1 Problem, 1 Concern with the H100i v2

[Daykeem]

Problem: After a short while gaming, the H100i v2 will stop responding to the fan curves I set. It'll set itself to about 1360 RPM and stay there no matter what I do. Ending the iCue process doesn't do anything. It was not doing this when I managed it using the old Link software. FYI, in my BIOS, the CPU fan header is set to 100% until iCue takes over at login. Only a full computer restart fixes the issue. I'm 100% positive I'm using the program as intended as I use the same program to set my keyboard and mouse profiles without issue. My setup works the way I want it to up until the point when software decides it doesn't want to.

 

FYI I do use these programs to manage the other fans in my system: EVGA's precision XOC for the GPU (no LED control) and SpeedFan for the chasis fans.

 

Any ideas?

 

Concern: i understand that this cooler has its own temperature sensor. It's been my experience that the cooler's sensor reports temperature 30 degrees Celsius cooler than the Package temperature reported by other hardware monitoring tools I use; for example, the H100i v2 will report 30 degrees Celsius while every hardware monitoring tool I use will report 60 degrees Celsius on the Package. I use the package temperature to determine my fan curve, because I know better, but I'm curious what the point of the H100i v2's sensor is for considering this very large discrepancy.

 

Edit: After doing some more reading it seems the temp that the h100i v2 reports is the 'coolant tempature'. So is it better to set the curve based on the coolant or the CPU package?

 

Thanks.

Edited October 1, 2018 by Daykeem

[c-attack]

Coolant temperature. This is how the process works. All of the heat transfer out of the cpu is conductive. No fan or pump speed settings affect that rate. It is all about cpu design, material, and voltage. The cooler is the heat disposal part of the process. Heat is dumped into the loop. Radiator needs to expel it at the same rate or coolant temperature will increase. If coolant temp does not increase, all heat is being removed from the loop. When it is not and the temp rises, the extra heat is transferred back across the cold plate to the cpu. Coolant temp is minimum cpu temp. Actual cpu temp is coolant temp + voltage based heat, determined by Vcore load. No cooler or fan setting will make up for 1.50 Vcore. The heat must always pass through the cpu first.

[Daykeem]

Thanks for taking the time to reply, but your answer doesn't quite answer my question or concern.

[DevBiker]

Thanks for taking the time to reply, but your answer doesn't quite answer my question or concern.

 

How not? That was your question - is it better to use coolant or CPU temperature. Coolant ... and c-attack explained why.

 

Also, when you set it to the coolant temperature, you can save the fan curve to the device. Yes, that's right ... Link and iCue won't be needed for your fan curve.

[c-attack]

After doing some more reading it seems the temp that the h100i v2 reports is the 'coolant tempature'. So is it better to set the curve based on the coolant or the CPU package?[/i]

 

Coolant temperature. This is how the process works...

 

 

+1C coolant temperature = +1C CPU temperature. The coolant temperature represents the minimum possible CPU temp. As coolant temperature increases, so does the CPU temp baseline.

 

Coolant temperature + voltage based temperature = Final CPU temperature. Sitting on your desktop in a high C-state with Vcore 0.00 (not really possible these days) your CPU temp equals the coolant temperature. Then you create a task that requires full core commitment and the CPU temp jumps +30-50C as the Vcore goes from 0 to 1.xx volts. That is the voltage based temp increase and determined by CPU make, design, and mostly voltage. The cooler cannot help with this and you will get the same additive with the same CPU and Vcore, no matter if you have a tiny 120mm radiator or a giant 2 meter external panel.

 

Since the coolant temperature change part of the equation is the smaller additive, you don't need to to make steep or aggressive fan curves for most usage scenarios. A gentle upward slope that meets your noise requirements will be effective without being loud. For most people, blasting the fans at 2000 rpm to reduce their CPU temp by 2C is a bad trade, but of course you can decide this for yourself.

[Daykeem]

I believe I understand now. Thanks for taking the time to explain again. I just needed some more education on voltage to understand you better.

 

Do either of you have a fan curve suggestions for a user who wants to keep their package temps under 80 degrees while basing their fan curve on the coolant temp?

 

Also, how do we know that our fan curve was saved on the device using the coolant temp?

[c-attack]

Saving a curve to the device (the H100i v2) can only be done with coolant temperature as the control variable. That can be read by the cooler without software assistance. For anything else, you must have Link or iCUE open to make the fan curve work with other variables. This also means if you create a curve for CPU temp with higher speeds in the 60-80C range, it will probably result in very low fans speeds if saved to the cooler and the same scale will be used with coolant temp and its much smaller range (20-40C).

