Overclock.net banner

1 - 20 of 62 Posts

·
Registered
Joined
·
1,342 Posts
Discussion Starter #1
Hi Guys,

I'm seeing a lot of "1.xx is the max safe voltage for [insert CPU]" and was wondering what your experiences are with this. Have any of you ever killed or damaged a CPU due to overvolting?

A lot of the numbers for max safe volts I'm seeing seem to be personal preference with little or no facts to back it up, the range of "safe" volts i've seen goes from 1.35-1.52 which is a little broad!!

I'm no expert in this field and my overclocking 'technique' is a little odd, I always go for voltages that many would deem too high, my 8700K is at 1.47 volts and my 4790K before this was at a high voltage too (can't remember off the top of my head) and both are running fine... the 8700K has only had a couple of months of use so there's plenty of time for degradation to show but the 4790K has been going strong for years with no issues.

So please share your experiences... I'd love to see some examples of degradation especially with recent intel cpus, I'd be happy to pump 1.55 volts through this 8700K if i was relatively sure it'd last 2-3 years
 

·
Registered
Joined
·
1,374 Posts
My understanding with degradation is it depends on the Amperage you push through the CPU more than the voltage, the voltage only serves to enable a higher Amperage draw for a given workload. High amperage flooding down fine circuit traces wear them out through a process called electromigration. This process is accelerated by temperature (hence why the mantra of keeping the temps under control while overclocking) and finer manufacturing processes. Intel's latest 8700K is rated for 138A which translates to 186W of safe continuous operation at the typical 1.35V which Intel is willing to stand-by. The higher your voltage, the lower your power draw must be to reduce the amperage going through the CPU circuit. This means you either use less stressful applications on the CPU that won't force high power draws, use the CPU less often to minimize the time it spends at/beyond 138A or use a lower voltage (and potentially clockspeed).

My interpretation of Intel's 1.55V specification is that is the highest voltage allowed in a transient spike scenario. Overly high voltages can have to potential to cause damage to the CPU circuits by literally punching holes in things or arcing current across junctions/traces. In this context, the 1.35V maximum overclock voltage makes a lot of sense. Depending on the stability of the VRM, switching frequency and Vdroop allowed (which is typically .1V on most Intel CPUs by default), it doesn't take much to see that 1.35V can potentially spike to 1.45V without vdroop under load. This leaves a healthy .1V from 1.45V to the maximum limit for transient spikes which the VRMs didn't compensate for. With better mobos with more potent voltage controllers and VRMs I would imagine it would be safer to skirt the limit to 1.4V or higher.
 

·
Registered
Joined
·
2,302 Posts
Temperature is the main factor in electromigration. Voltage plays the second most important role because of higher potential energy overcoming the resistance in the circuit (current literally jumps between adjacent traces).
 

·
Registered
Joined
·
73 Posts
Disclaimer: The following experiment can merely guess the short/midterm voltage induced degradation and only the voltage induced degradation of 14 nm Intel CPUs at room temperature with air cooling. It does not reflect the combined degradation mechanism resulting from electromigration and temperature induced effects.

I'd refrain from dialing in 1.55 V as you'll most likely have to add another 10 mV or so to keep your super tight overclock stable within two weeks. Let me explain the experiment I've been doing for the past weeks before bothering you with numbers and details. I was about to start a dedicated thread to it but when I saw this one on the first page I thought that's the perfect place and timing.

So I had a few dud 7700Ks lying around which I decided to sacrifice for the sake of science :) I was really curious about the rate of degradation in new generation CPUs at crazy high voltages. One by one, I tested them to determine the amount of time (under constant high core voltage) it takes to see a noticeable change (which is 5 mV or 0.005 V) in the Vcore one needs to apply to stabilise a mediocre, rather low voltage overclock. My procedure follows as such:

First, I determined a very tight overclock and two corresponding core voltages namely Vcore_max_unstable and Vcore_min_stable. As the names suggest, there happens to be a rather sharp transition point in Vcore, below which I can repeatedly prove the overclock is unstable and above which it can repeatedly pass the stability test (that is the Vcore_min_stable) when using V-Ray benchmark which is a miraculously effective tool to roughly check the stability in a very quick fashion. Since the chips were dud and I intended to isolate the influence of other factors like current and temperature on degradation, I've set the fixed overclock speed for my experiment to 4.6 GHz. Here are the initial voltages for the following CPUs:

CPU1 (batch L709C966):
Vcore_max_unstable = 1.195 V
Vcore_min_stable = 1.200 V

CPU2 (batch L710C662):
Vcore_max_unstable = 1.175 V
Vcore_min_stable = 1.180 V

CPU3 (batch L709C966):
Vcore_max_unstable = 1.190 V
Vcore_min_stable = 1.195 V

Now it's time to apply those crazy voltages. Degrading new generation Intel chips wasn't really a distant concept to me as I'd already played with and completely degraded (complete degradation would mean the required minimum voltage gets shifted so much that it fails to boot even at its stock speed on a fresh motherboard) numerous CPUs before like several G4400s, many i3s and some i5 6400s so I already knew what to expect and that anything between 1.75-1.85 V could bring sudden death. That's why I limited myself to 1.7 V. But first I started with 1.55 V. I dialed it in, got to the Win10 desktop and simply waited. I didn't run any intentional tasks but only some background Windows processes were running. I just interrupted the experiment every 12 hours to check the stability at those low voltages I mentioned above. At first I thought it would take weeks to see any change in my stable min. Vcore but then I was surprised to witness that all 3 chips were unstable at the initially stable voltages after 36 hours. Here are the results of the 1.55 V experiment:

CPU1: new Vcore_min_stable = 1.205 V after 24 hours
CPU2: new Vcore_min_stable = 1.185 V after 36 hours
CPU3: new Vcore_min_stable = 1.200 V after 24 hours

Next up was 1.6 V. I checked the stability every 4 hours during the experiment which yielded:

CPU1: new Vcore_min_stable = 1.210 V after 8 hours
CPU2: new Vcore_min_stable = 1.190 V after 12 hours
CPU3: new Vcore_min_stable = 1.205 V after 12 hours

Then I went on to 1.65 V and checked the stability every half an hour

CPU1: new Vcore_min_stable = 1.215 V after 1 hour
CPU2: new Vcore_min_stable = 1.195 V after 1.5 hours
CPU3: new Vcore_min_stable = 1.210 V after 1 hour

Finally 1.7 V. I guess I didn't check the stability frequent enough but the outcome was still obvious

CPU1: new Vcore_min_stable = 1.220 V after 15 mins
CPU2: new Vcore_min_stable = 1.200 V after 30 mins
CPU3: new Vcore_min_stable = 1.215 V after 15 mins

Well, can we deduce anything from all that? I guess so. Just like the mean time to failure rates of silicon chips vs. temperature follow the Arrhenius curve (failure time decreases exponentially with increasing temperature), the degradation rate vs. core voltage could also be an exponential function as the activation energy for voltage induced degradation/breakdown depends on the electric field strength in the transistor. As a matter of fact, even those very roughly determined degradation rates in my experiment follow a very clear exponential trajectory as you can see below.



So I took the liberty to extrapolate it further into the lower voltages to predict the time it would take for the required min. stable Vcore to degrade by 5 mV. Of course I'm not suggesting that at 1.45 V your CPU will degrade by 5 mV after 1000 hours since there must be significant amount of uncertainty margin in this tiny experiment but at least it gives some abstract picture. Assuming the uncertainty is less than 10x I can pretty much say that you're sure to lose a very tight overclock after a year or so (10000 hours) if you set a 24/7 fixed Vcore of 1.45 but then again, it's only 0.005 Volts. Then if I assume that the amount of voltage shift degradation has also an exponential behaviour, I can predict another trajectory namely the 50 mV degradation curve. What this curve shows at 1.5 V is particularly interesting to me as it would suggest the stable Vcore to shift by 50 mV after about 20000 hours which is 2.5 years. Given that Intel insists on the absolute max. voltage rating of 1.52 V for 3 years of warranty, the prediction doesn't seem too far fetched. And I also think Intel sets the stock VID to around 50-100 mV higher than what the chip can actually remain stable. Considering the 7700K I think every single chip can operate stable below 1.2 V at a stock turbo of 4.5 GHz but the stock voltage for it seems to be around 1.25-1.27 V. Which means even if you push your chip to the limits for 2.5 years with a constant 1.5 V, Intel has to make sure it's still stable at 1.25 V.

