2600k Voltage Issue

[st0ne]

Okay... Hopefully someone can help me understand and/or fix this issue, let me get started with my system:

Corsair Obsidian 650D Case
Corsair Professional Series Gold AX850 850W PSU
Corsair H100 Cooler
Asus Maximus IV Extreme
Intel Core i7-2600K 3.4GHz OC'd @ 4.5 GHz (originally at 4.8)
Corsair Vengeance 8GB 1600
OCZ Vertex 3 SSD 250GB
2 Sapphire Radeon HD 6970's in Crossfire

The issue I'm having is with voltage fluctuations. I understand this as vdroop, and is quite normal, however my problem is that when under full load the voltage drops significantly to keep the higher clock going. I've been trying to achieve a good mix of overclocking and temps. I was originally running @ 4.8 Ghz with a voltage of 1.44. Under a stable full load (prime95 blend, IBT) I was seeing the voltage drop to 1.424 with load temps in the low to mid 70's.
http://valid.canardpc.com/show_oc.php?id=2045196

I decided to sacrifice 300 mhz and drop my clock to 4.5 Ghz and my voltage to 1.370. Again, under a stable full load (prime95, IBT) I have a stable clock and my load temps are 60c, and the voltage shows 1.344.
http://valid.canardpc.com/show_oc.php?id=2047112

My problem is the fact that if I drop the voltage to anything under 1.370 it's no longer stable at the 4.5 clock. So why on earth do I need to needlessly run my pc at a higher resting voltage when the computer is stable at a lower voltage of 1.344 under load?

What I would ultimately like to do is have the voltage sit at 1.345, or even 1.35 at idle/load so I don't need to have extra needless power and heat running to the proc.

Right now I have:

vcore pwm mode set to extreme
vcore phase control set to extreme
overcurrent protection set to 120%
LLC is set at 50%
pll overvoltage is enabled
Bios is updated to the latest build

I also read somewhere that i should take my voltage while under load (1.344) and subtract it from my VID (1.3611) and set the voltage offset (rounding +0.020) to keep the voltage from jumping around. Honestly, I don't understand it all. I've been reading for HOURS trying to figure out the issue. (i don't even know how to set this with the mobo I have.)

So my problem (regardless of the actual overclocking i'm trying to achieve) is that i'm getting too much resting/idle voltage, while the pc is stable under load at a much lower voltage.

Should I not worry about this so much? My temps are fine @ 60c under load. I guess the point is I'd like to drop the voltage to the absolute minimum needed for the stable overclock, while keeping my temps at the lowest level possible.

 

[Dekar12]

Very interesting question which has often puzzled me as well. Sub'ed to the thread so that hopefully someone smarter then myself posts an answer.

 

[st0ne]

Thanks!

I will tell you this, for anyone else who may be interested:

I decided to play with CPU PLL Voltage lastnight. It was set at auto pumping out 1.8-ish volts. I manually set the PLL voltage to 1.525 to see what would happen. Not only was I still stable, but it allowed me to move the vcore from 1.375 (previously couldn't go below before immediately BSOD) down to 1.325. There is still tweaking to be done, as I will bsod with prime 95 8-ish hours into stress testing. But I think i'm onto something

 

[Copyright]

First time I have seen someone take PLL down to get stability... odd.

 

[dnottis]

First time I have seen someone take PLL down to get stability... odd.

This has been know for awhile. Many people were able to get stable by lowering PLL.

 

[st0ne]

Not only stable, but able to drop core voltage down and stay stable.

I'm down to 1.325 from 1.375 @ 4.5ghz.

 

[spine]

Don't overclock via manual voltage; always use offset.

You'll get higher stable clocks with less voltage and the chip can properly ramp up speed/voltage when needed.

It's win win.

Sandybridge's are truly amazing chips.

FYI, these chips will automatically increase VID when you increase the multiplier. It's by design. Right now I'm using an offset of +0.20 and my chip runs at 1.49v under load at 51x multi. The reported VID at this speed is 1.3811 according to Coretemp (it's much lower at stock, but I can't remember exactly what off hand). So you can't measure the VID at stock and make a calculation; you have to get there slowly, in steps, via trial and error.

 

[/usr/sbin]

Use offset voltage.

 

[st0ne]

How do I calculate what I should use as offset? Was my original post right?

Do I take the voltage given by cpu-z under load and subtract it from what it states the vid is in real temp?

 

[spine]

How do I calculate what I should use as offset? Was my original post right?

Do I take the voltage given by cpu-z under load and subtract it from what it states the vid is in real temp?

As I said, the VID increases as the multiplier increases, so you can only guess what the resultant final voltage will be.


Leave everything at stock, set the off set to +0.1, boot into windows, and check the voltage under load now as compared to stock voltage.

Then, slowly increase the multipler in the bios, booting into windows each time, and observe the resultant increase in voltage under load. This will give you a feel for your chip about where its voltage is headed.

And just basically work your way slowly up and up.

That's it.

Oh yeah, one last thing; read this;
http://hardforum.com/showthread.php?t=1578110

 

[st0ne]

Thanks so much spine!

 

[/usr/sbin]

How do I calculate what I should use as offset? Was my original post right?

Do I take the voltage given by cpu-z under load and subtract it from what it states the vid is in real temp?

-Coretemp 0.99.8 will display VID the processor requests
-CPU-Z will display actual voltage to the chip.

You calculate the differnce.

I'm running -0.045 offset, which if you do that math is about right, when you factor in a slight vDroop. Coretemp won't give you the right clock speed, but the VID is accurate.

 

[st0ne]

Ahhh, understood. Thanks so much