H115i with i7 7700K - CPU overheatting

Hello,

I am new to liquid cooling. Installed my new H115i with a lot of expectation and excitement. But the excitement did not last too long since the cooler performed worse than my previous air cooler.

 

The cooler temperature does not increase with the CPU temperature. With stock frequency the CPU goes around 65 c under load. But H115i stays close to room temperature of 22.

 

When I try to over clock the cpu to 4.8, the cpu temp goes to 100 instantly under load, but again cooler temp stays under 30.

 

I read about the loose backplate issue and added washer to the back to make it fit better. But the results did not change.

 

Any suggestion to improve the cooler performance will be greatly appreciated.

Did you mount the cooler on to CPU equally ?Maybe you installed wrongly somehow.Check all connections

I did mount the cooler properly. I am new to liquid cooling not PC building. The connection are simple. Single wire pump power goes to the CPU Fan header on the motherboard, The 2 wire power connector goes to the Sata power from power supply and the fans connect to the fan header from the pump unit. The last is the usb 2.0 cable.

 

I tried to change the pump speed in Corsair link software. I changed the speed from 1980 rpm (quiet) to 2940 rpm (performance), but i do not hear and pump noise from the cooler head. I did change the UEFI setup to keep the FAN on to have constant supply for the pump. How do I make sure the pump is actually running and is not broken?

How do I make sure the pump is actually running and is not broken?

 

By comparing the H115i Temp (coolant temp) to both local environmental and CPU temperatures. In your case, the pump is not broken. When this is the case, that H115i temp will start climbing the moment you hit power and slowly build up into the 50C range and beyond.

 

You mentioned at the stock frequency the CPU temp goes to 65C, but the H115i Temp did not change. Can you clarify if the H115i Temp never changes or is very slow to change? The later is normal and represents the amount the energy it takes to raise the volume of water by 1C while the radiator is simultaneously releasing heat. We don't have a lot of data on the new 7700's, but the TDP appears to fairly low, even at full tilt. As such, you won't see a lot of change to the coolant temperature and the limitation in temperature will be voltage and CPU design, as it usually is.

 

However, if the H115i Temp does not change at all, that would suggest the heat is not transferring through the cold plate into the system. You shouldn't need washers and those posts are related to some specific older motherboards. Keep in mind, it might take a minute for the H115i temp to start climbing. Neither it nor the fans on the radiator will instantly react to dynamic load.

 

Your 65C core temps at the stock settings seem to be in line with published results (see here). If you had a contact issue between the cold plate and CPU lid, it would be evident there as well as rather tall, spiky CPU temps for ordinary activities like web browsing, opening programs, etc. Before you drive yourself batty fiddling with the silly back plate, take a second look at your overclock settings to make sure they were within bounds. There may indeed be a contact issue, but given the tedious nature of trying to troubleshoot it, you might want to cross off some other possibilities along the way.

Thanks for your detail reply.

 

The cooler temperature does go up very slowly when doing load test, so the pump is working.

 

My only concern is why the CPU temp spikes on load and the water block does not absorb the heat more efficiently.

 

I will check the vcore and try to manually reduce it and see if that makes any difference.

That heat transfer from pins to CPU to cold plate is going to be the limitation on most every recent Intel CPU. This is not a controllable feature and the heat transfer is going to be dictated by the materials used in the CPU and cold plate. The contact metal is about the same in every after market cooler, until you start getting into really exotic elements with even more exotic prices. However, you certainly do want to determine if you are not getting the full contact between pieces.

 

One way to do that is to take of the pump and have a look at the TIM. It should be evenly spread and thin at this point. If there are solid patches still present, then you've found the issue.

 

A more passive approach would be to compare your CPU temps for basic non-maximal loads. With a contact issue, you usually get very spiky CPU temperatures for basic tasks, like opening browsers or other programs. Usually 2 or more cores will pop off on that simple task up to that 65C level or you may have more than one core running abnormally high with basic loads.

 

The 65C appears to compare with early 7700K results, but certainly the 100C is way off for 4.8 and we know 4.8 is attainable. I don't know if this was voltage related or the contact issue, but it could also be program related as well. What did you use to apply load?

What are you using to test load? Tests that use AVX will run hotter than those that don't. That said, your stock temps are about on par with my experiences with my 7700K and the H100.

More concerning is the overclocked temp. But there aren't enough details (vCore, type of test) to determine if what you are seeing is normal or if it's unusual. And most of the "auto" settings on the motherboards (at least, on my Asus) seem to crank the vCore up higher than it really needs to be, even according to their own guides. (see http://edgeup.asus.com/2017/01/31/kaby-lake-overclocking-guide/ ... 5Ghz/1.32 vCore) Also, what kind of TIM did you use and *how much*? Kaby Lakes seem to have very good contact with the plate so anything more than just a tiny bit will hurt your heat transfer.

I used cinebench to test the load and the CUP temp goes upto 85 to 90 with the CUP overclocked to 4.8. But with Intel burn test the cup temp jumps to 100 within couple of seconds.

 

I am using Arctic Silver 5 and used the pea application method. I also removed the cooler and confirmed that the Thermal Compound spreaded evenly.