Then what's the bottom line? To me, the bottom line is unless you're sure to upgrade your CPU within the current year, going above 1.5 V doesn't seem to be a good idea. And 1.55 V is surely a no go for long term. After all, it only takes 36 hours to lose 5 mV off of your overclock.
 

Attachments

·
OC...D
Joined
·
27,888 Posts
^^ crazy good post! I wish the +rep system was working!

A frequently overlooked contributor to CPU degradation (or a shift in the Hz/mV response curve) is Load line Transients. Under voltage "clamped" conditions (eg, fixed vcore or vcore at load with dynamic voltage and frequency) when the current (read "load") changes there is an unavoidable oscillation of the voltage in that circuit - called load line overshoot (and undershoot). These transient spikes occur on the microSecond scale and are referred to as V_ovs by Intel in processor spec sheets. They are not something you can detect with a DMM (or OS tool), but require specific equipment, and ideally an Intel socket-specific tool. Anyway, MB mfrs added LLC to "compensate" for the load-line overshoot, but all LLC really does is allow for (or not) sufficient vdroop to give some headroom for the V_ovs with regard to the bios specified vcore/vccin delivery. It does not change the magnitude of the overshoot, just lowers the peak spike (by the amount of vdroop). So, long story short, defeating vdroop completely with LLC (eg... "holding a constant vcore idle vs load as measured by CPUZ for example) actually allows for some serious uSec voltage excursions well above the bios setting(s), leading to what we call degradation (or accelerated "aging" of gates/transistors). a uSec is a long time from a transistor's point of view. So, Vdroop is a good thing.

Depending on the vdroop allowed in the above experiment, the actual peak voltage is likely much higher that the set vcore, eg,. 1.7V (by more than 200mV during load transitions).
 

·
9 Cans of Ravioli
Joined
·
19,873 Posts
Had a Sempron 140 ages ago I was trying to get >5Ghz with and killed it with 1.85v.

Other than that, never killed a CPU from voltage. Ran my i5 6400 at 1.475v (set in BIOS) for a few benches and it still made it 2+ years before I retired it.
 

·
OC...D
Joined
·
27,888 Posts
yeah - killed is different from degraded... guess it is the ultimate degradation. :p
 

·
Banned
Joined
·
3,726 Posts
^^ crazy good post! I wish the +rep system was working!

Depending on the vdroop allowed in the above experiment, the actual peak voltage is likely much higher that the set vcore, eg,. 1.7V (by more than 200mV during load transitions).
well I have a 3930K that I set the voltage in BIOS to 1.73 volts for a 4Ghz clock on the CPU. the CPU has been running this way for 6 years now. the top voltage that the CPU hits at load is 2.03 volts, now I have been told that this is to high, but my CPU is very stable so it works for me. now on the down side is that I have had to replace my memory every year as the memory keeps failing on me.
 

·
Registered
Joined
·
173 Posts
well I have a 3930K that I set the voltage in BIOS to 1.73 volts for a 4Ghz clock on the CPU. the CPU has been running this way for 6 years now. the top voltage that the CPU hits at load is 2.03 volts, now I have been told that this is to high, but my CPU is very stable so it works for me. now on the down side is that I have had to replace my memory every year as the memory keeps failing on me.
Uhhh...id love a pic of that bios screen? 6 years at 1.7 volts on a 3930k seems...insane. UNless you are on phase change cooling?
 

·
Registered
Joined
·
42 Posts
very good info in this post, I have been wondering the same thing myself, I almost wanted to run my 8700K at 1.45-1.5V just to see how long it would last, I decided against it though. Maybe a year from now I will get crazy with this chip and really push it, but for now I'm happy with it. I'm running 1.375v @5GHZ 23ish hours a day minning on this computer for several months now.
 

