First time AIO cooler, concerns

[ritchiedrama]

The System Agent Voltage and I/O voltage will have an impact on your Memory overclocking and not CPU temperature. Setting fixed values was likely a smart move and most every platform on DDR4 has needed manual settings for those values when using higher frequencies for the DRAM. However, I have no idea what values are appropriate for Z270 and this can be both frequency and individual CPU specific. That raises another question. What speed is your memory? If you enabled XMP for some frequencies (like 2800 or 3000), the BIOS may have automatically set your BCLK to 125 which would force you to use manual/fixed voltage (40x125=5000MHz instead of 50x100=5000).

 

EIST/Speedstep will allow your CPU to work at intermediate frequencies for low level background tasks. You can usually see the clocks working at the default, non-turbo level (4.2GHz). However, since you are still on manual/fixed voltage, it's going to provide 1.30v with the lower frequency anyway, so this is wasted on the current set-up. If there is one power saving feature you could disable, EIST would be it. What you really want is the adaptive voltage so your voltage will step down to about half that value at idle. That might knock your idle temps down by 10C and there is no reason to have it sitting at 5.0/1.30v every moment it is running.

 

I don't know what the TMPIN3 represents. This is motherboard and/or platform specific.

 

If you found your way into the BIOS to set the manual voltage, find your back to set it to adaptive. It will be in the same place. When you select adaptive from the pop-up menu, it should create two additional boxes. Usually you input the the max value you want (1.28-1.30) into the 'Additional Voltage Box' while leaving the 'Offset' to AUTO. This will cause the CPU to use the standard voltage table when below the turbo frequency (4.5) and your voltage whenever it jumps to 5.0. Again, this does not do anything for your synthetic stress tests and is for normal usage power/temp reductions. In fact, the synthetics might be worse when using adaptive. I don't see any reason to keep running them unless you are doing so for stability. You know they are going to be warm. Focus on the programs you actually use.

 

 

[ame]

[/ame]

 

These are my temps running Aida64 currently at 1.32vcore and 5ghz oc. I think its ok?

 

When I'm gaming as I said it does similar to this video hits 78-79 for one second then back down, so I think its fine..

 

I will try find this adaptive voltage setting..

 

What I really want is when it's idle to drop to like 1000mhz, with lower voltage, then when im using it, back to 5ghz with 1.32v is this do-able?

 

It would mean I can happily leave my PC on over night to download big games then.

[c-attack]

Yes, this is what setting the adaptive voltage and EIST will do. You can also set adaptive voltage without EIST. If you still enable basic C1/C1E c-states, the cores will drop down momentarily at idle saving you some heat, although not nearly as much with speedstep enabled. It kind of pings back and forth between the minimum (1-1.2GHz probably) to 5.0. With both enabled, it will run flat at the minimum while on the desktop then perk up when a task is assigned.

 

These two graphs are old and from a slightly different discussion, but it does illustrate what you are looking for.

 

Here is Adaptive, C-states C0/C1 limit, and EIST while at the desktop. You can see a core fire off every now and then when some task is assigned. The blue line is my cache frequency. You can ignore that and it will not completely flatten out without some deeper c-state packages.

 

These are the same settings while playing a CPU limited MMO. The first third is the pre-event or normal gameplay. It is not a challenging program to run, so two of my 6 cores are assigned as workers. Those are the two at the 4.5GHz line. The other 4 cores are running background tasks, monitoring programs, or other game program related subroutines. They bounce between the inactive state (1.2GHz) and the 3.5GHz non-turbo maximum. Then the action starts and all 6 cores go into full throttle 4.5 to support the work. The one core doing meaningless windows stuff drops down occasionally, but if the load is sufficient, it stays up too. At the very end of the graph I go to the game main menu and then down to the desktop to take the picture. All other cores downshift, and the one core looking after the game (still running) stays active.

 

This last one is what happens when you enable the Performance Power Plan in Windows, which has a 100% processor up time. I am not actually doing anything, other than sitting on the desktop. Half way through, I turn it back to Balanced. This is similar what your processor will do when set to adaptive with no speed step or on manual/fixed with C1E enabled. However, it is preferable to use the Power Plan to create the 100% uptime state. First, it can be toggled on and off with two clicks from the desktop, while in game. No reboot required to test differences in performance. Secondly, there are some hidden parameters related to other CPU functions that are put on full alert in the Performance plan. I don't know them all and am not qualified to discuss how much real benefit they have, however disabling C-states/EIST does not increase the performance of those other areas.

 

I don't have Overwatch, so I don't know how to gauge the CPU behavior. It is not a difficult game to drive, but it is frequently cited by 7700K owners as causing higher than expected temps. Whether or not this is something in the programming or just that Overwatch is quite popular is another matter. I would not be terribly concerned about momentary spikes. With a program like AIDA in the screen above, you might be able to get a better understanding of when these occur and under what circumstances. There are other programs that can do the same.

[ritchiedrama]

Yes, this is what setting the adaptive voltage and EIST will do. You can also set adaptive voltage without EIST. If you still enable basic C1/C1E c-states, the cores will drop down momentarily at idle saving you some heat, although not nearly as much with speedstep enabled. It kind of pings back and forth between the minimum (1-1.2GHz probably) to 5.0. With both enabled, it will run flat at the minimum while on the desktop then perk up when a task is assigned.

