Overclock.net › Forums › Intel › Intel CPUs › (Updated -Part II)= = Offset Mode Overclocking Starter Guide and Thread= =
New Posts  All Forums:Forum Nav:

(Updated -Part II)= = Offset Mode Overclocking Starter Guide and Thread= = - Page 24

post #231 of 369
Thread Starter 
Quote:
Originally Posted by unclewebb View Post

Since my Asus board always boots up with EIST enabled, I wasn't sure if toggling EIST when in Windows was actually doing anything. After doing some more testing, yes you can enable or disable EIST when you are in Windows by toggling bit[16] in MSR 0x1A0 but C1E / C3 / C6 all seem to be able to over ride your EIST selection. Whether EIST is enabled or disabled, if you enable any of the low power C states (C1E/C3/C6), the CPU will start to rapidly transition between the maximum turbo multiplier and the 16 multiplier.

With the C6 core C State enabled, I can not measure any significant difference in power consumption whether C1E is enabled or disabled. Enabling C1E makes a measurable difference to power consumption if you are not using C3/C6. When C3/C6 are enabled, there isn't much time left over when the CPU will choose to use C1E over one of the other C states. It would rather be in C3/C6 to maximize power savings.

load03p.png

In the above example at idle, the cores are spending 99.5% in either C3 or C6. The RealTemp Load meter shows the percentage of time in the C0 state so the CPU is in C0 or C3 or C6 99.8% of the time which leaves 0.2% unaccounted for. That leaves 0.2% of the time left over when the CPU is in the C1 state. Whether you have C1E enabled or not, there is only 0.2% of the time when the CPU would have the opportunity to use it. If you enable C1E, it might end up spending 0.1% of the time in C1 and 0.1% of the time in C1E. The rest of the time the CPU is going to be in one of the other preferred C States. When your CPU is partially loaded the same thing happens. There is not a lot of time left over when C1E will actually be used when C3 or C6 are enabled.

I originally turned C1E off because it was preventing Windows from resuming from sleep when overclocking. With C1E enabled, Windows would hang with a black screen when trying to resume from sleep. With C1E disabled and C3 and C6 enabled, it resumes from sleep every time with no issues. Asus might have fixed this in the most recent bios.

Based on that, I consider C1E to be optional when using either C3 or C6. Sure C1E reduces power consumption in theory but the reality is that the difference is going to be difficult to measure whenever you have C3 or C6 enabled.

Edit: The power consumption numbers provided by all monitoring software that I know of are based on VID voltage and not actual voltage. You can not use software power consumption data to come to any conclusions about anything. A change in the VID voltage may or may not result in a change in the actual vCore voltage. In the above example the cores are getting virtually 0 volts 99.5% of the time because they are in C6. That only leaves 0.5% of the time when the cores are getting the vCore that software reports. With C6 enabled, the reported vCore when idle starts to become meaningless because it is barely used. High or low reported vCore at idle becomes insignificant. Going with a fixed vCore does not have the penalties that you initially assume.

Thanks for posting this thumb.gif
    
CPUMotherboardGraphicsRAM
i7 4930K Rampage IV Gene SLI Evga GTX 780 Classy 16Gb TeamX2400c9 
Hard DriveCoolingOSMonitor
Samsung 840 Pro/ Hitachi Deskstar 1TB  H20 (NexXxoS 420+280) w7x64 VG248QE 
KeyboardPowerMouseAudio
Logitech G110 SuperNOVA 1300 G2 G502 Proteus Core Foobar2K/WASAPI >NuForce Icon HDP >Crack with S... 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
i7 4930K Rampage IV Gene SLI Evga GTX 780 Classy 16Gb TeamX2400c9 
Hard DriveCoolingOSMonitor
Samsung 840 Pro/ Hitachi Deskstar 1TB  H20 (NexXxoS 420+280) w7x64 VG248QE 
KeyboardPowerMouseAudio
Logitech G110 SuperNOVA 1300 G2 G502 Proteus Core Foobar2K/WASAPI >NuForce Icon HDP >Crack with S... 
  hide details  
Reply
post #232 of 369
Quote:
Originally Posted by kevindd992002 View Post

2.) Also, why isn't your system going to the Package C state for C6? Isn't that better?

