Post your New Ivy Bridge Processor Thermals, Type of Cooling, and First Impressions:

i7-3570k_4400Mhz_IntelBurn.jpg


Had to set an offset of +0.40 to get it stable on an ASUS Z77 WS (offset at auto sets the vcore to over 1.3v).
This is with a Corsair H80 in a Silverstone FT-02.

I'm not sure if the voltage is too high for 4,4Ghz, it was at 1.192v most of the time during IBT and Prime95. Setting affinity to one core to emulate single core usage pushes the voltage to 1.224v.

Seeing 69°c max on one core temps could be a little lower maybe for a H80, I used MX-4 when I re-applied the cooler for the 3750k. Ambient temperature is about 26°c.
 
Last edited:
that sound about right. offset of 0.03-0.04 was required for 4.4ghz for me.
now try 4.5ghz.... its crazy how much more voltage i had to pump in that chip for 100mhz lol
 
hey milena. how do you like your motherboard so far?i ran tri sli with my gtx 280s before i went to my gtx 680 and i think in 2 years or so i will probably follow the same path with the gtx680s. the extreme9 and maximus aren't out yet but I anticipate they are more money. Let me know what you think and I will keep it in mind :) thanks
 
Last edited:
With just these three steps from bios.

Ai Overclock Tuner X.M.P
Ratio Limit 47
CPU Manual Voltage 1.275

ibt47.png
 
Hit my target of 4.2Ghz, all settings on Auto, ambient temperature is about 75F, H100 fans running on low.

BurnTest.jpg
 
zpaf said:
With just these three steps from bios.

Ai Overclock Tuner X.M.P
Ratio Limit 47
CPU Manual Voltage 1.275

ibt47.png
Click to expand...

Nice! I have 1.25 for 4.7 and I put that a couple notches high on purpose. You can probably get lower.
 
Guys, I'm finding to be stable for P95 custom blend using 85% or more of system ram, you need your vcore to be higher.

The default blend mode utilize much less ram. Stressing system to the max with nearly all ram is much much more difficult.

Before, I was stable in P95 blend with about 1.256-1.264v under 100% load. WIth memory set to max, I need around 1.288v under load to be stable.

Of course everyone's idea of stability is different but I don't fully trust default blend mode.
 
christpunchers said:
Guys, I'm finding to be stable for P95 custom blend using 85% or more of system ram, you need your vcore to be higher.

The default blend mode utilize much less ram. Stressing system to the max with nearly all ram is much much more difficult.

Before, I was stable in P95 blend with about 1.256-1.264v under 100% load. WIth memory set to max, I need around 1.288v under load to be stable.

Of course everyone's idea of stability is different but I don't fully trust default blend mode.
Click to expand...

I use Intel Burn Test on maximum for 5 passes which ends up being about 10 minutes total. According to the software 20 minutes is equivalent to 48 hours prime so I would guess 10 minutes is as good as 24 hours. Everyone has different opinions on the comparisons between these two programs. I haven't crashed yet so IBT is fine for me.
 
chronicfx said:
I use Intel Burn Test on maximum for 5 passes which ends up being about 10 minutes total. According to the software 20 minutes is equivalent to 48 hours prime so I would guess 10 minutes is as good as 24 hours. Everyone has different opinions on the comparisons between these two programs. I haven't crashed yet so IBT is fine for me.
Click to expand...

IBT is a good test but I don't think it's better than Prime, especially not compared to a 12-24hour+ Prime run.

I've passed LinX, which is pretty much the same thing as IBT, for about 5 runs or so. But the very same overclock using max memory in Prime will not be stable (rounding error, one thread failing, or app crashing, etc).

For everyday use, it may not make that big of a difference if it is 90% stable instead of 99.9% stable. But I think it's best to be safe. From what I know data can corrupt over a period of time if the CPU is constantly making errors without the user's notice. Depends on what you do with the system I suppose.
 
christpunchers said:
IBT is a good test but I don't think it's better than Prime, especially not compared to a 12-24hour+ Prime run.

I've passed LinX, which is pretty much the same thing as IBT, for about 5 runs or so. But the very same overclock using max memory in Prime will not be stable (rounding error, one thread failing, or app crashing, etc).

