H115 + Non OC 6700K = 80s temps

Been dealing with some event id 41 crashes while gaming, and while trying to pinpoint the root cause, I noticed my CPU temps were in the 80s, despite not being OC'd.

 

My parts list:

• Motherboard = ASUS Maximus VIII Hero Alpha
  
• CPU = i7-6700K
  
• PSU = Corsair HX850i
  
• Cooler = Corsair H115i
  
• PC Part Picker List Here
  

 

The H115i water temp progressively goes up (I've seen it hit up to 40 degrees while gaming), so I'm wondering if it's a flow issue. Also tried pushing down on the cooler and didn't get any temp reductions, so it doesn't look to be a contact issue. What could my problem be?

 

See below for Corsair Link stats from 3 separate runs of Intel Burn Test. One is with all mobo settings set to auto (i.e. non-OC), One is All Auto with voltage manually set to 1.35, and one is 4.6 OC with voltage manually set to 1.35 (I cut that run short because my temps started getting into the 90s).

 

Corsair Link stats - all auto in MoBo settings

 

Corsair Link Stats - All Auto Except manual voltage to 1.35

 

Corsair Link Stats - 4.6 OC and manual voltage to 1.35 (I cut this test short because of fear of the high temps)

 

Corsair Link stats - all auto in Mobo, 30 minute Borderlands 2 session

 

Also, if there's a different way for me to test/log, please share and I'll be happy to do so.

I don't think you have a flow problem. IBT would be untenable in a matter of moments and with an H115i Temp (water temp) in the 50-60C+ range. Your water temp goes up steadily over the 10-12 minutes of the test. This is very normal for an extreme load like Prime or IBT.

 

Skylake CPU behavior on AUTO for stress tests is pretty well documented. It looks like you have demonstrated it again. Your fixed voltage test cuts 20C off your core temps and that also tells you where the real problem lies. You can't run those types of program on auto voltage. It will add as much as it thinks is necessary to keep it stable, up to silly levels.

 

In your gaming, you started off at 31C water temp and rose about 8C over the course of the session. This is fairly normal, especially if you are running some AAA level graphics and moderate watt GPU. It's not just the water that rose to 38C, but the entire ambient internal temp. I can never make sense out of the multiple motherboard temps in Link, but #2 is likely the actual board sensor and tracks right along with it. Your CPU temps are low and the CPU load is low, but the overall case heat keeps the baseline water temp up.

 

When did the 80C CPU temps happen? Gaming? Was it a spike or an sustained period at that temp?

 

I don't think there is anything wrong with the cooler or your contact. Unfortunately, I can't shed any light on your error 41 either, however I don't think there is any reason to start ripping the cooler apart.

Thank you for the well thought out reply! Your explanation covers 3 of the 4 links I provided and gives me some peace of mind, but then why when I OC to 4.6 and have a manual voltage of 1.35 do I see temps in the 90s when stress testing?

 

Also, to answer your question, the 80 degree CPU temps came during stress testing. When I tried a test OC to 4.6 with a manual voltage of 1.35 is when I saw temps shoot to the 90s during IBT which are outlined in the link below. At idle I usually hover in the 30s.

 

See below:

 

Corsair Link Stats - 4.6 OC and manual voltage to 1.35 (I cut this test short because of fear of the high temps)

So it turns out that's just the nature of the Intel Burn Test. Ramps temps up dramatically. When I tested with RealBench I was seeing temps in the mid-70s with a 4.6 OC and 1.35 volts.

 

Sorry for the freakout on my end, I'm now officially comfortable with the H115 again :)

 

Now to just figure out where these event id 41s are coming from...

