H115i higher than expected temps

I just bought am H115i. I'm looking for help or opinions on whether these temperatures are expected, or if I need to RMA the H115i, or my CPU.

 

My idle temps are roughly the same as when I had a Hyper 212+. They're bouncing between 31-50c, but usually in the 35-40c range. Again, this is while idle. With the Hyper 212+, on load, my computer would actually shut down due to temps. So far, the H115i seems to prevent that, so that's good. While just idling, I have tried setting the fan speed anywhere from balanced to max, and it does not affect the temperature of the CPU.

 

My system is not overclocked. In fact, I undervolted the CPU a little to try to help my temperatures.

 

Extra information that may help: I previously had to RMA my 4790k, as it was crashing my system, unrelated to temperatures-my replacement one no longer does that. The temperatures for my first one were fine, and did not have such a high fluctuation in idle temperatures; with the Hyper 212+, it sat in the low 30's +/- 5c (no random 20c jumps).

 

My replacement 4790k is nowhere near that consistent, which is why I bought the H115i. Nothing is suddenly increasing load on the CPU when it jumps up in temperature on idle. I don't even want to try to overclock it. I did a test with a full load on the CPU, for 10 minutes, and it was holding at 77-83c with the H115i on max (it didn't turn off though! Better than with just the Hyper 212+!). Room temperature is 22-24c.

 

So the questions are: Do I have unreasonable expectations? Should I be expecting more from the H115i and should RMA it, or should I RMA my CPU again, this time due to excessive temperatures? I'm pretty sure what I'm experiencing is not normal. These temperatures are more like what I'd expect with a moderate overclock, not an undervolted CPU.

 

Thanks!

The key to making sense out the numbers is your water temperature (H115i Temp - last box in Link main display) as well as your room temperature. Comparing the two usually can help us identify if there is a cooler problem, a physical contact problem, environment issues, or just plain old voltage settings. The water temperature should be relatively steady and is usually 3-7C over your room temp at idle, case and environment dependent. Even at full load, it should slowly increase -- no jumps or spikes.

 

CPU core temps are something different. The voltage is applied and removed in milliseconds and there are a lot of reasons for their behavior. Windows 10 in particular is quite "jumpy" at the desktop. You need to distinguish between 1 core spiking in temperature (1 core loaded and doing work) and all 4 four cores jumping up and down or all over the place.

 

What stress test did you run? Not all 100% loads are created equal and 77-83C may not be so bad at the auto setting.

Water temperature is just as you said, roughly 5-6c over room temperature. The water temperatures do not suddenly spike in temperatures even under load, it's gradual, again as you said.

 

While idle, it's usually 1-2 cores jumping up the extra 20c randomly and briefly-not all of them at once. If this is normal, I'd really love to know that. All I know is that I did not notice my previous 4790k doing this, and I was closely watching it for awhile, as I suspected the issue was overheating (it was not). That CPU actually was defective, though.

 

Under load of the stress test, they're all, usually, within 5c of each other.

 

I set the voltage on my CPU down to 1.193v. Stock was just under 1.3v, if I recall correctly. This made no noticeable difference on temperatures.

 

For the stress test, I used Prime 95 (version 28.9, build 2), and ran the torture test "Small FFTs" for around 10 minutes, default 8 threads. I had no interest in anything except CPU temperature, so my (brief) research made this option seem to be a decent one. Do you recommend something different?

 

Really, I'm mostly concerned if I'll be able to overclock this system. I've had this system, in some form, for somewhere in the 1.5-2 year range, and I've just had problem after problem. This is the last (just cursed myself) potential issue.

Small FFT prime95 is quite a load and well above most other stress tests out there. If you have a practical need to run that kind of test, then it may be of some value. However, for most people it puts an extreme load on the CPU side of the system and thus makes for a lousy cooling system diagnostic tool. Remember, the heat always has to pass through the CPU to get to the cold plate and water. You can't bypass it. Extreme voltage and crazy instructions can cause the heat to accumulate in the CPU shell faster than the materials will allow heat to transfer out.

 

Strictly from a cooling diagnostic frame, I like Intel XTU. It offers a fairly low 100% load compared to Prime so people with cooling issues can run it without going into thermal shutdown 1.5 seconds in. It also has a lower core temperature variability compared to other tests like AIDA or OCCT. The line graph will look like a very smooth sine wave when all is well and stray or misbehaving cores are easy to pick out. You will need to configure the settings to show cores 0-3 individually. Package temps or "CPU temp" don't always tell the story and can be misleading.