For everyday use, it may not make that big of a difference if it is 90% stable instead of 99.9% stable. But I think it's best to be safe. From what I know data can corrupt over a period of time if the CPU is constantly making errors without the user's notice. Depends on what you do with the system I suppose.
Click to expand...

Yeah I agree it is best to be safe. My only hangup is that no-one can really tell if prime crashed because they were unstable or if the program itself crashed. Is there any resolution to that question? This is just what I have heard and correct me if I am wrong that you can fail prime through no fault of your system but through the program itself just crashing and maybe people will up voltages that never needed to be pushed up.
 
chronicfx said:
Yeah I agree it is best to be safe. My only hangup is that no-one can really tell if prime crashed because they were unstable or if the program itself crashed. Is there any resolution to that question? This is just what I have heard and correct me if I am wrong that you can fail prime through no fault of your system but through the program itself just crashing and maybe people will up voltages that never needed to be pushed up.
Click to expand...

I was running the latest version of prime and the app kept crashing regardless of my clock speeds and it would crash at the same time every time. I could have set a stop watch to it and be within a few minutes of predicting the hang. I downloaded a different version and passed well over 24hrs after which I stopped it myself.
 
chronicfx said:
Yeah I agree it is best to be safe. My only hangup is that no-one can really tell if prime crashed because they were unstable or if the program itself crashed. Is there any resolution to that question? This is just what I have heard and correct me if I am wrong that you can fail prime through no fault of your system but through the program itself just crashing and maybe people will up voltages that never needed to be pushed up.
Click to expand...

there are so many other programs as well. I remember JJ from Asus youtube videos mentions aida64 as his stress test tool. Try that if you are un sure about prime
 
chronicfx said:
hey milena. how do you like your motherboard so far?i ran tri sli with my gtx 280s before i went to my gtx 680 and i think in 2 years or so i will probably follow the same path with the gtx680s. the extreme9 and maximus aren't out yet but I anticipate they are more money. Let me know what you think and I will keep it in mind :) thanks
Click to expand...
So far I can say it's an amazing board. I like the z77 ws much more than the deluxe and other versions, it's prepared for a future tri or quad gpu setup and can still serve cards like a raid controller very well. Some of the features like dual intel server class lan are nice to have and the workstation series got cross-shipping rma support by asus, thats something we usually don't get at all here in europe. US$339 at newegg, I paid €289 at my place.

I'm probably leaving the 3570 at 4,4 for now as its stable and fast enough.
 
Hi folks.

A few question for those overclocking 3770K.

My setup.
3770K
AsRock Fatality Z77 Professional
4 x G.SKILL ARES 2133MHz (16GB)

I've found a stable overclock for 3770K @ 4.5GHz
Offset = +0.050V
Load-line calibration = Auto (Level 5 which is the lowest/mildest)
All power management options are at default (enabled) (C1E, C3/C6 etc)
Under Prime95 and LinX load, CPU-Z reports 1.200V
This passed 20 passes of LinX and 10 hours of Prime95 (Temps ~ 93C under full load)


Now, my questions are :

1. Is it better to *raise* load-line calibration and *lower* offset voltage ?
For example : Offset = +0.040V with LLC at Level 2 ?
Not sure if this would make difference at all since my 3770K seems to require 1.200V under full load to be stable @ 4.5GHz. (Be it with LLC or not, as long as 1.200V is provided to the 3770K, it's the same, right ?)

2. When I went with *Fixed* Mode (rather than *Offset* Mode), below were the value/settings in BIOS which gave 1.200V under full load (according to CPU-Z).
Fixed = 1.215V (BIOS)
Load-line calibration = Level 2 (BIOS)
All power management options are at default (enabled) (C1E, C3/C6 etc)
Under idle as well as Prime95 and LinX load, CPU-Z reports 1.200V
And this gave me BSOD in Prime95 load.
The same voltage was applied to the 3770K. (Offset Mode and Fixed Mode) Why would Fixed mode give me BSOD while Offset Mode goes stable ? (After all, the same amount of juice is provided to the CPU (1.200V)).

Thanks!
 
