VCore decreases under load

[PersonWithTech]

So I've decided to find the max safe clock of my CPU for benchmark purposes, and I am currently at 4.7GHz (is that good for a 3570K?). Unfortunately, 1.4V of Fixed voltage didn't get me there so I've ramped it up to 1.5V. I am noticing, however, that the actual vcore goes down to 1.44V on load from 1.48V when idle. Why is this happening? Frequency remains static (exc. after decimal point Mhz fluctuation) and performance is an improvement, but why is my VCore going down. I do have turbo boost enabled (as I am forced to) and the "additional turbo voltage" is the minimum of +0.004V . I don't think this is throttling or SpeedStep as those just reduce the frequency. What is it? How do I get a higher, more fixed voltage?

 

[FX-GMC]

So I've decided to find the max safe clock of my CPU for benchmark purposes, and I am currently at 4.7GHz (is that good for a 3570K?). Unfortunately, 1.4V of Fixed voltage didn't get me there so I've ramped it up to 1.5V. I am noticing, however, that the actual vcore goes down to 1.44V on load from 1.48V when idle. Why is this happening? Frequency remains static (exc. after decimal point Mhz fluctuation) and performance is an improvement, but why is my VCore going down. I do have turbo boost enabled (as I am forced to) and the "additional turbo voltage" is the minimum of +0.004V . I don't think this is throttling or SpeedStep as those just reduce the frequency. What is it? How do I get a higher, more fixed voltage?

Load Line Calibration.

.04 isn't much of a drop though. I always see a slight bit of variance in my core voltage.

 

[Jack1n]

Thats really high voltage,you should keep it under 1.4v if you want the cpu to last.

 

[PersonWithTech]

Load Line Calibration.

.04 isn't much of a drop though. I always see a slight bit of variance in my core voltage.

I suspected it was LLC. I just don't understand the technology. Why would you want LESS voltage under load? Shouldn't it be vice versa.

Thats really high voltage,you should keep it under 1.4v if you want the cpu to last.

I know it is high. I was only using it at that voltage to get me to 4.7GHz for benchmarking. For general use, I run 4.5GHz at 1.3ish v

 

[Aquinus]

You're noticing Vdroop and LLC which is to protect your CPU from transient current/voltage when a CPU switches between low power (idle) and high load. Wikipedia does a half decent job describing why this is employed and what it's supposed to do.

Voltage droop is the intentional loss in output voltage from a device as it drives a load. Employing droop in a voltage regulation circuit increases the headroom for load transients.
...
In a regulator not employing droop, when the load is suddenly increased very rapidly (i.e. a transient), the output voltage will momentarily sag. Conversely, when a heavy load is suddenly disconnected, the voltage will show a peak. The output decoupling capacitors have to "absorb" these transients before the loop has a chance to compensate...

...and from Intel.

2.2 Load Line Definitions – REQUIRED

To ensure processor reliability and performance, platform DC and AC transient voltage regulation must be contained within the VCCMIN and the VCCMAX die load line boundaries, except for short burst transients above the VCCMAX as specified in Section 2.4. Die load line compliance must be guaranteed across 3-sigma component manufacturing tolerances, thermal variation and age degradation. The following load line contains static and transient voltage regulation data as well as maximum and minimum voltage levels. It is required that the regulator’s positive and negative differential remote sense pins be connected to both the VCC_SENSE, VSS_SENSE pin pair of the processor socket, see Figure 5-1. The prefix VCC is designated for the positive remote sense signal and the VSS prefix for the negative remote sense signal.

The upper and lower load lines represent the allowable range of voltages that must be presented to the processor. The voltage must always stay within these boundaries for proper operation of the processor. Operating above the VCCMAX load line limit will result in higher processor operating temperature, which may result in damage or a reduced processor lifespan. Processor temperature rise from higher functional voltages may lead to dynamic operation to low power states, which directly reduces processor performance. Operating below the VCCMIN load line limit will result in minimum voltage violations, which will result in reduced processor performance, system lock up, “blue screens” or data corruption.

Source

...and an example for skt1366 Xeons.

 

[PersonWithTech]

You're noticing Vdroop and LLC which is to protect your CPU from transient current/voltage when a CPU switches between low power (idle) and high load. Wikipedia does a half decent job describing why this is employed and what it's supposed to do.



...and from Intel.

Source

...and an example for skt1366 Xeons.
View attachment 57173

So, in a nutshell: LLC = CPU Vdroop = Protection from oscillation

My guess is the transients are more violent with higher load?