 

I have to do a little more learning on the vcore setting. I have a Gigabyte board and I an just changing the multiplier in the UFEI. But I can see the vcore jumps to 1.4 under load which is causing the high temperature spikes. I will try to manually adjust the vcore and see if I can get a stable overclock at lower voltage.

OK, I think we have our 1 + 2 problem. First, don't use Intel Burn Test for anything. I am not sure what is proves anymore and the load is not indicative of any normal program. If you are trying to see how much heat the cooling system can handle versus how much heat the CPU can endure, use a lower intensity test for a longer duration. I like Intel XTU for diagnosing cooling problems since it won't fry you chip if you have poor contact. For general stability and routine benchmarks, AIDA64 and the Asus Real Bench have a lot more utility.

 

There is nothing wrong with running Cinebench, but I think you found your other issue. You can't leave Skylake and Kaby Lake processors on Auto voltage and overclock. This was always generally true of prior models, but those two really pile on volts with any kind of synthetic load. The "adaptive" part of the AUTO voltage is a little too compliant. To solve this, you only need to set a specific value and you can still set it as "adaptive" so it will throttle down at idle. If you are going to do a bunch of serious synthetic benchmarks, set the Vcore to a fixed value and then put it back on adaptive when you are finished. You should be able to get a general target voltage from some of the 7700K overclock guides out there.

I saw you wrote you have a Gigabyte board. I'm in the same situation as you, although i admit, my temps are way cooler than yours. One of the first things you HAVE to do (especially if you own a Z270x board) is to flash the newest BIOS version available (IIRC it should be F4e or something). That takes care of the widespread issue with the BIOS adding too much voltage when the CPU goes into TURBO.

 

After you flash (if you dont already have it) the newest BIOS, i recommend tinkering with the VCORE a bit. Try lowering your VCORE by using the adaptive setting. Modifying the base VCORE will supply your CPU with a constant voltage, regardless if its in IDLE or doing tasks.

 

Since there are no 2 equal situations, i'll give you mine as a start point (using a 7700k with a Gaming 7 Z270) - Vcore is set to auto (which is set to go as high as 1.3) and adaptive is -0.090. This pushes a maximum of 1.240 volts into the cpu, when in TURBO but with significant temperature drops (as far as 10 degrees, depending on the scenario of use).

 

Again, i can't write it enough: it is highly recommended to lower the voltage when using recent 270 boards from Gigabyte.

 

I am using Arctic Silver 5 and used the pea application method. I also removed the cooler and confirmed that the Thermal Compound spreaded evenly.

I really hope that you didn't put a pea-sizes glob of TIM on there. It really should be along the lines of a grain of rice or a lentel. You just want it to fill in the microscopic gaps and ensure that there is no air - even a little - between the CPU's heat spreader and the cooler plate.

This is probably the best discussion of TIM application that I've seen: http://www.tomshardware.com/reviews/thermal-paste-heat-sink-heat-spreader,3600.html.

You should also give your TIM a little time to "break in" and fill in all of the gaps before beating up on your CPU. IIRC, Arctic Silver has a relatively high viscosity so it'll take a bit to flow into the gaps.

And ... finally ... you really shouldn't see temps that high. Just test, I used the auto settings on my Asus to clock to 5Ghz and only hit 90C after running Prime95 (with AVX) for over 30 minutes.

Finally manually tuned the Vcore to 1.32 and got a very stable overclock of 5GHz. Temps are within 80C under load in a 20 c room.

 

Thanks everyone for your suggestions. For Gigabyte owner here is link that I followed.

https://www.joomag.com/magazine/gigabyte-z270-overclocking-guide-gigabyte-200-series-overclocking-guide/0767815001483933769

Out of curiosity, what did you use to test stability? For how long? With or without AVX?

Out of curiosity, what did you use to test stability? For how long? With or without AVX?

 

I tested with Aida64 for 1 hrs, fixed vcore 1.32v.

@pbcal Would you be able to test with OCCT? I have tested with AIDA64 and everything goes very smoothly with temperatures mainly around 70C with a 4.8ghz overclock. But when testing OCCT the temperatures are more around 85C and then it says there is an error about 3 mins in. I thought setting your cpu to 100% was setting your cpu to 100% but obviously not... lol :)

 

Also the stability test mentioned in the link you sent also fails badly!

 

EDIT:

 

I was using stress CPU should have checked FPU?

No, the FPU only AIDA test would have been a rude awakening. Most people do not need to run it. It's not entirely intuitive, but if your check off all four boxed of CPU/FPU/cache/system memory you will get a nice blend test. That is a good measure for general stability. FPU and cache has some specific uses, but most people don't need to run them. OCCT is generally a harder test and has some settings as well. Stick to large data set. It sound like you might be just a tick or two short on voltage, but keep in mind some synthetic stress tests become very voltage hungry at the highest possible clocks. I need another 0.11v to run Prime 95 for my 5930K at 4.5. No other program needs anything close to that for stability, so unless you need to run Prime, that is wasted. How are your normal use temps and stability?