So i have 3570K at X45 (@4.5 ghz). CPU voltage at 1.2V and PLL at 50%. Max temp with Hyper 121+ is 76C for one of the cores and rest at 71C max. Looks pretty good to me, but should I try to lower voltage or try for higher clocks? I feel like this is a good spot, but just want second opinions. I'll post screens once i figure out where i want to stop.

Autotune went pass 4.6 and was unstable at higher than 4.6 then throttled it down to 4.55 ghz with higher voltages (1.3V and PLL 1.8V) and temperatures went almost 87C. But I think i'm okay with 4.5ghz.
 
Last edited:
ShePearl said:
Hi folks.

A few question for those overclocking 3770K.

My setup.
3770K
AsRock Fatality Z77 Professional
4 x G.SKILL ARES 2133MHz (16GB)

I've found a stable overclock for 3770K @ 4.5GHz
Offset = +0.050V
Load-line calibration = Auto (Level 5 which is the lowest/mildest)
All power management options are at default (enabled) (C1E, C3/C6 etc)
Under Prime95 and LinX load, CPU-Z reports 1.200V
This passed 20 passes of LinX and 10 hours of Prime95 (Temps ~ 93C under full load)


Now, my questions are :

1. Is it better to *raise* load-line calibration and *lower* offset voltage ?
For example : Offset = +0.040V with LLC at Level 2 ?
Not sure if this would make difference at all since my 3770K seems to require 1.200V under full load to be stable @ 4.5GHz. (Be it with LLC or not, as long as 1.200V is provided to the 3770K, it's the same, right ?)

2. When I went with *Fixed* Mode (rather than *Offset* Mode), below were the value/settings in BIOS which gave 1.200V under full load (according to CPU-Z).
Fixed = 1.215V (BIOS)
Load-line calibration = Level 2 (BIOS)
All power management options are at default (enabled) (C1E, C3/C6 etc)
Under idle as well as Prime95 and LinX load, CPU-Z reports 1.200V
And this gave me BSOD in Prime95 load.
The same voltage was applied to the 3770K. (Offset Mode and Fixed Mode) Why would Fixed mode give me BSOD while Offset Mode goes stable ? (After all, the same amount of juice is provided to the CPU (1.200V)).

Thanks!
Click to expand...

I would like question #1 answered as well. I do have my LLC set to extreme for 4.7GHz and 1.25v. The largest "spike" of voltage I have ever seen is 1.256v (this normally happens for a split second when the load drops off) and normal load is 1.240v (like cinebench) or 1.248v (IBT).

I could see when someone is pushing the voltage for way upper limits (like 1.35v+ they would not want any spikes of overvoltage whatsoever, but being at a modest 1.256v for spikes is this really a system killer? Please don't skip this one :)

2. I don't know whats up with prime. We discussed it before if you look up a few posts, try aida64 or trust IBT very high or maximum in my personal opinion.
 
chronicfx said:
I would like question #1 answered as well. I do have my LLC set to extreme for 4.7GHz and 1.25v. The largest "spike" of voltage I have ever seen is 1.256v (this normally happens for a split second when the load drops off) and normal load is 1.240v (like cinebench) or 1.248v (IBT).

I could see when someone is pushing the voltage for way upper limits (like 1.35v+ they would not want any spikes of overvoltage whatsoever, but being at a modest 1.256v for spikes is this really a system killer? Please don't skip this one :)

2. I don't know whats up with prime. We discussed it before if you look up a few posts, try aida64 or trust IBT very high or maximum in my personal opinion.
Click to expand...

From reading Anandtech's article on regarding LLC and vdroop, it was my understanding that the voltage spikes do not show up on CPU-Z and the only way you would be able to 'see' the spikes and drops in voltage is with an oscillator. So basically whatever is showing up in CPU-Z does not show the actual additional voltage that LLC is providing. Someone correct my understanding if I'm mistaken?
 
The voltage spike itself happens in micro/milliseconds and cpuz's refresh is in second(s), so it is highly unlikely you would ever see it...
 