 

These two graphs are old and from a slightly different discussion, but it does illustrate what you are looking for.

 

Here is Adaptive, C-states C0/C1 limit, and EIST while at the desktop. You can see a core fire off every now and then when some task is assigned. The blue line is my cache frequency. You can ignore that and it will not completely flatten out without some deeper c-state packages.

 

These are the same settings while playing a CPU limited MMO. The first third is the pre-event or normal gameplay. It is not a challenging program to run, so two of my 6 cores are assigned as workers. Those are the two at the 4.5GHz line. The other 4 cores are running background tasks, monitoring programs, or other game program related subroutines. They bounce between the inactive state (1.2GHz) and the 3.5GHz non-turbo maximum. Then the action starts and all 6 cores go into full throttle 4.5 to support the work. The one core doing meaningless windows stuff drops down occasionally, but if the load is sufficient, it stays up too. At the very end of the graph I go to the game main menu and then down to the desktop to take the picture. All other cores downshift, and the one core looking after the game (still running) stays active.

 

This last one is what happens when you enable the Performance Power Plan in Windows, which has a 100% processor up time. I am not actually doing anything, other than sitting on the desktop. Half way through, I turn it back to Balanced. This is similar what your processor will do when set to adaptive with no speed step or on manual/fixed with C1E enabled. However, it is preferable to use the Power Plan to create the 100% uptime state. First, it can be toggled on and off with two clicks from the desktop, while in game. No reboot required to test differences in performance. Secondly, there are some hidden parameters related to other CPU functions that are put on full alert in the Performance plan. I don't know them all and am not qualified to discuss how much real benefit they have, however disabling C-states/EIST does not increase the performance of those other areas.

 

I don't have Overwatch, so I don't know how to gauge the CPU behavior. It is not a difficult game to drive, but it is frequently cited by 7700K owners as causing higher than expected temps. Whether or not this is something in the programming or just that Overwatch is quite popular is another matter. I would not be terribly concerned about momentary spikes. With a program like AIDA in the screen above, you might be able to get a better understanding of when these occur and under what circumstances. There are other programs that can do the same.

 

There was a setting in my BIOS which someone said to change.

 

CPU PLL OC Voltage: [Auto] - so I set it to 1.150 and my temps dropped 10-15c

 

I can't find any solid info as to why, some people say it makes the sensors read less, than the actual temps (why would that even exist? sounds wrong to me).

[ritchiedrama]

***POSTING THIS AS ITS MY COPY PASTE FROM ANOTHER FORUM, WITH MORE DETAIL***

 

Hello!

 

New to this forum but I've been trying to learn/find an answer all day about my KabyLake 7700K.

 

First I'll list my specs, just so everything is known:

 

CPU: 7700K @ 5GHz w/ 1.328 vCore

MOBO: MSI Z270 Gaming Pro Carbon

GPU: 970 GTX (upgrading, soon).

MEM: Corsair Vengeance LPX DDR4 3200MHz

COOLING: Corsair H100i V2 w/ pre-applied thermal paste

 

http://www.userbenchmark.com/UserRun/3051401

 

 

 

Ok, so firstly these are the only things I have changed in BIOS:

 

CPU Clock: 50 (5GHZ)

EIST: Disabled

Extreme Memory Profile: Enabled

CPU Core Voltage: 1.325 (1.328V displaying in BIOS) - http://i.imgur.com/3W2CL4z.jpg

CPU SA Voltage: 1.200 (1.208V displaying in BIOS as shown in the image after Core Voltage) - This value was set to AUTO previously, and over 1.3V

CPU IO Voltage: 1.150 (1.160V displaying in BIOS as shown in the image after Core Voltage) - This value was set to AUTO previously, and over 1.3V

 

Now - originally that is all I had changed, and my temps were a little too high for my liking though.

 

So I reached out for help on a different forum called overclock, and a user recommended I change this setting below:

 

CPU PLL OC Voltage - It was set to AUTO, and he told me to change it to 1.150. My motherboard says default is 1.2, and I can use anywhere from 0.6 to 1.5 (that's what it says on the "HELP" on the right hand side).

 

After changing this setting, my temps dropped 10-15c. Now when gaming they only hit 65 maximum.

 

I was then told by two other users of the same forum that it is just the dts sensors reporting FALSE readings, or so they believe. Now, I find this strange as I can't imagine any bios setting would do such a thing without ANY documentation anywhere on the internet (I've been looking all day and seeking as much help as possible).

 

Now, my overclock is stable, or at least it is stable right now. With these current settings, I only ran a few stresses, x264 for over an hour, no crashes, gaming all day with no crashes. Obviously it could crash tomorrow, next week and so on but for now, we're good.

 

I am concerned however about these claims of it reporting FALSE temperatures since changing that setting off AUTO to 1.150.

 

Help would be very much appreciated, and/or some understanding of this setting. The only information I have found on it so far (which is old, and possibly false, also is here):

 

"CPU PLL Voltage Override (Overvoltage): What the Heck does it do?