I'd also like to see this, I think he has similar board to mine (Z77-V Pro) so in theory I'm expecting to see what I see, no Package C6 state for you! (unless I've read what his mobo is wrong).
Quote:
Originally Posted by kevindd992002 View Post

3.) If we do not have a way to stop CPU frequency throttling and no way to stay at a fixed frequency, wouldn't it better to use less voltage for frequencies in between 1600MHz and, say, 4.7GHz? Because for Fixed vcore, you force the voltage to a high voltage even at frequencies like 3.3GHz when the CPU stays at that frequency. If you use Offset, the voltage scales according to the frequency used at one moment because it is based on the current VID. Isn't that better?

This is where I get confused, right. We know the CPU - as per the example he provided - is spending 99.3% of it's time at that specific time, in C6 pulling ~0v. To me that means that those threads in C6 state cannot be in the multi all the programs show (ie: fluctuating). They'd have to be in the LFM which I never see in any program otherwise you'd just BSOD. This in turn means that all the programs are wrong. If all the programs are wrong about the Multi (ie: 34.7x in that screenshot example) then offset becomes meaningless.

It's either doing work maxed out per required threads or not doing anything sitting in C6 state pulling ~0v, there is no in-between?
Just a 'puter
(18 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i7 6700K ASUS Maximus VIII Hero iGPU thanks to Nvidia 970 3.5GB Corsair Vengeance LPX CMK8GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Samsung 950 PRO 512GB Crucial M4 256GB RX360 V3 Koolance CPU-380I 
CoolingCoolingOSMonitor
Koolance RP-452X2 Reservoir Koolance PMP-450 12V Variable Speed Pump Windows 10 Pro x64 benq XL2411Z 
KeyboardPowerCaseMouse
logitech G510 Seasonic X-760 Corsair 800D Mionix Castor 
AudioAudio
Creative Sound Blaster X-Fi Titanium HD DT 990 Premium 250Ohm 
  hide details  
Reply
Just a 'puter
(18 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i7 6700K ASUS Maximus VIII Hero iGPU thanks to Nvidia 970 3.5GB Corsair Vengeance LPX CMK8GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Samsung 950 PRO 512GB Crucial M4 256GB RX360 V3 Koolance CPU-380I 
CoolingCoolingOSMonitor
Koolance RP-452X2 Reservoir Koolance PMP-450 12V Variable Speed Pump Windows 10 Pro x64 benq XL2411Z 
KeyboardPowerCaseMouse
logitech G510 Seasonic X-760 Corsair 800D Mionix Castor 
AudioAudio
Creative Sound Blaster X-Fi Titanium HD DT 990 Premium 250Ohm 
  hide details  
Reply
post #233 of 369
ok so I have started testing with unclewebs method, I have the asus z77 sabertooth with most recent bios 1805. I am able to turn off eist and my cpu doesn't downclock even in balanced power plan, so it seems my board is different than other asus boards. Anyway I am still testing this method now so far I am 10 runs in on IBT and because I am using a fixed voltage I am able to run at less voltage and lower temps, plus I can have c3,c6 and package cstates enabled, where when I was using offset I followed the asus guide and left everything except c1e and eist disabled to be stable. so If the cores still go to 0volts in c3/c6 I think It would be smarter to use unclewebbs method because I can't use c3/c6 and be stable at 4.8


just something I have noticed so far using this method I see a total loss of 5 points on IBT gflops using this, but the gflops are way more consistant with a manual vcore it seems. on offset it ranged from 117-124 and I am getting a consistant 112 during my current tests.
Edited by enigma7820 - 2/20/13 at 5:46pm
post #234 of 369
also if using offset can you use eist, c1e and package c-states, or if you have c3 and c6 disabled does package c-states not work?
post #235 of 369
Quote:
Originally Posted by enigma7820 View Post

ok so I have started testing with unclewebs method, I have the asus z77 sabertooth with most recent bios 1805. I am able to turn off eist and my cpu doesn't downclock even in balanced power plan, so it seems my board is different than other asus boards.