Nearsite said:
From reading Anandtech's article on regarding LLC and vdroop, it was my understanding that the voltage spikes do not show up on CPU-Z and the only way you would be able to 'see' the spikes and drops in voltage is with an oscillator. So basically whatever is showing up in CPU-Z does not show the actual additional voltage that LLC is providing. Someone correct my understanding if I'm mistaken?
Click to expand...

Thanks for the info. I just found that article. If it was the one with the qx9650. It seems that those spikes can be pretty serious, they couldn't be sure if they weren't in excess of 1.6v!! I only scanned the article for a total of two minutes but thats the jist I got. So I guess I will drop my LLC to level 3 or so.
 
fps4ever said:
Why is what reading the same?
Click to expand...

When I do IBT and the load lets off I always see that exact same value just appear and disappear for a split second. Every pass at the end. So I was under the impression I was "seeing it" but as you said it makes sense that this is happening in microseconds and there is no way something that polls in seconds would see this.

Anyone want to recommend an LLC (1 thru 5 with 1 being extreme) setting for 4.7GHz???

I will try to make it stable tonight using offset, I don't want to hurt my chip with these spikes as I have it on extreme LLC right now.
 
I have a similar question to ShePearl.

Last night I started using the UEFI to dial in my overclock using offset voltage. Previously I had been using TurboV Evo in Windows to apply an OC profile. I knew that my CPU would handle 4.5GHz at around 1.22V load based on the Asus auto-tune in TurboV Evo, so my approach was to go into BIOS, and figure out what offset to hit around 1.2V (turned out to be 0.07V) and work back from there.

My goal is to achieve the lowest voltage at load and idle possible, while maintaining stability with turbo/power saving enabled. I'm not sure what to do with LLC/PLL in this case? My other concern is that at max load (LinX) the CPU voltage is lower than at my realistic max load which is BF3 gaming (and rare video encoding/transcoding). Is there a way to use LLC to tighten in the droop so my LinX and BF3 load voltages are closer (see results below - they are about 0.04V apart right now).

When I get some more time I will try LLC on regular rather than auto and see what it does. I think this CPU can handle slightly lower idle voltage (0.97ish maybe less?). Also I realize my load testing is not enough to conclude on stability. I will do a 5-10 pass LinX run at 100% memory, followed by 12+h Prime95 as well.

Settings:
Asus mode: XMP
CPU multiplier: 45
Voltage: Offset +0.06V
CPU Spread Spectrum: disabled
Everything else is auto (PLL, PLL Overvoltage, LLC, all turbo/power saving enabled)

Test:
LinX (using latest libraries/AVX = 104GFlops at 4.5GHz HT on) - 30min 2.5 passes
Diablo 3 - 1 hour
BF3 - 1 hour

Results:
ambient room temp 22c
idle voltage (1600MHz on desktop for a few minutes): 0.984V
idle core temps 29/24/26/30
load voltage during LinX (using all memory available): 1.184V
load core temps LinX 69/76/72/71
load voltage during BF3/Diablo 3: 1.232V / 1.224V - note these are just the max recorded by HWiNFO64, I was not actively monitoring them so might have been temp spikes
load core temps during BF3/Diablo 3: 63/64/60/60 and 53/53/47/50

Overall I am very happy so far with this, given all the negative sentiment around IB at release. I may try for higher frequency, but am happier if I can keep LinX below 80c and my "realistic" gaming loads in the mid to low 60s. This way my system can remain near silent even during gaming.
 
Just re-overclocked using offset of 0.100v and LLC = 3

before I had 0.055v and LLC = 1.. Is this really going to be safer?

This is for 4.7GHz
 
chronicfx said:
Just re-overclocked using offset of 0.100v and LLC = 3

before I had 0.055v and LLC = 1.. Is this really going to be safer?

This is for 4.7GHz
Click to expand...


What does LLC of 1 equate to? For my Asus board, LLC at 'Regular' means no LLC at all, while the next step up is LLC Medium which equates to 25%, meaning (I'm assuming here) that it will allow the voltage spikes to go past the required mobo voltage by 25%. Next is high which equates to 50%.