Also, to make things more annoying and complicated, how does LLC function with offset voltage (considering voltage initially increases under load when voltage is set in offset)

 

[Aquinus]

So, in a nutshell: LLC = CPU Vdroop = Protection from oscillation

My guess is the transients are more violent with higher load?

Also, to make things more annoying and complicated, how does LLC function with offset voltage (considering voltage initially increases under load when voltage is set in offset)

Yes and no.
Yes: LLC is Vdroop, Vdroop is to reduce the max voltage experienced by the device when transience is encountered.
No: Higher load doesn't mean more transience, however the bigger change between idle draw and loaded draw can impact it depending on how quickly the VRMs respond to changes in load. A little more information about why transience arise will clarify this.

Transient response definition is stated as:

In electrical engineering and mechanical engineering, a transient response or natural response is the response of a system to a change from equilibrium.

Transience occurs particularly when the VRM hasn't sensed that the circuit is demanding less power (or more) but is still producing output for a particular current draw. So what happens is that since I drops, V increases proportionally since switched mode PSUs pull voltage up using a voltage higher than what you want and also the opposite when you're going from low load to high load. As I increases, VRMs don't keep up right away so there is a voltage drop before the VRMs have time to respond.

 

[brandonwh64]

1.4V will kill it. I would just back it down to a safer voltage and clocks

 

[RCoon]

I've ramped it up to 1.5V

You're crazy. You're not OC'ing an FX CPU, tone that voltage down. That's extreme cooling voltages, not the 4.7Ghz area.

 

[Aquinus]

1.4V will kill it. I would just back it down to a safer voltage and clocks

You're crazy. You're not OC'ing an FX CPU, tone that voltage down. That's extreme cooling voltages, not the 4.7Ghz area.

I'm with these two. A 22nm CPU doesn't need that much voltage and on air that's a lot of heat to move. 1.4v on a SB or SB-E isn't unreasonable and 1.5v on an FX or Phenom II isn't unreasonable either, but you're asking to damage that CPU but putting that much volts through it. IVB-E doesn't clock like a champ and when it doesn't, raising the voltage isn't always the answer. Some CPUs OC better than others and even if you do get higher clocks from the extra voltage, you're probably going to degrade your CPU so there will come a point when either the CPU will stop working or you'll never reach those kinds of clocks again and it will take more voltage to maintain lower clocks.

So in summary: Pull down those volts before you irreversibly damage something.

 

[PersonWithTech]

Yes and no.
Yes: LLC is Vdroop, Vdroop is to reduce the max voltage experienced by the device when transience is encountered.
No: Higher load doesn't mean more transience, however the bigger change between idle draw and loaded draw can impact it depending on how quickly the VRMs respond to changes in load. A little more information about why transience arise will clarify this.

Transient response definition is stated as:

Transience occurs particularly when the VRM hasn't sensed that the circuit is demanding less power (or more) but is still producing output for a particular current draw. So what happens is that since I drops, V increases proportionally since switched mode PSUs pull voltage up using a voltage higher than what you want and also the opposite when you're going from low load to high load. As I increases, VRMs don't keep up right away so there is a voltage drop before the VRMs have time to respond.

So transients occur as a result of a lag between voltage/load change. Ahhh.

I'm with these two. A 22nm CPU doesn't need that much voltage and on air that's a lot of heat to move. 1.4v on a SB or SB-E isn't unreasonable and 1.5v on an FX or Phenom II isn't unreasonable either, but you're asking to damage that CPU but putting that much volts through it. IVB-E doesn't clock like a champ and when it doesn't, raising the voltage isn't always the answer. Some CPUs OC better than others and even if you do get higher clocks from the extra voltage, you're probably going to degrade your CPU so there will come a point when either the CPU will stop working or you'll never reach those kinds of clocks again and it will take more voltage to maintain lower clocks.

So in summary: Pull down those volts before you irreversibly damage something.

Thanks. You will all be releived to know that I don't run it at 1.5V/4.7GHz full time. I was simply trying to find the max OC. I am actually run it at 4.5GHz with 1.3v (offset) full-time. Is there anyway I can achieve the highest clock without giving it a deadly voltage? I know 4.5GHz is the most common full time OC on a 3570K, but I think I've also seen quite a few 4.6GHz OCs on the 3770K
How does the i7 manage to have better overclocking performance? Isn't it essentially a hyperthreaded i5?

Also, my dad runs a 2500K at 4.6GHz and he seems to still beat me in cinebench. Would it have been smarter to have gotten the 2500K? Was the 3570K a dumb option for overclocker/gamers?