It's a PITA to try and figure out the mood of the asus people and what they choose to remove/add. For example I can see using AMIBCP that power state string is in the BIOS but the actual value is not defined anywhere so of course there's no way to change it. It's same deal with the 'additional turbo voltage', it's there as a string but not defined anywhere.

Re: the package state, AFAIK though someone correct if wrong, if you don't have core C states enabled (C3/C6) then there's no way for them to go into Package C state either. I actually have no idea what benefit Package C state does besides supposedly telling me that the cores are in sufficient C state for Package C state to kick in.
Just a 'puter
(18 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i7 6700K ASUS Maximus VIII Hero iGPU thanks to Nvidia 970 3.5GB Corsair Vengeance LPX CMK8GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Samsung 950 PRO 512GB Crucial M4 256GB RX360 V3 Koolance CPU-380I 
CoolingCoolingOSMonitor
Koolance RP-452X2 Reservoir Koolance PMP-450 12V Variable Speed Pump Windows 10 Pro x64 benq XL2411Z 
KeyboardPowerCaseMouse
logitech G510 Seasonic X-760 Corsair 800D Mionix Castor 
AudioAudio
Creative Sound Blaster X-Fi Titanium HD DT 990 Premium 250Ohm 
  hide details  
Reply
Just a 'puter
(18 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i7 6700K ASUS Maximus VIII Hero iGPU thanks to Nvidia 970 3.5GB Corsair Vengeance LPX CMK8GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Samsung 950 PRO 512GB Crucial M4 256GB RX360 V3 Koolance CPU-380I 
CoolingCoolingOSMonitor
Koolance RP-452X2 Reservoir Koolance PMP-450 12V Variable Speed Pump Windows 10 Pro x64 benq XL2411Z 
KeyboardPowerCaseMouse
logitech G510 Seasonic X-760 Corsair 800D Mionix Castor 
AudioAudio
Creative Sound Blaster X-Fi Titanium HD DT 990 Premium 250Ohm 
  hide details  
Reply
post #236 of 369
Ok what about this if in my windows power plan I put CPU to sleep after x amount of time do the cores goto 0v still without using cstates or does it just sit at 1600mhz and the 1v I see when im at idle in windows monitoring it?
post #237 of 369
Quote:
Originally Posted by enigma7820 View Post

ok so I have started testing with unclewebs method, I have the asus z77 sabertooth with most recent bios 1805. I am able to turn off eist and my cpu doesn't downclock even in balanced power plan, so it seems my board is different than other asus boards. Anyway I am still testing this method now so far I am 10 runs in on IBT and because I am using a fixed voltage I am able to run at less voltage and lower temps, plus I can have c3,c6 and package cstates enabled, where when I was using offset I followed the asus guide and left everything except c1e and eist disabled to be stable. so If the cores still go to 0volts in c3/c6 I think It would be smarter to use unclewebbs method because I can't use c3/c6 and be stable at 4.8.

Can you check the power use with a Kill-A-Watt or something and compare it to the other method (to see if it is actually going into C3/C6 with EIST off)? Does Realtemp show it using C3/C6 states with EIST off?

My only knowledge about package C states is the understanding that all the cores have to be in a given C state before the package can go there. Or at least that's how I read the Intel white paper. I don't know what advantages it offers.
post #238 of 369
Quote:
Originally Posted by error-id10t View Post

This is where I get confused, right. We know the CPU - as per the example he provided - is spending 99.3% of it's time at that specific time, in C6 pulling ~0v. To me that means that those threads in C6 state cannot be in the multi all the programs show (ie: fluctuating). They'd have to be in the LFM which I never see in any program otherwise you'd just BSOD. This in turn means that all the programs are wrong. If all the programs are wrong about the Multi (ie: 34.7x in that screenshot example) then offset becomes meaningless.