BTW - that is an awesome OC. I can't get mine to 4.4ghz stable for some reason, but I did get mine to 4.3ghz stable with a -.030 offset and LLC set at Medium. When I idle, the voltage drops down to .922-.936, and during load (IBT at high) it goes up to 1.18-1.19
 
Nearsite said:
What does LLC of 1 equate to? For my Asus board, LLC at 'Regular' means no LLC at all, while the next step up is LLC Medium which equates to 25%, meaning (I'm assuming here) that it will allow the voltage spikes to go past the required mobo voltage by 25%. Next is high which equates to 50%.

BTW - that is an awesome OC. I can't get mine to 4.4ghz stable for some reason, but I did get mine to 4.3ghz stable with a -.030 offset and LLC set at Medium. When I idle, the voltage drops down to .922-.936, and during load (IBT at high) it goes up to 1.18-1.19
Click to expand...

thanks! on my board LLC 1 is the highest or extreme. 3 would be medium and 5 would be lowest.
 
chronicfx said:
thanks! on my board LLC 1 is the highest or extreme. 3 would be medium and 5 would be lowest.
Click to expand...

I would just use the lowest LLC setting that still keeps your OC stable. If you could not use it at all, that's probably even better.
 
Nearsite said:
I would just use the lowest LLC setting that still keeps your OC stable. If you could not use it at all, that's probably even better.
Click to expand...

I will give it a shot but it seems like the offset is going to be pretty large. Is that ok?
 
chronicfx said:
I would like question #1 answered as well.
Click to expand...

grambo said:
I have a similar question to ShePearl.
Click to expand...

chronicfx and grambo,

Upon my further googling, I've found close answers to my questions. Let me share and I hope this helps you guys, too.

Please read 2 quotes below.

Offset voltage is a value added to the base voltage of your CPU. If at stock you have 1.1V, and you set +0.2 to the offset voltage, you are stating that the maximum voltage is going to be 1.3

Load Line Calibration forces the voltage you set in the BIOS through the chip, and gets rid of Vdroop. Vdroop is a drop on voltage which follows Intel specifications to avoid damaging the CPU. WITHOUT LLC, when you set 1.35V on the BIOS, while you stress test the chip, voltage might drop to 1.3 - 1.25 V. When the test stops it will go back up. So Vdroop only is enabled while your CPU is under stress. This means you will have a larger voltage when idle than while doing work.

Therefore you need a method of having a lower voltage when you CPU is idling. Enter C1E. With this feature, the CPU lowers it's multiplier when it`s idling, but also lowers the voltage! This is why Offset voltage is useful. It dynamically adjusts the clock AND voltage to what the chip needs.

Without Offset you CPU doesn't know what maximum and minimum voltage you want. It will just stay at the same voltage the whole time, while idling and while doing work. If the voltage is not very high, the degradation shouldn't be substantial, but it's always nice having lower voltage when you don't need alot of power. Why would you want to use 1.45V when browsing the web when you could do the same at 1.1?

Also, this inderectly lowers the temps a bit. With lower voltages the chip is less hot, but when it idles it also underclocks and the temperature difference is very noticable.
Click to expand...

Offset mode gives you more stability at lower maximum voltages. When settings the voltage change it from fixed to offset. You will often get lower stable voltages than with fixed, that alone is a great reason to do it. The other reason offset is sweet is that it doesn't keep your processor at a set voltage. This allows you to save power, heat, and add longer life to your processor. Just start at +0.005V and work your way up. I was able to get it working at -0.010V (which when under full load was 1.23V) but it wasn't stable enough.

For my setup +0.005V = 1.28V under full load.
Click to expand...

From this thread.
http://www.overclock.net/t/999060/offset-vs-fixed-vcore-2600k-ocing

And a good post on Vdroop.
http://www.overclock.net/t/197804/what-is-vdroop#post_10334580
 
Thank you so much for the clarification. I just re-overclocked again as per-nearsites advice. I have LLC now set to 5 which on the Asrock board I assume to be "off" and my 4.7GHz overclock just passed IBT on maximum stress for 5 passes which is my OK to go ahead and use for gaming.

My offset is +0.145 for LLC=5 which according to CPU-Z gives a vdroop from 1.298v which is the highest voltage I see but it is only for a sec at the very begining of the pass, all the way down to 1.216v under extreme load in the middle of the IBT pass, and 1.240v & 1.248v interchangeably under normal load near the end of the IBT pass. My idle at 1600MHz is 1.104v.