So I asked that question to an Intel Overclocking Engineer his explanation was roughly: We went through the BIOS settings trying to find setting that if changed could help overclock our CPUs further. We came across this setting. Think of the CPU PLL voltage as a voltage that is provided to the CPU, but then “clipped” down to an approximate voltage. No matter what that input is whether 1.3v or 1.9v it is clipped (hypothetically let’s say 800mv after clipping (he didn’t say how much)) that way other devices can use the PLL voltage and clip to what they need. The CPU PLL Overvoltage allows for less clipping of that voltage. It can also reduce the lifespan of the CPU, but nothing noticeable."

 

"CPU PLL Voltage: Sets the voltage for the internal phase locked loop. The role of the PLL is to ensure that the output clock of the internal processor frequency synthesizers maintains phase coherency with the reference clock signal (supplied from a clock generator located within the PCH). "

 

 

Thanks for any help;

 

Richard

[c-attack]

Unfortunately, that something I have not tinkered with before. On Asus, I am accustomed to it being represented as way to change the oscillation to match an unusual base clock setting. This is your second definition. I don't know how to interpret it as a voltage and the Z170/270 may have new tricks up its sleeve. A 10-15C reduction sounds quite nice, but this generally an area you don't tweak unless you are pursuing radical overclocks. Although, in those cases you might be adding rather than subtracting. The adjustment is small and if you can run handbrake for an hour it is not completely topsy-turvy. I guess at this point proceed with caution until more information becomes available.

 

Are you still on fixed/manual voltage?

[ritchiedrama]

Unfortunately, that something I have not tinkered with before. On Asus, I am accustomed to it being represented as way to change the oscillation to match an unusual base clock setting. This is your second definition. I don't know how to interpret it as a voltage and the Z170/270 may have new tricks up its sleeve. A 10-15C reduction sounds quite nice, but this generally an area you don't tweak unless you are pursuing radical overclocks. Although, in those cases you might be adding rather than subtracting. The adjustment is small and if you can run handbrake for an hour it is not completely topsy-turvy. I guess at this point proceed with caution until more information becomes available.

 

Are you still on fixed/manual voltage?

 

 

Yes, I have been on 5ghz 1.328 voltage all day (it boosts to 1.336 though when I use prime95 (old version before bad updates) and aida64.

 

I've ran x264 (for an.hour) and prime95 aida64 and games all day and not crashed (yet).

 

My BIOS says default CPU PLL is 1.2 - but I manually changed it off auto to 1.2 and temps dropped so I think AUTO setting from MSI was trying to reach too high because of the overclock (it was doing the same in the others I mentioned in the previous post as it listed the values for those for some reason).

 

The only concern is just that some people said it could be the sensors are changed when the setting is changed but they provided zero.evidence all day when I asked.

 

I have been using 1.150 CPU PLL all day.

 

I cannot get speedstep working though even when I enable it in BIOS and set minimum processor state in windows plan to 50%

 

Thanks for your replies.

[c-attack]

The little overshoot on the Vcore target is normal and necessary. Don't worry about that. It is also OK if you leave in on manual/fixed voltage while you are still running tests at a frequent rate. The adaptive voltage setting can be a little more elastic when running these synthetic stress applications. Once you are satisfied, you can turn that into an adaptive voltage that will drop frequency and voltage when not tasked. As long as your are on manual voltage, it will not drop both frequency and voltage, thus speedstep/EIST will not work as expected. I think you are fine to switch to adaptive and stop testing, but ultimately that is your decision.

 

I don't write BIOS programming, but I cannot imagine any way the temp sensors would connected to PLL voltage. There is no connection between those two elements. Part of the discussion you cited was in reference to the Sandy Bridge platform. I don't know if those factors still hold true for this one.

[ritchiedrama]

The little overshoot on the Vcore target is normal and necessary. Don't worry about that. It is also OK if you leave in on manual/fixed voltage while you are still running tests at a frequent rate. The adaptive voltage setting can be a little more elastic when running these synthetic stress applications. Once you are satisfied, you can turn that into an adaptive voltage that will drop frequency and voltage when not tasked. As long as your are on manual voltage, it will not drop both frequency and voltage, thus speedstep/EIST will not work as expected. I think you are fine to switch to adaptive and stop testing, but ultimately that is your decision.

 

I don't write BIOS programming, but I cannot imagine any way the temp sensors would connected to PLL voltage. There is no connection between those two elements. Part of the discussion you cited was in reference to the Sandy Bridge platform. I don't know if those factors still hold true for this one.

 

 

For now the only thing I care about is temps.

 

Since I changed to 1.150 CPU PLL OC Voltage I've not had a crash once, and my temps haven't gone over 69c when gaming, usually sit around 45-60 somewhere depending on the game.

 

My vent at the top blows out warm air (I assume that's a good indication, rather than bad) but the air is never *hot*.

 

I need to get to the bottom of the CPU PLL OC setting and see if it really can change the sensors readings. I agree with you, I really doubt it can, but my concern is because of another forum where they said it can, yet I've asked for proof for 2 days now and received none.