 

Generally, on a H100i v2 I would expect a coolant temp rise in the +6-8C range when at 100% load and no GPU heat present to additionally warm the case (that last part is often significant when gaming or any combo type load). We would need to know your CPU model and settings to be more specific. Regardless, that also means unless your current package temp is right at 80C, blasting the fans won't do much for you. With the coolant temp part abound 6-8C and the voltage part typically +30-50C, it is quite obvious where the bulk of heat comes from. The only way to address that is with voltage, BIOS, and power level settings. If you are on a 6700-8700K CPU, there are quire a few things you can (must) do to tame this down. Frankly anyone running Auto voltage on a Kaby or Coffee Lake CPU is in for a rude awakening the first time they fire up a stress test. Fortunately, a lot of this can be tuned out.

 

As for creating the coolant based fan curve, power on your PC and let it sit for 15 minutes, web browsing or other light work. This is your baseline and you should set a quiet comfortable fan speed to match. About 3-5C down the line, start to slope the curve up. At +6-8C, you will want a moderate speed that you can stand when using the PC. For 120mm fans, that is likely in the 1300-1500 rpm range, but every fan/case/ear combination is unique, so decide for yourself. Small changes in fan speed do not make a difference, so if 1400 rpm is clearly less irritating than 1500 rpm, set it lower. Finally you can set a high speed blast for somewhere near 50C coolant temp. This will give you an auditory warning if something goes wrong. You should not make it to 50C coolant temp without considerable additional heat sources. The real monkey wrench in all of this is room temperature. If you have a normal room that for instance gets a lot of afternoon sun and warms up a lot, it can throw your curve off. In Summer, my office and will warm from 23 to 28C, no matter what I do with the AC. I have to build that into my fan curve and the +5C rise in room temp is equal to the typical rise I would get from m 8700K at full blast.

[Daykeem]

Excellent advice. This is what I have to work with:

- i7-7700k, 4.7Ghz, 1.248V Stable

- GTX 1080 ti

- 22.5 degrees C ambient

 

I left my computer idle for 30 minutes with Chrome open while running the minimum 420RPM with two 120mm fans. Under these conditions, the max coolant temp was 39 degrees C. Therefore, my coolant fan curve should be the following based on your advice: 420RPM/39degrees to 1560RPM/47degrees (+8).

 

I've adjusted my fan curve accordingly, as shown below. I do prefer a linear curve. Does this look right?

 

[c-attack]

Oooh... 39C idle in a 22C room? That is considerably higher than expected. 39C is about what you should see at max under load. Can you set the fans to a fixed speed of 1300 or so at idle and see if the coolant temp changes? Have you been having temperature issues at load?

 

*Can you also detail your case layout? Case model, H100i is top mounted as exhaust, etc. Is there a dust filter on the radiator?

Edited October 7, 2018 by c-attack

[Daykeem]

I don't believe I'm having temperature issues at load -- when the iCUE software isn't malfunctioning. Sometimes my CPU package will spike to 80 degrees while gaming but I was told this is normal. However, the reason for this post was to address the fact that my cooler doesn't always respond to my iCUE software's package temperature fan curve. When this happens, I do have issues with +85C temps. I'm switching to coolant temperature because DevBiker said the curve is saved on the cooler instead of relying on the software. It may also be worth noting that my 2nd core is always the hottest core by about +3 to +5 degrees.

 

At a fixed 1320RPM, my idle temp is exactly 30.1C.

 

The easiest way to show you my build would be to refer you to my PC Part Picker list, if you don't mind.

 

My radiator is top mounted as exhaust using a pull configuration and there is a dust filter there. I don't have the option to front mount the radiator with the case I chose.

 

[DevBiker]

And idle temp of 30.1C is fine. Where did the idle temp of 39C come from? As c-attack mentioned, that's far too warm for the coolant - you should only see that coolant temp (if at all) at load and then only rarely and after heavy, sustained load.

 

Also, your GPU should be installed in the top slot to get the full 16 PCIe lanes rather than limiting it to 8 lanes in the second slot. It'll also get the heat exhausted quicker - it'll be sucked right out the rear almost immediately rather than dissipated throughout the case.

 

You also need to put some thought to the control variable for the case fans. The default for those is the package temperature but that's not a good indicator of the internal case temperature (since you are liquid cooled). As a result, they may not be spinning up fast enough when the internal case temperature rises - primary due to exhaust heat from the GPU. If the internal case temperature rises, this warm air will then get exhausted through the radiator ... which could actually wind up increasing the temperature of your coolant. PC Part Picker lists a 1080Ti for your build but that photo shows a 1060. I can say (from personal experience) that the 1080Ti can really put some heat into a system.