So are you guys are absolutely sure this is the way to go? Higher offset > Higher LLC?

Highest core is 79°C. But my central air is pumping as it is almost 90°F today that may have something to do with it being lower.
 
Last edited:
FYI nearsite. In your sig you have a 3750k listed as your cpu. Figured I could help you out a bit too :)
 
Generally speaking, higher offset with a lower setting of LLC is preferred. The final voltage is going to be the same, but you can reduce the voltage spikes that LLC can cause and you can avoid problems with lightly loaded tasks crashing because LLC isn't compensating enough.. And by lower setting I mean something like regular is okay - you don't need to go all the way to off.
 
chronicfx said:
I will give it a shot but it seems like the offset is going to be pretty large. Is that ok?
Click to expand...

You need to find the equilibrium. Thats the game. At a certain point you may want to apply low level LLC depending on just how much offset you need to give it otherwise. I personally will not go higher then the lowest level of LLC my board provides. The higher you go the higher the voltage spikes and the greater chance of instability during idle/light/moderate usage. The flip side is, if you don't apply any at all you might be providing more voltage than really necessary at light/moderate loads. Just gotta find that sweet spot.
 
RamonGTP said:
You need to find the equilibrium. Thats the game. At a certain point you may want to apply low level LLC depending on just how much offset you need to give it otherwise. I personally will not go higher then the lowest level of LLC my board provides. The higher you go the higher the voltage spikes and the greater chance of instability during idle/light/moderate usage. The flip side is, if you don't apply any at all you might be providing more voltage than really necessary at light/moderate loads. Just gotta find that sweet spot.
Click to expand...

Thats exactly how i feel right now. The voltage is up at almost 1.3v (1.288v) when there is not a heavy load but the proc has gone to 4.7GHz from idle. I don't want to have such a high voltage especially when ivy is an unknown right now. Not sure what to do now.. I seem to get conflicting advice.
 
How come every overclocking guide for ivy bridge recommends being one step down from extreme for load line calibration on air for a 4.7GHz overclock?
 
Playing a bit more with LLC and Offset:

I had LLC at Auto which is probably same as Regular (0%) for my MB (saw no difference in voltages) and an Offset of +40 to get stable at 4,4Ghz. Max voltage during IBT was 1.192v which is fine but max voltage during light desktop load went up to 1.232v - more than necessary imo and thats because Vdroop isn't lowering it for light load situations.

1.232v is the real voltage I set with +40 Offset and when I tried to get stable at 4,4Ghz I took Vdroop into account to get 1.192v under heavy load.

Setting LLC to Medium (25% for ASUS Z77 MB's) allowed me to drop my Offset from +40 all way down to a negative setting of -10. Heavy load voltage is still about the same due to Medium LCC compensating the Vdroop but light desktop load doesn't spike so high anymore like it did with the high Offset voltage.

I guess LLC is a neat way to get optimal idle/light/heavy load voltages.
 
Milena said:
Playing a bit more with LLC and Offset:

I had LLC at Auto which is probably same as Regular (0%) for my MB (saw no difference in voltages) and an Offset of +40 to get stable at 4,4Ghz. Max voltage during IBT was 1.192v which is fine but max voltage during light desktop load went up to 1.232v - more than necessary imo and thats because Vdroop isn't lowering it for light load situations.

1.232v is the real voltage I set with +40 Offset and when I tried to get stable at 4,4Ghz I took Vdroop into account to get 1.192v under heavy load.

Setting LLC to Medium (25% for ASUS Z77 MB's) allowed me to drop my Offset from +40 all way down to a negative setting of -10. Heavy load voltage is still about the same due to Medium LCC compensating the Vdroop but light desktop load doesn't spike so high anymore like it did with the high Offset voltage.

I guess LLC is a neat way to get optimal idle/light/heavy load voltages.
Click to expand...

Well, Auto is not actually off. Auto is letting the board decide which LLC level it requires to maintain stability. I would set it one up from Off.
 
Last edited:
You must log in or register to reply here.
Back
Top