/1475351/i-think-my-4770k-gave-out

WARNING: HardOCP.COM state that anything over 1.35V VCORE 24/7 will degrade an Haswell pretty fast when conjoined with heavy loads and lots of working hours every day.



http://www.hardocp.com/article/2013/08/21/asus_maximus_vi_hero_lga_1150_motherboard_review/7
 
I actually had a similar question only in regard to GPU memory a few days ago.

From what I've experienced and read, I think they are both detrimental. They are somewhat related in that higher voltage usually causes higher temperature. I think this can make it difficult to identify exactly what causes any observed degradation / damage.

I do know of cases where too much voltage causes significant damage / degradation while temperatures are perfectly fine though.

Take Sandy Bridge-E chips for example. I've read quite a few reports of people experiencing significant degradation due to VTT and VCCSA voltages being set too high. Given that most people who purchase these enthusiast chips use pretty robust cooling systems (closed loop liquid cooling solutions or custom loops) and the relatively small impact VTT and VCCSA have on temperature, I think this is evidence that voltage alone can be a problem.

Also, given that Intel / AMD usually list maximum safe operating temperature as well as voltage would lead me to believe they can both cause problems independent of the other.

Hopefully somebody more knowledgeable comes along with a detailed explanation. I'd like a definitive answer to this as well.
 
It's not an either or thing, nor are voltage and temperature the only factors.

Voltage, temperature, and current all play major roles in electromigration damage. Thermal cycling and other mechanical stresses can also result in failure.
 
Quote:
Originally Posted by Arxontas View Post

From the Haswell Death/Degradation thread:

CPU: i7-4770k

DELIDED: YES

REPORTED CONDITION: DEAD, R.I.P., MUERTO, KAPUTT

CAUSE OF DEATH: EXCESSIVE VOLTAGE, 1.37-1.4+V VCORE

HOW LONG OVERCLOCKED: Approx one (1) year

OP DESCRIPTION: I had ran a daily overclock of 4.4, but that wasn't the problem. My voltages were off the charts. My vcore was set on adaptive instead of manual and it would range anywhere from 1.375 to 1.4+ My input voltage was 1.9 and my uncore voltage was somewhere up there. I don't remember. I had this chip delided and OC'd for almost a year.

THREAD: http://www.overclock.net/t/1475351/i-think-my-4770k-gave-out

WARNING: HardOCP.COM state that anything over 1.35V VCORE 24/7 will degrade an Haswell pretty fast when conjoined with heavy loads and lots of working hours every day.



http://www.hardocp.com/article/2013/08/21/asus_maximus_vi_hero_lga_1150_motherboard_review/7
Keep in mind that is only one report. I would take that with a grain of salt. If we see more reports like that I would start to take notice.

Ivy bridge (22nm) could handle 1.4 vcore easily (combined with good temps), and 1.9v Input Voltage is not high at all.
 
Quote:
Originally Posted by Anth Seebel View Post

Keep in mind that is only one report. I would take that with a grain of salt. If we see more reports like that I would start to take notice.

Ivy bridge (22nm) could handle 1.4 vcore easily (combined with good temps), and 1.9v Input Voltage is not high at all.
Alright, so i guess i will keep it running 24/7 at 1.405V. My max temp in the IBT is 74C, so i think i am good
thumb.gif
 
Quote:
Originally Posted by Anth Seebel View Post

Keep in mind that is only one report. I would take that with a grain of salt. If we see more reports like that I would start to take notice.

Ivy bridge (22nm) could handle 1.4 vcore easily (combined with good temps), and 1.9v Input Voltage is not high at all.
HardOCP.com have noted and documented degradation in their chip @1.35V.

Since these guys know very well what they say and write, I don't think we need any more reports. In fact, I don't think that Haswell degradation at or above 1.35V VCore is up for debate.
 
Quote:
Originally Posted by Arxontas View Post

HardOCP.com have noted and documented degradation in their chip @1.35V.

