Jump to content
Corsair Community

Corsair H115i AIO on a i7-6850k @ 4.0GHZ


Recommended Posts

Hey folks,


this is my system for reference: http://www.modsrigs.com/detail.aspx?BuildID=36010


I decided to do a prime95 torture test and got some interesting results. My CPU package and some cores were at over 90 degrees C within 10 minutes, my Corsair H115i was at 1800 RPM on both Fans and 3090 on the pump.


During the Prime95 test, Corsair Link was reporting the CPU and some cores at around 95C but the Corsair H115i is reporting it's temperature at around 45C.


NOTE: I am using the thermal compound that was pre-applied and provided by Corsair. I also made sure that the AIO is firmly mounted and all the mounting screws are very tight.


Furthermore my CPU package idles around 50-55c while individual cores are around 36-42c idle. My room temperature is 24-25c currently in a dry climate (Calgary, AB Canada). My system is perfectly clean and dust free. I have 2 intake fans and 1 exhaust fan that is not including the AIO. The AIO is set on a push exhaust configuration and mounted to the top of the case.


My Corsair H115i Fans that came with them were replaced after an accident lead to a blade breaking of one of them. The Original SL140 PWM fans were replaced with the ML140 PWM fans.


Do I have a problem? is there some kind of issue with my AIO or CPU?

Link to comment
Share on other sites

No, all of those values are exactly as expected.


1) Stop using Prime for any CPU from Haswell forward. It's like applying a blowtorch to one side of your hand and putting ice on the other and expecting things to balance out painlessly. Prime still has some value for binning chips or in select circumstances with custom runs, but it is not a good tool for cooler or system evaluation. It will get too hot at the CPU itself and the heat builds up inside the CPU faster than it can be transferred out through the CPU heat spreader.


2) CPU package on HW-E and BW-E - These CPUs have specific sensors for the "CPU" (somewhere in the lid), CPU package (socket), and of course individual cores. CPU package or socket temp is a weird one. I have owned 3 HW-E models and all of them have the same characteristics. Package temp is always going to be +10-15C over the cores or "CPU" temp at idle, particularly if you have basic SpeedStep/EIST and c-states enabled. It has no real value for me in this context, except as a maximum value. Asus' suggested peak is 80C. Cooler material and TIM should have no effect on the pin side, so you are effectively at your voltage limit at that mark. The reasons above make this a less than ideal variable for cooler control. Use the H115i coolant temperature.



3) H115i Temp (coolant temperature) - Temperature of the fluid as it passes through the cold plate and exits the pump housing. This is your default variable for radiator fan control and you should use this, although it takes some getting used to.


Think of the cooler as a two stage heat remover. Stage 1 is CPU lid to TIM to cold plate. This is straight material heat transference, non-adjustable, and a two way street. Heat goes both ways. If the fluid in the cooler is 45C, then the lowest "CPU temp" you can see is also going be right around 45C. After a stress test, you can sometimes watch the three CPU values trying to find equilibrium. The CPU Package temp will drop as soon as you cut the test and resume it's normal, voltage dominated value. The cores will drop and resume their usual pinging, often lower than the CPU lid temp. The CPU lid temp will drop down with the cores, but then rebound up to the same as the coolant temp. For a moment, the cores will read much cooler than the CPU temp which receives all the heat from the cores. Makes sense, except when you are running the full 100% load and your cores are pinging between 60-70C, but the "CPU temp" is only 50-55C. That is the result your heat being conducted away from the CPU lid, through the cold plate, and into Stage 2.


Stage 2 is the coolant stream. It's purpose is to transport the heat somewhere else. Pick it up at the cold plate, move it to the radiator, and with the help of the fans, release it out of the case. This is why fan speed should be linked to coolant temperature. It only affects that value and can do nothing for CPU package or cores directly. Their temps are all down to thermal transfer. However, as mentioned the cold plate is a two way transfer, so if you don't dump the heat that's in the coolant stream, it will add heat back into the CPU. Effectively, for every +1C rise in coolant temp, you will see a +1C rise in CPU temp. The cores are way to dynamic to every see this, but you can see it at the CPU temp lid sensor.




