strange temperature behaviour

[THRESHIN]

built a new skylake system couple months ago, and i've honestly been too busy to try overclocking until now. the temperatures are being very strange....

cpu is i5 6600k, on a gigabyte GA-Z170MX-gaming 5 board. cpu is water cooled with an old apogee gt block....custom mounting lol http://hardforum.com/showthread.php?t=1884610&highlight=

so here's where it gets strange...idle temps are very normal, around 23ºC. i've got a good water setup, swiftech 655 pump and a 3x 120mm rad as you can see in the pics. video is on water and doesn't seem to have these issues. i'm using prime95 to stress test, but the test program doesn't seem to matter either.

the very instant i put full load on the cpu, the temperature jumps to 55ºC and goes up a little from there....go back to idle and again it instantly dumps down to 23ºC again. that doesn't seem physically possible....

i've tried multiple test programs. CPUID hardware monitor, realtemp, OCCT and the gigabyte software. all report the same. wtf is going on here???

 

[Araxie]

so, what's the strange behavior in temperature? I see all absolutely fine...

hmmm. After reading a bit of your other thread I see that you may need to familiarize a bit with newer tech and understand a bit that since Ivy Bridge intel doesn't utilize anymore soldered internal CPU heatsink, so instead they use a cheap low thermal conductivity paste, which explain your exactly behavior with temps, temps doesn't gradually increase as old tech it just go straight up to the limits of the TIM, this have been improved a bit since devil's canyon (haswell rehash) as haswell was so even crap that with even great custom loops temps in a 4790K would just instalock at 100C... all it's fine with your machine and with your setup. just enjoy it, if you want to really fix it google a bit of Intel CPU Delidding methods, it seems that you have the tools for do it fast and safe =).

 

[THRESHIN]

yeah its been a long time since i've upgraded....just 55 on water seems silly. the instant jump is very alarming in my mind...

the heat spreaders really that bad? and just for fun, whats normal load temps for these things? i'm used to barely over 40ºC on this setup with my old hardware....even overclocked to the moon.

thanks for your input! i added that thread as an afterthought...glad i did

 

[Ocellaris]

yeah its been a long time since i've upgraded....just 55 on water seems silly. the instant jump is very alarming in my mind...

the heat spreaders really that bad? and just for fun, whats normal load temps for these things? i'm used to barely over 40ºC on this setup with my old hardware....even overclocked to the moon.

I think you've been out of the game a bit long. 55C load temp is great.

What are you using to read temps? I haven't found a single app that reported consistent numbers for Skylake yet.

 

[unclewebb]

Try flowing about 70 Amps through your pinky fingernail. That is about how big a Skylake CPU is and that is what happens when you start running Prime95. Of course the core temperatures are going to instantly jump. Your fingernail would look like burnt toast by the time you were able to reach for the switch and turn the power off. The moment you stop Prime95, core temperatures instantly drop.

All monitoring apps are reading the same core temperature sensors and have been using the same method for the last 7+ years to determine the core temperature. Intel standardized this so the location of the register that holds TJ Max information and the other register that contains the thermal sensor information are in the exact same location that they were in back when the first gen Core i7 was released.

This means that when a CPU is fully loaded, all programs should be showing pretty much the same thing. Use Prime95 - Small FFTs to equally load all cores, give the CPU a chance to stabilize and then post a screenshot of all of the monitoring programs so we can see all of the monitoring programs that are not reading the core temperatures correctly. When a CPU is partially loaded, peak core temperatures will be bouncing around so it is not unusual if different programs reading the sensors at slightly different time intervals show slightly different results. Some programs have a hard time gaining access to the temperature sensors when the CPU is fully loaded. If you see one program consistently reporting higher temperatures, that's usually why.

 

[THRESHIN]

just for fun, here's a screenshot of with prime95 load applied to applied to all cores (small FFTs) using 3 hardware monitors:

and then less than a minute later right after load removed:

i do have a slight overclock, bumped it to 3.9Ghz. no voltage bump or anything else at this point. it was the multiplier, so mainboard and ram are not a problem. i figure thats the point of owning a K cpu right?

everyone seems to be saying the same thing here, so thank you all for your input. it's probably just a case of being out of the game for a long time as suggested. i do get what you're saying unclewebb about putting a lot of power through a small die, just surprised that the temp change is so fast both ways. i honestly figured there was something wrong with the temp reporting.

when i build a system i generally run it into the ground hard. i don't really like spending the money on parts until i see it as necessary....and i seem to be able to handle playing at lower framerates when i have to without much difficulty. turn down a few settings, etc. i think my previous core2 system go through 8 years with only a couple upgrades.

so the last question is....what should i consider my upper limit for temperature when i'm overclocking? 70ºC? 80ºC? i wonder if it'll even jump up much with water...?