 

Give this a run and see how it turns out. I am expecting a massive difference in end CPU core temperatures compared to Prime. Also, 5-10 minutes is plenty for this purpose. This isn't a stability run.

The temperatures are much lower with Intel XTU, so I feel better about that. I wouldn't call the graph super smooth, but there are no giant spikes.

 

Assuming I'm understanding everything correctly, I'm feeling a lot better now about the temperatures, even though the idle temperatures are about the same as the Hyper 212+. This test makes it look like I have a lot more headroom for overclocking than it looked like before (none). A couple games were shutting me down due to heat before, and that's not happening now.

 

I'll attach a screenshot of the test, towards the very end of a 5 minute run, just in case it should be a smoother line than it is.

 

Thank you so much for your help!

 

http://i.imgur.com/hcsRQYs.jpg

 

http://i.imgur.com/S5wL4uJ.jpg

Oh, I wanted to you to show the individual cores in the line graph display. The 4 of them would make the smooth wave pattern. Package temp can be a strange variable. Other programs like AIDA and OCCT give large core temps swings every second. XTU is a less variable. Anyway, there is no need now. Clearly all 4 cores are in the 50's where I would expect. Everything is set-up and functioning properly. Green light for overclocking, unless you actually intend to use Prime95 to hunt Mersenne numbers.

 

On that note, I would leave Prime95 behind. It still has some uses (beyond the original) like stability tests for specific values, but not as a general stability or heat test tool. It is even worse on the newer CPUs. I need an extra 0.11v beyond every other test out there to make Prime95 run on HW-E. XTU is good for this cooling tests. I like AIDA for general purpose and stability testing. OCCT is a definite step up in stability difficulty. There are others too, but Prime95 should be for select purposes only.

Great information, very much appreciated!

 

I won't be doing any overclocking like that, so I'll skip Prime95 next time. I generally go for fairly modest OC's, basically just used to drag out good performance and delay building new systems. So nothing I'm even hugely concerned about yet, but I know I will be, so I wanted to deal with it now.

 

Overheating on a different cooler, then still seeing similar idle temps and fluctuations with the new cooler made me nervous. Already having one bad CPU when that almost never, ever happens, made me paranoid too, lol. I'm completely comfortable with it now, though!

 

Thank you! I feel much better about it all now, and I've learned a lot.

Yes, having bad turn on a CPU can really throw you for a while. It is a very tedious troubleshooting process and "bad CPU" is always the last box on the checklist.

 

Going from air to water, there are a couple of things to keep in mind.

 

1) Your CPU cores temps can never be below the water temperature. Your water temp can never be below the room temperature. This is the same with an air tower, but it's importance is in the next rule.

 

2) Fans control water temperature, not CPU core temps. Your core temps will be the result of the voltage applied, the CPU instructions, minus the heat transferred through the CPU lid/TIM/cold plate into the water stream. That water stream is the holding tank for excess heat until you can expel through the radiator fins. That 'holding tank' and the nature of water gives these systems an advantage over air. You don't need the fans to react to constantly changing CPU loads. The thermal transfer rate into the water stream is constant. The water temperature is your basis or starting point for all CPU core temps, with the final result dependent on load/instruction.

 

If you run XTU at 30C water temperature and you get back core temps of 55-60C, in order to lower core temperature by 5C, you must lower the water temperature by 5C. That's not always possible. If you run long enough at high loads, the water temperature will increase. However, a 2C increase in water temperature only results in a +2C increase to core temperature at that same load. All of which is an elaborate way of explaining, you don't need to blast your fans. The best you can do is return the water temperature to its warm idle starting point. So if your starting water temp was 30C and it goes to 34C while gaming, the most you can reduce your core temps, with any fan speed, is 4C. That is usually not worth the noise. You can keep the fans low nearly all the time and you should never need speeds of more than 800-1000 for sub-maximal activities like gaming.

77-80°C sounds about right for a 4790k running Prime95 under water. My lowly 6600k runs about that hot with my H115i if I use the latest Prime95.

 

The reason P95 heats those CPU's up so much is because it uses AVX instructions and Intel CPU's automatically increase the voltage to handle AVX. As had been suggested already, there's other stress tests you can use that do not utilize AVX. These will probably give you a better approximation as to the temps you can expect when gaming, etc.

 

TL;DR -- 80°C is not a worry when using Prime95. TJMax is like 100°C on most Intel CPU's anyway. As long as your water temp is about 5-10°C above ambient at idle, you're good.

Awesome, thank you. Additional confirmation that everything's fine is always appreciated!