4) Fans and fan speed - Because of the above, you don't need the fans to react to every little blip and bob of the cores. You only need more fan speed when the stage 2 coolant stream is holding too much heat. A normal full AIDA64 blend test will have a coolant delta of about +6C on my 5930 clocked at 4.5@1.275 and 4.0 on the cache. That's about 205-215W. That's not huge delta and the 280mm radiator can easily handle that wattage. I generally cap my 140mm fans at 1350 rpm, even when benchmarking. It doesn't make much difference and I am not going to run the fans at 2000 rpm to reduce the CPU temp by 1C. That 6C delta is the theoretical maximum you can reduce your CPU temp, regardless of fan speed. You can't go lower than the original starting value. In reality, nothing is 100% efficient and you while you might see several degrees difference between 500 rpm and 1000, you don't get the same gains when doubling up higher. Fan speed is almost never the limiting factor in your CPU temps. The voltage and resulting socket temps are going to be the limiter on almost modern CPU.


Where things get complicated is the CPU is not the only heat source in your case. The GPU is by far the largest and when loaded it will affect case temperature, which in turn affects everything else. If you heat the case up to 45C, the lowest possible coolant temp inside the radiator will also be 45C, before you add any CPU heat to the mix. For this reason, you are likely to see higher coolant temp deltas when gaming or GPU load rendering than on a straight 100% CPU test. This in turn will make the fans run faster. You can look at this in two ways. 1) Good - more heat in the case means I need more exhaust. My radiator fans exhaust and this is helping. 2) Annoying - now it's too loud! That's fine, set the maximum speed you can tolerate to a few degrees above your normal peak value. If you get out of range, you will hear the effect likely before you notice on the monitor. Remember, the fans don't have to react to CPU temp. A moderate and steady speed will always do the job. No one's CPU ever overheats because the fan speed was just a little too low. We are talking about coolant deltas in the 6-10C range, so you would need to already be in the 70's on the CPU temp before this becomes relevant in crossing the 80C barrier. That generally only happens in stress tests, but if you do long encoding runs you may get there as well.


Nothing wrong with the ML140's. That is an excellent replacement for the stock fans, which are not popular with everyone. Call that a happy accident. You can experiment around and find your sweet spot for fan noise versus cooling. The ML140s are quite reasonable at 1000 or under and that should do for most anything except long, constant workloads like rendering, mining, etc. No need for 2000 rpm, even then. You likely will want to make your own fan curves. The presets are based on a general 20-23C room temp, but because case and room environments vary greatly, there is no way to make a perfect curve for everyone. Someone in a tropical location would be running the fans at max all the time, even when the coolant delta was zero. During your Winter, you can probably put in on anything an not notice the fans, but case airflow matters too and ultimately it is the case temp that dictates a lot of the end coolant temperature. In a micro-box, you are going to be hot all the time. In a 900D suitcase, it may be almost a complete non-factor.





If you want to run a stress test to check cooler function, try something with a steady load for longer duration. Linpack, AIDA or even something mild like XTU for long duration can give a better idea of how the cooling system works without making the CPU and voltage the limiting factor. In reality, your best tests are doing what you normally do and being happy with the results. Stability testing is different, although you certainly can do both at once. AIDA64 and OCCT work well. Even the Asus Real Bench has some value as a general stability test, less so for cooler function.


The only thing I noticed is your core temps seem a touch higher than expected for your room temperature. Usually this is down to one of a few things. 1) You have C-states/EIST off and the cores are always fully active; 2) Windows power plan is set to performance with 100% up time and the cores are always active; 3) Your case has some type of restriction at the radiator exhaust point that slows down the heat release. Could be a dust filter or even very thick mesh. In this last instance, the coolant temp would also be a bit higher. Normally, I would expect to see your H115i temp float 4-7C above the room temp. If it's 22C and your coolant temp is 36C, then there are some questions to be asked - mostly about case flow. Compare the coolant temp to other case temp values (non-voltage affected), like the motherboard temp sensor, drive temps, etc.

Link to comment
Share on other sites

Those temperatures are about right for how high your ambient room temperature is. I keep my place around 68F/20C. My Package Idle is 40c, w/my cores between 31-34c. My package temperature is often 10c hire then my lowest core, if I have any background tasks going during idle. With my H115i is Liquid at 30-32c. I'm running my 6850k @ 4.2ghz, @ 1.3v. My 1080ti idles at 28C and peaks at 50C w/fans topping out at around 1100rpm.


I have 3 Noctua F-12 for intake, with another F-12 and EVGA Hybrid kit running Push/Pull off the middle fan for my 1080ti, w/3 Noctua NF-A14, 1 running exhaust and 2 running pull config for the top mounted h115i.


Side view Front View

Link to comment
Share on other sites


This topic is now archived and is closed to further replies.

  • Create New...