 

[Araxie]

it's not really quite strange once you understand how tiny is the CPU die even in comparison to the iGPU, the smaller size in the die make really difficult to dissipate all that heat gradually so it make a big jump in temperature once stressed but the low power nature of intel chips make it to be able to be cooled faster, so that's the wild variation in temps from idle to load and vice-versa, take in consideration that the advertised TDP include CPU+iGPU load scenarios, so, when the iGPU isn't used the power consumed by the CPU is just minimal and can be cooled extremely fast.

And well, first of all I think you should stop using prime95. since the launch of haswell prime95 has been a trouble as it can really damage a chip and with Haswell-E chips it can even fry both chip and motherboard, so that's an important thing to consider, if you want to use Prime95 use the 27.7 as the latest one possible as that doesn't include the newer AVX instruction badly implemented that can harm a CPU, lets be honest, to this date P95 is just really outdated and need to be re-coded for newer tech. actually Aida64 is the recommended tool for stressing and stability test. also Asus RealBench it's probably my favorite stability test suite since launch and its actually becoming quite popular as it stress and test the CPU (also GPU if you want) with realworld applications that will also tell you how your system behave in realworld workloads, also take in consideration that a big difference with modern chips is that none have a truly "stock" voltage under load and that voltage can vary from one chip to other depending on the luck of the draw in the golden CPU lottery, this apply specially for "Auto" voltage CPU settings some apps as Prime95 can make the CPU to work wild and suddenly apply an instantly injection of high voltage that can damage it, so google a bit about auto, adaptive, offset, manual, every motherboard have it's pro and contras regarding those CPU voltage options.

about limits, the limit it's really the TJmax so 100C for your skylake chip, so you are far to be worried about reaching unsafe temperatures under heavy stress and stability, you will see that in realworld things are gona be completely different.

My general rule its to keep any CPU in the range under 70C-75C under heavy workloads for 115X systems and under 85C for X79 and X99 systems just for peace of mind, of course normally under gaming, those intel chips are cool I think in your system you will rarely see it over 45C in heavy gaming sessions, nothing to worry really, so have a bit of fun and overclock it more you shouldn't have any trouble with getting it to 4.5ghz - 4.7ghz.

 

[THRESHIN]

it's not really quite strange once you understand how tiny is the CPU die even in comparison to the iGPU, the smaller size in the die make really difficult to dissipate all that heat gradually so it make a big jump in temperature once stressed but the low power nature of intel chips make it to be able to be cooled faster, so that's the wild variation in temps from idle to load and vice-versa, take in consideration that the advertised TDP include CPU+iGPU load scenarios, so, when the iGPU isn't used the power consumed by the CPU is just minimal and can be cooled extremely fast.

And well, first of all I think you should stop using prime95. since the launch of haswell prime95 has been a trouble as it can really damage a chip and with Haswell-E chips it can even fry both chip and motherboard, so that's an important thing to consider, if you want to use Prime95 use the 27.7 as the latest one possible as that doesn't include the newer AVX instruction badly implemented that can harm a CPU, lets be honest, to this date P95 is just really outdated and need to be re-coded for newer tech. actually Aida64 is the recommended tool for stressing and stability test. also Asus RealBench it's probably my favorite stability test suite since launch and its actually becoming quite popular as it stress and test the CPU (also GPU if you want) with realworld applications that will also tell you how your system behave in realworld workloads, also take in consideration that a big difference with modern chips is that none have a truly "stock" voltage under load and that voltage can vary from one chip to other depending on the luck of the draw in the golden CPU lottery, this apply specially for "Auto" voltage CPU settings some apps as Prime95 can make the CPU to work wild and suddenly apply an instantly injection of high voltage that can damage it, so google a bit about auto, adaptive, offset, manual, every motherboard have it's pro and contras regarding those CPU voltage options.

about limits, the limit it's really the TJmax so 100C for your skylake chip, so you are far to be worried about reaching unsafe temperatures under heavy stress and stability, you will see that in realworld things are gona be completely different.

My general rule its to keep any CPU in the range under 70C-75C under heavy workloads for 115X systems and under 85C for X79 and X99 systems just for peace of mind, of course normally under gaming, those intel chips are cool I think in your system you will rarely see it over 45C in heavy gaming sessions, nothing to worry really, so have a bit of fun and overclock it more you shouldn't have any trouble with getting it to 4.5ghz - 4.7ghz.

great explanation, thank you. i have a good friend who has pretty much the exact same rig - turns out we think alike when it comes to system building. got him to play around and he had identical results to me. i'm just happy to know that this isn't something wrong.

 

[Nenu]

Temps are fine.
My 6600K clocked to 4.7GHz with an Apogee XT block performs like yours.
When gaming for a long time my CPU rarely gets hotter than 55C.