·
OC...D
Joined
·
27,888 Posts
well I have a 3930K that I set the voltage in BIOS to 1.73 volts for a 4Ghz clock on the CPU. the CPU has been running this way for 6 years now. the top voltage that the CPU hits at load is 2.03 volts, now I have been told that this is to high, but my CPU is very stable so it works for me. now on the down side is that I have had to replace my memory every year as the memory keeps failing on me.
:tiredsmil
 

Attachments

·
TEC Veteran
Joined
·
2,007 Posts
My first 4790k was ran at 4.9ghz w/ 1.37v and showed zero degradation signs after a year and a half of use. My current 4790k has degraded by ~200mhz in the 3 years I have had it. Was ran at 5ghz w/1.4v 24/7, with sub ambient temps on a TEC setup (avg load temps in the 40s). Now I run it at 4.7ghz since 4.8-5ghz is no longer stable on feasible voltages. I usually only keep CPUs for ~2yrs, so not a big deal for me. It's past due for an upgrade, I'm just waiting to decide on Ryzen2 or wait for the TR refresh. To be fair, it survived a pin rotted motherboard and has been through a lot of scares, so I'm just happy it still works :laughings.

Before those, I had a 3770k that lost roughly 150mhz in 2 years. That was one a turd though and only ran at 4.6ghz...any higher needed TONS of voltage.

Every CPU and setup will be different but depending on how hard you are running these chips, I would expect some slight degradation after a year or more. Don't be surprised if you have to drop your clocks 100mhz or so, or give another click more vcore to compensate.

All of my chips for the last 6-7 years have been on TECs though, so YMMV.
 

·
Banned
Joined
·
3,726 Posts
Uhhh...id love a pic of that bios screen? 6 years at 1.7 volts on a 3930k seems...insane. UNless you are on phase change cooling?


CPUVCORE is 1.728 to start but goes all the way up to 2.014 at load, still runs the same way today.
 

·
OC...D
Joined
·
27,888 Posts
lol - and for a frequency of 4GHZ? You should send that to Intel so they can study that anomaly. ;)

Peace. Out.
 

·
Banned
Joined
·
3,726 Posts
yea I get that it is not the norm, but it's stable, and not fried, so I leave it the way it is. it does run hotter than the normal 3930K CPU with a 4Ghz clock on it. maybe when I am done with it I will, that is not a bad idea.
 

·
OC...D
Joined
·
27,888 Posts
yea I get that it is not the norm, but it's stable, and not fried, so I leave it the way it is. it does run hotter than the normal 3930K CPU with a 4Ghz clock on it. maybe when I am done with it I will, that is not a bad idea.
yeah, bottom line is current kills, not voltage. So, load and frequency play an important part. :)
 

·
Facepalm
Joined
·
8,614 Posts
^^ crazy good post! I wish the +rep system was working!

A frequently overlooked contributor to CPU degradation (or a shift in the Hz/mV response curve) is Load line Transients. Under voltage "clamped" conditions (eg, fixed vcore or vcore at load with dynamic voltage and frequency) when the current (read "load") changes there is an unavoidable oscillation of the voltage in that circuit - called load line overshoot (and undershoot). These transient spikes occur on the microSecond scale and are referred to as V_ovs by Intel in processor spec sheets. They are not something you can detect with a DMM (or OS tool), but require specific equipment, and ideally an Intel socket-specific tool. Anyway, MB mfrs added LLC to "compensate" for the load-line overshoot, but all LLC really does is allow for (or not) sufficient vdroop to give some headroom for the V_ovs with regard to the bios specified vcore/vccin delivery. It does not change the magnitude of the overshoot, just lowers the peak spike (by the amount of vdroop). So, long story short, defeating vdroop completely with LLC (eg... "holding a constant vcore idle vs load as measured by CPUZ for example) actually allows for some serious uSec voltage excursions well above the bios setting(s), leading to what we call degradation (or accelerated "aging" of gates/transistors). a uSec is a long time from a transistor's point of view. So, Vdroop is a good thing.

Depending on the vdroop allowed in the above experiment, the actual peak voltage is likely much higher that the set vcore, eg,. 1.7V (by more than 200mV during load transitions).

Wow....
this basically proves Raja's point that "Vdroop is a good thing."
Or maybe, SOME vdroop is a good thing.
I mean, if a chip required 1.4v for full stability at load, and you turn off all Loadline Calibration, you're going to be putting 1.55v IDLE into that thing to get 1.4v stable at load. And there's your degradation at very low temps and current.

