Haswell voltage help?

[JimmiG]

Haven't used an Intel CPU since 1998 so I'm completely new to overclocking on Intel.

-When I set everything to Auto, Auto, everything works pretty much as expected. At full load, CPU-Z shows 1.104V, and the CPU Turbo Boosts to 3.7 GHz. I guess the stock voltage of my CPU is 1.104V, then.

-I then left everything at Auto and increased the multiplier to 42. The system booted, but showed a constant 4.2 GHz, at 1.36V . Did the motherboard or the CPU pick this voltage? I thought, if I didn't change anything, it would just use the stock VCore?

-I then changed all the Turbo multipliers to 42 and set the CPU multiplier back to default - The CPU now correctly throttles to 0.8V and 800 MHz at idle. At load, it has decided to run at 1.1238V for whatever reason. I didn't change anything as far as CPU voltage since I wanted to try it at stock voltage first.

-I then used "voltage offset" instead of defining the voltage. I set it to -0.05V. Voltage at full load (8 threads Prime95) is now 1.189V. This makes sense since without voltage offset, it was at 1.1238, a difference of 0.0652V (close enough to 0.05V). But why did it base this offset on the 1.238V I was getting at 4.2 GHz Auto, and not the 1.104V at stock? What causes this auto-magic bump in VCore from just increasing the multiplier?
Also, this affected idle - at 800 MHz, it now idles at 0.65V instead of 0.8V (~ 0.05V offset again...). This may cause stability issues at idle...

-I then set the Turbo multipliers to 40x, no voltage offset. The powers that be, decided to set my VCore to 1.1233V.

What's the proper way of defining voltages? I just want to bump the full-load voltage slightly and try increasing the full-load multiplier to 42-44x without affecting idle frequency and voltage... But the VCore seems to live a life of its own.

 

[JimmiG]

Well, I think I figured it out. I found this posted at Overclock.net

DVID Offset: With SB and SBe we have SVID, and SVID will change the VCore unless the VCore is explicitly set. At 4 GHz my VID will be 1.30v, at 4.1 it will then be 1.315v, at 4.3 it will be 1.335v. Setting DVID is setting a positive or negative offset onto that VID. The issue is that VID changes with frequency, so if you set a +0.100v offset, at 4GHz it would be 1.4v, at 4.1 GHz it would be 1.415v, and at 4.3 GHz it would be 1.435v. You can’t control the final VID on these CPUs, so I do not recommend using DVID offset, but if you want the voltage to drop when the frequency drops then you need to use it.

That explains the massive confusion I suffered. I thought my CPU was possessed LOL

So the "easy" way is to define a voltage and stick with it (equal to running an AMD chip without Cool n Quiet), or if you feel like it, you can use dynamic voltage and try to make it hit your desired voltage at full load and the downclocked frequency..

The problem with Haswell seems to be that the voltage wants to increase too much. I'm currently testing at 1.21V at 4.3 GHz, but with dynamic voltages, it wanted to set 1.36V already at 4.2 GHz. Setting a negative offset might affect stability at downclocked speeds (800 MHz at 0.6V might not be stable).

 

[allenk09]

If you don't need power management, just turn auto off and set the voltages yourself. My motherboard set mine to 1.296V stock, which is ridiculous . I went in and set it to 1.090V stock frequencies and it was idle and cool.

 

[JimmiG]

If you don't need power management, just turn auto off and set the voltages yourself. My motherboard set mine to 1.296V stock, which is ridiculous . I went in and set it to 1.090V stock frequencies and it was idle and cool.

Yeah, that's what I'm doing now. Too many variables with Haswell to worry about power management too, at least in the initial stages.

Currently getting close to the 1-hour mark in Prime95 at 4.3 GHz/1.21V. Core temps are hovering around the 69C mark, about 13C higher than at stock speed/VCore.

 

[allenk09]

Yeah, that's what I'm doing now. Too many variables with Haswell to worry about power management too, at least in the initial stages.

Currently getting close to the 1-hour mark in Prime95 at 4.3 GHz/1.21V. Core temps are hovering around the 69C mark, about 13C higher than at stock speed/VCore.

Nice. My boyfriend just ordered me a Phanteks PH-TC12DX, and I'll have it tomorrow. We'll see how the temps drop from the Evo 212 and maybe I'll be able to get it a bit higher!

 

[rgallant]

Yeah, that's what I'm doing now. Too many variables with Haswell to worry about power management too, at least in the initial stages.

Currently getting close to the 1-hour mark in Prime95 at 4.3 GHz/1.21V. Core temps are hovering around the 69C mark, about 13C higher than at stock speed/VCore.

-re power management on ib ,haswell should be the same.
-I did not like the clock\voltage swings from 1.6 to 4.6 so using a tip on xs I found by turning off all the PM options in the bios except EIST ,then turning up all win7 PM options [to 100% for all options] , gives me a idle clock of 3.5 then to 4.6 in non idle.
-works well for me , runs at a higher idle voltage [1.192v] but much better than 1.312v for 4.6 if fixed voltage was used.