The 4.6 - overclock link is still locked in Google Drive. That was why I didn't comment on it earlier. However, at the risk of a crude analogy, using IBT is like using a blowtorch when you need a scalpel. It's just too much on the CPU side of the equation. CPU architectures have changed and the days where you could run Prime for hours on an overclocked 930 at 55C are long gone. You could get away with running that level of program at stock frequencies on the Haswells, but Skylake (and likely all the future 'Lakes') are different in the way they handle these types of loads. I am half expecting a warning label directing the end user not to use synthetic benches on future chips. As for the current difference between 4.6/1.35 and 4.2/1.35v, there may be some other BIOS setting besides Vcore that have an impact on the testing. I would really need to crawl through the BIOS and test to find out. It's also possible the water temp was higher than in the other test to start and your reached the limit in a shorter amount of time. Either way, I am not sure IBT is the best tool for doing a health check. It's too severe and puts too much load on the front side of the heat exchange. The voltage/heat always has to pass through the CPU before transferring through the lid into the cooling the system. If you apply a 1.80v Vcore load, your cooler design won't matter. The silicon will be damaged. You are not there, but not all 100% loads are created equal.

 

If you want to run a test to check on your cooler functionality, I like Intel XTU. It's a mild test, although its real value is very even, sine wave like loading. If you see a core temp jumping out of line, you know you have a TIM/contact problem and the temps are low enough people with flow problems can run for more than 20 seconds without going into thermal shutdown. AIDA64, and OCCT in particular, use a more varied loading pattern and make for better stability tools, with OCCT being a step up in difficulty. There certainly are other tools as well. However, at the risk of another analogy, if you get a new sports car, you want to see how fast it will go or how well it drives. You don't want to see how hot you can make the engine as the primary goal. That is Intel Burn Test. It's all in the name. I just don't think it has a lot of utility for the majority of users.

 

If you run OCCT, you will need to fix the voltage, just as with IBT. I think Intel XTU can manage on full auto, but likely also does better with a fixed or manually set adaptive voltage. Run a comparison test with another program. It's always a good idea anyway, but I think it will also give you some peace of mind.

So it turns out that's just the nature of the Intel Burn Test. Ramps temps up dramatically. When I tested with RealBench I was seeing temps in the mid-70s with a 4.6 OC and 1.35 volts.

 

Sorry for the freakout on my end, I'm now officially comfortable with the H115 again :)

 

Now to just figure out where these event id 41s are coming from...

 

Now, if only I had been less wordy... I would have beat you to it.

I've stopped using stress/burn tests, as they seem to cause too many problems. I set my system up, overclock it and set the voltage and use it, do some browsing, play videos and play a game or two and see what happens. I do sometimes run 3D Mark to test my graphics and I keep an eye on the temps when I do so, but that's about it.

If you want to run a test to check on your cooler functionality, I like Intel XTU. It's a mild test, although its real value is very even, sine wave like loading. If you see a core temp jumping out of line, you know you have a TIM/contact problem and the temps are low enough people with flow problems can run for more than 20 seconds without going into thermal shutdown. AIDA64, and OCCT in particular, use a more varied loading pattern and make for better stability tools, with OCCT being a step up in difficulty. There certainly are other tools as well. However, at the risk of another analogy, if you get a new sports car, you want to see how fast it will go or how well it drives. You don't want to see how hot you can make the engine as the primary goal. That is Intel Burn Test. It's all in the name. I just don't think it has a lot of utility for the majority of users.

 

If you run OCCT, you will need to fix the voltage, just as with IBT. I think Intel XTU can manage on full auto, but likely also does better with a fixed or manually set adaptive voltage. Run a comparison test with another program. It's always a good idea anyway, but I think it will also give you some peace of mind.

 

Thank you for this! I'll definitely look into XTU, OCCT, and AIDA64 for proper stress testing. I wasn't a fan of how my CPU temps would ramp up into the 80s with IBT (even with manual voltage of 1.32 on a 4.6 OC). I'd definitely prefer something less harsh on the CPU.

Asus Real Bench is fine too. I wasn't a big fan originally, but it's a step up in difficulty from Aida without needing the additional voltage padding for OCCT when in that last tenth or two. It also might reveal issues that are not exclusively CPU in nature. I actually prefer it for one stop, quick testing on Haswell-E.