P.S. Just be thankful I'm not this guy : http://www.tomshardware.com/forum/357916-28-3570k-highest-overclock-stock-cooler

 

[Lopez0101]

Hey man, my 4770k can totally run 4.1Ghz with the stock cooler. Even stays below 95C! Though I went back to stock while I wait for my heatsink to get here.

 

[PersonWithTech]

Hey man, my 4770k can totally run 4.1Ghz with the stock cooler. Even stays below 95C! Though I went back to stock while I wait for my heatsink to get here.

Yes, but this guy is running is running 4.6 with 65C on IDLE! and thinks he can do 5GHz. He is wasting a good 3570k

 

[Lopez0101]

I know, wasn't saying he wasn't stupid. I was making a joke not to doubt the stock cooler...

 

[PersonWithTech]

I know, wasn't saying he wasn't stupid. I was making a joke not to doubt the stock cooler...

My dad used to OC his 2500k to 4GHz ocassionally. But those temps are a bit high (90C under load). Yeah. I just linked that to prove that I'm not the most retarded when it comes to overclocking.

 

[buildzoid]

I'm with these two. A 22nm CPU doesn't need that much voltage and on air that's a lot of heat to move. 1.4v on a SB or SB-E isn't unreasonable and 1.5v on an FX or Phenom II isn't unreasonable either, but you're asking to damage that CPU but putting that much volts through it. IVB-E doesn't clock like a champ and when it doesn't, raising the voltage isn't always the answer. Some CPUs OC better than others and even if you do get higher clocks from the extra voltage, you're probably going to degrade your CPU so there will come a point when either the CPU will stop working or you'll never reach those kinds of clocks again and it will take more voltage to maintain lower clocks.

So in summary: Pull down those volts before you irreversibly damage something.

1.5V for benching is safe as long as you don't stay there too long. I mean come on I ran my 3960X at 1.725V for a couple of minutes I haven't gotten any degradation at all. I won't be doing 1.725V again but once in a CPUs life to get that absolute best possible score in a short benchmark I think it's worth it. I even saw a Cinebench R11.5 score where some guy ran a 4930K on 1.84V using watercooling to get 5.3Ghz. Link

As for the voltage dropping I suggest setting the switching frequency on the VRM higher if your board allows it and setting a higher LLC. I sometimes run benchmarks only relying on high LLC settigns to boost voltage under load. For example the Utra settign on the Rampage IV extreme runs 25mv above the voltage you set in idle and increases by another 75mv under load. When I do this I run lower voltages because you want to avoid having the LLC boost the voltage so high that it kills the chip.

 

[PersonWithTech]

1.5V for benching is safe as long as you don't stay there too long. I mean come on I ran my 3960X at 1.725V for a couple of minutes I haven't gotten any degradation at all. I won't be doing 1.725V again but once in a CPUs life to get that absolute best possible score in a short benchmark I think it's worth it. I even saw a Cinebench R11.5 score where some guy ran a 4930K on 1.84V using watercooling to get 5.3Ghz. Link

As for the voltage dropping I suggest setting the switching frequency on the VRM higher if your board allows it and setting a higher LLC. I sometimes run benchmarks only relying on high LLC settigns to boost voltage under load. For example the Utra settign on the Rampage IV extreme runs 25mv above the voltage you set in idle and increases by another 75mv under load. When I do this I run lower voltages because you want to avoid having the LLC boost the voltage so high that it kills the chip.

I feel that reaching a certain voltage will cause instantaneous damage the second power is applied.

 

[buildzoid]

I feel that reaching a certain voltage will cause instantaneous damage the second power is applied.

Yes but the insta kill on LN2 happens at more than 1.9V for the 22nm parts so I think up to 1.7V on water/air should be safe for about 15 minutes maybe longer. I think an Ivy bridge will die instantly at above 2.1V bellow that it will degrade really fast.

BTW 1.725V is apparently the voltage at which 32nm intels die regardless of cooling according to intel. So I've got one really tough Sandy Bridge-e.

 

[Aquinus]

Yes but the insta kill on LN2 happens at more than 1.9V for the 22nm parts so I think up to 1.7V on water/air should be safe for about 15 minutes maybe longer. I think an Ivy bridge will die instantly at above 2.1V bellow that it will degrade really fast.

BTW 1.725V is apparently the voltage at which 32nm intels die regardless of cooling according to intel. So I've got one really tough Sandy Bridge-e.

It's not the voltage that kills LN2 chips, it's the rapid expansion and contraction of the metal due to extreme temperature changes which can change the physical characteristic of a circuit to be unfavorable for normal operation.