[c-attack]

OK, super low 120mm fan speed (400 rpm) + dust filter probably results in virtually no airflow. On the positive side, 700-800 rpm should be inaudible as well, but I suspect below that is going to be too slow for the massive restriction (radiator plus dust filter). Even at speed, the dust filter prevent airflow and increase the amount of retained heat in t he coolant stream. I typically recommend you not use it on the radiator unless you are in extreme environmental conditions. That decision is up to you.

 

The original method of developing a fan curve is still valid, but you need to find one appropriate for you actual desktop conditions. If you want to run super slow and take the heat at idle you can, but that also means you will surely go past the 80C package temp goal as your coolant/CPU is starting off at least +10C higher than it otherwise would. Experiment with and without the dust filter at acceptable speeds.

 

Most Z270/370 have some type of load line adjustment that makes adaptive voltage usable. I am not familiar with that ASRock board. The biggest gains will come from tightening your real use Vcore and CPU behavior, although the dust filter is a measurable impediment as well.

[Daykeem]

And idle temp of 30.1C is fine. Where did the idle temp of 39C come from? As c-attack mentioned, that's far too warm for the coolant - you should only see that coolant temp (if at all) at load and then only rarely and after heavy, sustained load.

 

I'm not sure what happened there... But after leaving it idle for about 1 hour without touching it, I got a new idle coolant temp of about 30.7C @ 480RPM as shown in the picture below. Maybe 30 minutes wasn't enough time. By the time I've finished writing this post (5 minutes), the coolant temp has increased to 33.3C @ 780RPM.

 

 

Also, your GPU should be installed in the top slot to get the full 16 PCIe lanes rather than limiting it to 8 lanes in the second slot. It'll also get the heat exhausted quicker - it'll be sucked right out the rear almost immediately rather than dissipated throughout the case.

 

I may have needlessly added to the confusion here. I upgraded from that old 1060 to a 1080 ti, and the 1080 ti is indeed installed on the top slot. Everything else is the same. It's just an older picture.

 

You also need to put some thought to the control variable for the case fans. The default for those is the package temperature but that's not a good indicator of the internal case temperature (since you are liquid cooled). As a result, they may not be spinning up fast enough when the internal case temperature rises - primary due to exhaust heat from the GPU. If the internal case temperature rises, this warm air will then get exhausted through the radiator ... which could actually wind up increasing the temperature of your coolant. PC Part Picker lists a 1080Ti for your build but that photo shows a 1060. I can say (from personal experience) that the 1080Ti can really put some heat into a system.

 

What do you use as your control variable for your chasis fans?

[Daykeem]

The original method of developing a fan curve is still valid, but you need to find one appropriate for you actual desktop conditions. If you want to run super slow and take the heat at idle you can, but that also means you will surely go past the 80C package temp goal as your coolant/CPU is starting off at least +10C higher than it otherwise would. Experiment with and without the dust filter at acceptable speeds.

 

Okay, so to sum everything up: coolant temperature of 30C is normal-ish at idle, and 40C is normal-ish under load. So I'll adjust my fan curve as follows: 30C/25%RPM and 40C/100%RPM. Based on what I've seen so far, this should keep my package below 80 under load. Once I've done this, as shown in the picture I posted above, I can quit iCUE and the cooler will remember the curve?

 

I'll remove the dust filter to give me some more headroom as per your recommendation.

 

If I find that I cannot keep the coolant below 40C at 100%(1560)RPM, does that mean my cooler is malfunctioning/must reapply thermal paste? My max ambient temperatures is about 22C no matter what time of day it is.

[c-attack]

That curve looks better and so does the 30C idle temp. I might even move the 100% fan point further down to 45 or 50C. There is nothing specifically bad about 40C, but with a mild overclock you likely will run into the 80C mark when above that base temperature. Our alarmed response was because a +18C level of the room temperature usually means to the cooler has a blockage and is malfunctioning. There are other possibilities, but questions need to asked. Most people run +4-7C idle above room temps and your CPU/cooler combo should go +6-7C further when under load. Ironically, you may see the highest temperatures when under mixed GPU/CPU load conditions as the Ti heats up the case.

 

You can find out your voltage additive is by running a quick stress test. Set the fans to a fixed speed (1300-1500), note the package temp and coolant temp, then initiate the test. CPU package should instantly spike then come down slightly. Over the next 10-20 seconds it should be steady. Subtract the initial coolant temperature from that number and that is the voltage part of the equation. The reason to do this is it will tell you exactly what coolant temp you can reach and still stay under 80C in the heaviest conditions. So if your voltage additive is +38C, then you know you can go up to 42C coolant temp before hitting your 80C limit.