It's either doing work maxed out per required threads or not doing anything sitting in C6 state pulling ~0v, there is no in-between?

Yeah, this is the only thing that's confusing me right now. The thing is we don't even know if at C6 the frequency "needs" to be at LFM. The LFM still needs around 1.000V as we know so 0V won't cut it. I want to understand if there are operations that will make the CPU run at frequencies in between the LFM and max mutli and if Offset voltage is better at those frequencies.

Here's an example assuming C3/C6 are enabled for all cases:

1.) For fixed vcore (say 1.45V), it will be applied as the vcore from frequencies 3.3GHz up to 4.7GHz (max turbo multi).

2.) With offset vcore, the vcore will be lower for frequencies less than 4.7GHz because it adds/subtracts the offset value from the VID of each multi. The lower the multi, the lower the VID, the lower the effective vcore.

In that example, what is the disadvantage of using offset then aside from the difficulty in making it stable with C3 and C6 enabled?
post #239 of 369
i wish I had a kilowatt haven't had a chance to pick one up yet. but it does show me in real temp just like in unclewebbs screenshot that the c3 and c6 where being using right away when booted into windows and cpu stayed at 4800 the whole time never downclocking or lowering vcore. only difference was idle temps were slightly higher. I personally don't feel comfortable with this way and will stay with my offset until I feel I know enough.
post #240 of 369
Quote:
Originally Posted by kevindd992002 View Post

1.) Thanks for the confirmation. If C1 is not a setting in the BIOS, is automatically enabled? In your example above, if I disable C1E then where would the 0.2% go? Will it be shared by C3 and C6?

2.) Also, why isn't your system going to the Package C state for C6? Isn't that better?

3.) If we do not have a way to stop CPU frequency throttling and no way to stay at a fixed frequency, wouldn't it better to use less voltage for frequencies in between 1600MHz and, say, 4.7GHz? Because for Fixed vcore, you force the voltage to a high voltage even at frequencies like 3.3GHz when the CPU stays at that frequency. If you use Offset, the voltage scales according to the frequency used at one moment because it is based on the current VID. Isn't that better?

Most computers are set up so when the CPU is working on something it will be in the C0 state and when a core has nothing to do, it will enter the low power C1 state so unless there is a problem with Windows or the bios, C1 is always available. There is actually an way you can disable C1 and prevent your CPU from using any C states. The performance improvement when benching is so tiny that I don't think anyone should bother but here's how if you are interested.

http://forum.techinferno.com/throttlestop-realtemp-discussion/2925-how-get-c0%25-%3D-100%25.html

Preventing a CPU from entering C1 mostly just creates heat. RealTemp will report the load as a constant 100%, even when idle, because the RealTemp Load value is really just showing the percentage of time the CPU is in the C0 state.

1) If you disable C1E then the CPU will spend its time in C1 instead. Either way, we are not talking about a huge percentage of time that a CPU will end up in C1 or C1E.

2) The package C States on my board are kind of interesting. When I enable that in the bios, the CPU power consumption might decrease but the power consumption at the wall increases. I haven't figured out why yet but it seems like it is another unresolved bios issue. Asus had a lot of BSOD issues on their G53 and G73 laptops when the C package states were enabled so I am starting to wonder about what new technology their engineering team have come up with. Power consumption at the wall should definitely not go up when you enable the C Package states but on this board it does.

3) The point of all this is to spend as little time as possible at all of those intermediate multipliers. When a CPU has something to do, get it up into the C0 state at the maximum speed to get the job done as quick as possible and then get it back into the low power C6 state. This minimizes power consumption and maximizes performance. Kind of a win win situation.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › (Updated -Part II)= = Offset Mode Overclocking Starter Guide and Thread= =