Since these guys know very well what they say and write, I don't think we need any more reports. In fact, I don't think that Haswell degradation at or above 1.35V VCore is up for debate.
I dont think you can say 1.35v will degrade in every application though. Your statement is too general. If I set my vcore to 1.45v and spend 99% or my time browsing the web and gaming 3hours a week my chip is obviously going to degrade slower than someone who runs prime95 60 hours a week( no idea why, just an example).
\
Whith cstates enabled the chip is spending 90% or the time at a lower vcore. I think every application is going to be different as well as each chip is different.
 
I have been running Haswell since launch of 2013.

I killed my first 4770K due to excessive voltage 1.8V. 1.8V or higher will most likely lead to death. I was only using watercooling. Thankfully intel accepted my cpu as per RMA and sent me a new one:)

in the summer I run 1.38V @ 4.6GHz

in the winter I run 1.45V @ 4.7GHz

I personally feel under 1.45V is acceptable for 24\7 Use if you can cool the beast.

I have benched both 4770k and 4670k at over 1.7V, no problems. both chips still kicking and I really dont think I degraded them at all.
 
Discussion starter · #14 ·
I'm only hearing isolated instances of examples for and against. No one is citing numbers in depth.

It's also been said in Darkwizzie's Guide and elsewhere that in point of fact, there might not be any good data (big numbers) out there. Who wants to test a lot of chips to death just to find exactly where they die? Further, we already know there is serious variation in OC ability. There's undoubtedly variation in longevity versus conditions, too.

On the average, my PC is off/standby two-thirds of the time, and practically idle all the rest. So my issue is whether volts but not temps matter, since I never run hot IRL. I don't need opinions; everyone has one. Can anyone point to studies or metaanalyses on at least 10 CPUs, preferably 100+? At what VIDs or other volts did the CPUs die?

I'm just asking. That's all. If there aren't hard numbers, there aren't.

Thanks ya'll
smile.gif
- RK7
 
Quote:
Originally Posted by RedKnight7 View Post

Can anyone point to studies or meta-analyses on at least 10 CPUs, preferably 100+? At what VIDs or other volts did the CPUs die?

I'm just asking. That's all. If there aren't hard numbers, there aren't.
I think you will be hard pressed to find this type of raw data compiled already. Haswell is still very much in its infancy, even tho it is widespread globally.

One may have to look at individual cases and chart the data themselves at this point of time.

Jut re-read your comment on ring voltage.

Currently I run my i7 ring voltage at 1.45V. On my i5 I run it at 1.4V. Both cache ratios are set for 4400MHz..

Been running this type of ring voltage for almost a year with no signs of degradation or negative effects at this point.
 
4770k 4.4 at 1.28 daily.. I won't go any higher than 1.3v personally. This thing is so fast as it is. I'm delided and max temp rendering or mining is about 55c.
I do a lot of video renders for youtube and solidworks. and my pc is on 24/7

1.5v is the most I'd ever push on a full water loop and that would be just for benching. I'm on a corsair H110, my chip will do 4.6 at 1.42 but I never tested for daily use cus the volts. I also didn't test any higher clocks after that. but it finishes all my benches in a row and 10x intel burn tests at 4.6
 
My current 4670K runs 4.6ghz core and 4.4 cache with 1.267 and 1.15 respectively. While this may be a pretty good chip by Haswell standards, I can't get it stable at 4.7 with as much as 1.4 into it. It does seem to be that as the circuits get smaller, they also become much more susceptible to decay.

Quote:
Originally Posted by ProKoN View Post

I personally feel under 1.45V is acceptable for 24\7 Use if you can cool the beast.
^ I agree with this. I ran my old i7-950 at 1.5 volts and 4.4ghz from 2011 until i got this haswell thing in mid 2013. That thing ran 24/7 folding at night and gaming/encoding during the day.
 
Discussion starter · #18 ·
Thanks for contributing everyone, whether you thought 1.4 V was high or low.

To me it seems like the conclusion is that ... there isn't really a conclusion. Yet.

But it's pretty clear one is at least in a grey area. So if you want to be cautious, do!

I'll probably keep my cache at 1.392 V so it can keep up with my Cores at 44.

Maybe I'll end up on the Haswell Death/Degradation thread.

Let's find out.
smile.gif
 
1 - 18 of 18 Posts