 

Not sure what options you have in Speedfan for case fan control, but I would suggest either GPU temp (for the rear fan certainly) or the coolant temp if Speed fan can read it. This helps balance the different sets of fans and both are relatively slow to change variables reducing "fan ramping" up and down.

 

*Looking at pictures of the case, the top dust filter may not be too bad if there is some space between the fan and the screen. Close proximity has a more dramatic effect. Another reason to try with and without and see.

Edited October 7, 2018 by c-attack

[Daykeem]

... Ironically, you may see the highest temperatures when under mixed GPU/CPU load conditions as the Ti heats up the case.

 

I observed this today. I ran a Superposition stress test (GPU) and a Prime95 Small FFTs stress test (CPU) at the same time for about an hour. GPU maxed out at 60C and my CPU at 84C on the second core only (about 80C package). Coolant temp got up to 40.1C quickly, but didn't move from there.

 

You can find out your voltage additive is by running a quick stress test. Set the fans to a fixed speed (1300-1500), note the package temp and coolant temp, then initiate the test. CPU package should instantly spike then come down slightly. Over the next 10-20 seconds it should be steady. Subtract the initial coolant temperature from that number and that is the voltage part of the equation. The reason to do this is it will tell you exactly what coolant temp you can reach and still stay under 80C in the heaviest conditions. So if your voltage additive is +38C, then you know you can go up to 42C coolant temp before hitting your 80C limit.

 

I'll play around with this experiment after work :)

 

Not sure what options you have in Speedfan for case fan control, but I would suggest either GPU temp (for the rear fan certainly) or the coolant temp if Speed fan can read it. This helps balance the different sets of fans and both are relatively slow to change variables reducing "fan ramping" up and down.

 

SpeedFan can't detect my coolant temp, but I'll set the rear to GPU for sure.

 

Thanks a lot to the both of you for your input. I'm quite confident I know what's going on now.

 

Now that my coolant temp is the variable that determines fan speed, can I terminate the iCUE process and be confident the cooler will remember the curve?

Edited October 7, 2018 by Daykeem

[c-attack]

I observed this today. I ran a Superposition stress test (GPU) and a Prime95 Small FFTs stress test (CPU) at the same time for about an hour. GPU maxed out at 60C and my CPU at 84C on the second core only (about 80C package). Coolant temp got up to 40.1C quickly, but didn't move from there.

 

Good, that pretty much eliminates any possibility of a cooler malfunction. You would not have lasted long if the coolant wasn't flowing as it should.

 

 

 

Now that my coolant temp is the variable that determines fan speed, can I terminate the iCUE process and be confident the cooler will remember the curve?

 

I believe with the H100i v2 you will need to use iCUE to manually save it to the device. Go to the H100i v2 tab, then click the three horizontal line menu at the very top next to "Profiles". It should be 4th down.

[DevBiker]

I may have needlessly added to the confusion here. I upgraded from that old 1060 to a 1080 ti, and the 1080 ti is indeed installed on the top slot. Everything else is the same. It's just an older picture.

OK, cool.

 

 

What do you use as your control variable for your chasis fans?

I have a Commander Pro and use one of the CoPro temp sensors measuring the internal case temperature. And the 1080Ti upgrade that I recently did (from a 1070) had me redo the curves due to the amount of heat that it put into the system. With a large case and lots of fans, I am able to keep that heat away from the cooler but that may be something that you are going to have to manage. The 1080Ti is a 250W card, compared to 120W for the 1060. That's a lot of additional heat.

[Daykeem]

The image below is a perfect example of why I started this entire thread. Whenever I push my PC to the max, the iCUE software seems to hang or crash somewhere along the way. The dashboard is reporting that I'm currently under 100% load and about 80C avg CPU temp. This is obviously a lie as RealTemp shows to the right.

 

If I was still using the package as my control variable for fan speed, my radiator fans would be running at 1560 like iCUE is showing below, but I think saving my coolant temp fan curve to the device has fixed the issue.

 

This seems silly.

 

[DevBiker]

Yeah, this is a known issue and happens pretty specifically with certain load testing packages (not all). It's an issue of thread starvation; the threads that monitor the sensors are background threads at a lower priority and, therefore, when there are threads with higher priorities scheduled that consume the CPU (and some stress testing packages specifically set their thread priorities to Above Normal as opposed to Normal), the iCue threads don't get enough CPU cycles.

 

This type of behavior doesn't usually happen during normal usage - only stress tests and, even then, only specific ones.

 

If, however, you write the cooler's settings to the device (go to the cooler, click on the hamburger menu ... there's an icon that looks like and SD Disk ... that'll write your curve to the device), you'll be fine.