Seems like it's best to use a small amount of LLC, so you have something like 1.42v idle, 1.40v load, for best of both worlds. any spikes you get from overshoot are going to have the cpu exposed to less high vcore than 1.55v idle...

Either way, vRISE at load 'seems' to be a good thing when you look at it, but then you have 1) voltage spikes can now actually get damaging, 2) idle BSOD problems.

Is it possible for you to test something on your test chips, like having idle and load voltage the same ? (like 1.4v idle, 1.4v load) and then check for degradation by rapidly shifting load cycles (doesn't have to be power viruses either).
 

·
Facepalm
Joined
·
8,614 Posts
My first 4790k was ran at 4.9ghz w/ 1.37v and showed zero degradation signs after a year and a half of use. My current 4790k has degraded by ~200mhz in the 3 years I have had it. Was ran at 5ghz w/1.4v 24/7, with sub ambient temps on a TEC setup (avg load temps in the 40s). Now I run it at 4.7ghz since 4.8-5ghz is no longer stable on feasible voltages. I usually only keep CPUs for ~2yrs, so not a big deal for me. It's past due for an upgrade, I'm just waiting to decide on Ryzen2 or wait for the TR refresh. To be fair, it survived a pin rotted motherboard and has been through a lot of scares, so I'm just happy it still works :laughings.

Before those, I had a 3770k that lost roughly 150mhz in 2 years. That was one a turd though and only ran at 4.6ghz...any higher needed TONS of voltage.

Every CPU and setup will be different but depending on how hard you are running these chips, I would expect some slight degradation after a year or more. Don't be surprised if you have to drop your clocks 100mhz or so, or give another click more vcore to compensate.

All of my chips for the last 6-7 years have been on TECs though, so YMMV.
May be a relationship between HIGH VID and LOW VID chips (automatic voltages/SVID).
Something about "leaky" vs "non leaky" chips?
and lower VID chips running cooler, better overclocking on air, but responding worse to subzero, while high VID chips run hotter, have higher power consumption, but react better when kept cold?
 

·
Registered
Joined
·
1,342 Posts
Discussion Starter #20
Disclaimer: The following experiment can merely guess the short/midterm voltage induced degradation and only the voltage induced degradation of 14 nm Intel CPUs at room temperature with air cooling. It does not reflect the combined degradation mechanism resulting from electromigration and temperature induced effects.

I'd refrain from dialing in 1.55 V as you'll most likely have to add another 10 mV or so to keep your super tight overclock stable within two weeks. Let me explain the experiment I've been doing for the past weeks before bothering you with numbers and details. I was about to start a dedicated thread to it but when I saw this one on the first page I thought that's the perfect place and timing.

So I had a few dud 7700Ks lying around which I decided to sacrifice for the sake of science :) I was really curious about the rate of degradation in new generation CPUs at crazy high voltages. One by one, I tested them to determine the amount of time (under constant high core voltage) it takes to see a noticeable change (which is 5 mV or 0.005 V) in the Vcore one needs to apply to stabilise a mediocre, rather low voltage overclock. My procedure follows as such:

First, I determined a very tight overclock and two corresponding core voltages namely Vcore_max_unstable and Vcore_min_stable. As the names suggest, there happens to be a rather sharp transition point in Vcore, below which I can repeatedly prove the overclock is unstable and above which it can repeatedly pass the stability test (that is the Vcore_min_stable) when using V-Ray benchmark which is a miraculously effective tool to roughly check the stability in a very quick fashion. Since the chips were dud and I intended to isolate the influence of other factors like current and temperature on degradation, I've set the fixed overclock speed for my experiment to 4.6 GHz. Here are the initial voltages for the following CPUs:

CPU1 (batch L709C966):
Vcore_max_unstable = 1.195 V
Vcore_min_stable = 1.200 V

CPU2 (batch L710C662):
Vcore_max_unstable = 1.175 V
Vcore_min_stable = 1.180 V

CPU3 (batch L709C966):
Vcore_max_unstable = 1.190 V
Vcore_min_stable = 1.195 V

Now it's time to apply those crazy voltages. Degrading new generation Intel chips wasn't really a distant concept to me as I'd already played with and completely degraded (complete degradation would mean the required minimum voltage gets shifted so much that it fails to boot even at its stock speed on a fresh motherboard) numerous CPUs before like several G4400s, many i3s and some i5 6400s so I already knew what to expect and that anything between 1.75-1.85 V could bring sudden death. That's why I limited myself to 1.7 V. But first I started with 1.55 V. I dialed it in, got to the Win10 desktop and simply waited. I didn't run any intentional tasks but only some background Windows processes were running. I just interrupted the experiment every 12 hours to check the stability at those low voltages I mentioned above. At first I thought it would take weeks to see any change in my stable min. Vcore but then I was surprised to witness that all 3 chips were unstable at the initially stable voltages after 36 hours. Here are the results of the 1.55 V experiment:

CPU1: new Vcore_min_stable = 1.205 V after 24 hours
CPU2: new Vcore_min_stable = 1.185 V after 36 hours
CPU3: new Vcore_min_stable = 1.200 V after 24 hours

Next up was 1.6 V. I checked the stability every 4 hours during the experiment which yielded:

CPU1: new Vcore_min_stable = 1.210 V after 8 hours
CPU2: new Vcore_min_stable = 1.190 V after 12 hours
CPU3: new Vcore_min_stable = 1.205 V after 12 hours

Then I went on to 1.65 V and checked the stability every half an hour

CPU1: new Vcore_min_stable = 1.215 V after 1 hour
CPU2: new Vcore_min_stable = 1.195 V after 1.5 hours
CPU3: new Vcore_min_stable = 1.210 V after 1 hour

Finally 1.7 V. I guess I didn't check the stability frequent enough but the outcome was still obvious

CPU1: new Vcore_min_stable = 1.220 V after 15 mins
CPU2: new Vcore_min_stable = 1.200 V after 30 mins
CPU3: new Vcore_min_stable = 1.215 V after 15 mins

Well, can we deduce anything from all that? I guess so. Just like the mean time to failure rates of silicon chips vs. temperature follow the Arrhenius curve (failure time decreases exponentially with increasing temperature), the degradation rate vs. core voltage could also be an exponential function as the activation energy for voltage induced degradation/breakdown depends on the electric field strength in the transistor. As a matter of fact, even those very roughly determined degradation rates in my experiment follow a very clear exponential trajectory as you can see below.



So I took the liberty to extrapolate it further into the lower voltages to predict the time it would take for the required min. stable Vcore to degrade by 5 mV. Of course I'm not suggesting that at 1.45 V your CPU will degrade by 5 mV after 1000 hours since there must be significant amount of uncertainty margin in this tiny experiment but at least it gives some abstract picture. Assuming the uncertainty is less than 10x I can pretty much say that you're sure to lose a very tight overclock after a year or so (10000 hours) if you set a 24/7 fixed Vcore of 1.45 but then again, it's only 0.005 Volts. Then if I assume that the amount of voltage shift degradation has also an exponential behaviour, I can predict another trajectory namely the 50 mV degradation curve. What this curve shows at 1.5 V is particularly interesting to me as it would suggest the stable Vcore to shift by 50 mV after about 20000 hours which is 2.5 years. Given that Intel insists on the absolute max. voltage rating of 1.52 V for 3 years of warranty, the prediction doesn't seem too far fetched. And I also think Intel sets the stock VID to around 50-100 mV higher than what the chip can actually remain stable. Considering the 7700K I think every single chip can operate stable below 1.2 V at a stock turbo of 4.5 GHz but the stock voltage for it seems to be around 1.25-1.27 V. Which means even if you push your chip to the limits for 2.5 years with a constant 1.5 V, Intel has to make sure it's still stable at 1.25 V.

Then what's the bottom line? To me, the bottom line is unless you're sure to upgrade your CPU within the current year, going above 1.5 V doesn't seem to be a good idea. And 1.55 V is surely a no go for long term. After all, it only takes 36 hours to lose 5 mV off of your overclock.
Thank you so much... that is exactly the sort of info i was looking for, fantastic post!!!

If i knew my vcore max unstable/vcore min stable of my 4790K i'd test this myself but i don't, i can only guess that some degradation has happened by now.

You're the first person to ever convince me to lower voltages, now I'm at 1.37v for 5Ghz on my 8700K... my CPU sends it's thanks :)
 
1 - 20 of 62 Posts
Top