4790K + H100i High Temps

[damien c]

Basically I have a few systems in my house 2 of them are cooled by H100i's.

 

The first system is a 3930K based system which is overclocked to 4.0Ghz and the second is my system which is a 4790k based system running at stock speed.

 

Now the 3930k's max core temps are 60c to 70c when using OCCT with AVX enabled, and the H100i on that has 4x Scythe GT1850 fans on it in push pull, with a ambient room temp of around 25c.

 

My system which is the 4790k is where the issue presents itself, originally with the Corsair SP120 Quiet Edition, fans on the H100i I was getting max core temps of 85c which is no better than the stock Intel cooler I would suspect.

 

Today I have swapped the fans to 4x Akasa Apache Black 1300rpm fans and the max core temps now are 81c which is lower but still stupidly high.

 

The CPU Core Voltage is 1.250v set the bios as that is the lowest I can get with the system remaining stable.

 

I have searched the internet for ideas on what could possibly be causing these temps and, eventually I found someone saying to put some washers on the back of the CPU Socket behind the motherboard on the backplate, for the cooler in order to make the backplate more secure.

 

I had previously done that which knocked the core temps down to what they are now, as prior to that they were hitting 90c.

 

I have checked that the pump is running and it is running at it's maximum speed according to Corsair Link when I managed to find a thread, with the registry changes required in order to get it to identify and pick up the H100i.

 

I have purchased a temperature monitor for the room with my system in it, and it is currently 23.6c ambient temp and the cpu is basically at idle now whilst I am posting this, and the core temps are 27c 27c 28c 29c which seem really good but as soon as the CPU usage goes up when either, gaming or encoding a video the temps just skyrocket.

 

 

Here is a video I uploaded showing OCCT running before changing the fans and then after.

 

[ame]

[/ame]

 

 

I honestly cannot understand why the temps are so high, I know that Intel have had issues with CPU temps since Ivy Bridge, but these temps to me are shockingly poor and something that I would not expect from a AIO and I would expect them more from the Stock Intel Cooler.

 

I have checked the mounting and the cooler is mounted securely it cannot be mounted any tighter than it currently is, without damaging the motherboard or the cpu iteself.

 

I have checked the TIM spread after deciding this afternoon after posting the video and the TIM spread was nice and even, not to thin and not to thick.

 

I have the fans plugged in to 12v molex adaptors with the fans on the H100i on one molex adaptor and then the case fans are plugged in to another, so they are not connected to the motherboard, the only thing connected to the motherboard fan headers in the little fan cable coming of the Pump/CPU block unit.

 

I thought as well that it might be because of the heat from my graphics cards which are 2 MSI GTX980Ti Gaming Twin Frozr cards, but even with my old GTX780Ti's which exhausted the air out of the case, the CPU core temps remained the same.

 

I thought it might be a airflow issue with the case, which previously was a Corsair Air 540 so I bought, a Phanteks Enthoo Primo case after selling some old hardware as I am currently on a limited, income due to being my mothers full time carer so I filled, the new case with fans but the CPU core temps remain the same, the only change that was made to temps was the GPU's are now 2c lower than in the Air 540 case.

 

I was even considering putting a cold air duct with a my desktop fan connected to it, and then connected to the fans on the inside of the case on the H100i, and bringing in cold air from outside but the current outside temperature is around 7c, and I really don't want to run the risk of condensation building up in the pc.

 

I do have the original packaged H100i fans, as well as some Scythe Kazi Jyuni 1900rpm fans that, I could try on the H100i but I really don't want the pc to sound like a jet getting ready for take off.

 

I am pretty much all out of ideas, and I am basically now looking at building a custom cpu loop as I have currently sat in a box, 2x XSPC 360x30mm rads, 1x XSPC 240x30mm rad and 1x Alphacool 240x60mm rad with resovoirs, tubing, barbs and fluid the only things I do not have yet is a pump and cpu block, as all that stuff is left over stuff from previous builds.

 

Does anyone have any ideas, before I go and spend up to £150 on a pump etc that I would have to save money for over the next 2 months.

[c-attack]

I had hard time reading the OCCT data during the video. Do your CPU temps instantly reach the upper 70's when you activate the stress test? Or is it a slower build-up to that level over several minutes? (with or without the 1 min warm-up) Fans and fan speed are the primary factors for water temperature, but they won't have any effect on your instant load temperatures. Fan choice may be significant to you for your encoding runs, depending upon run time, but if you instantly hit that 78-80+ range it has to be voltage (assuming no installation issues). You mentioned the 1.25 vcore, which is by no means extreme, but you didn't say weather that was fixed voltage, off-set, or adaptive. Combining adaptive voltage and OCCT Linpack + AVX is most certainly a guaranteed recipe for high temperatures no matter what you do. How do your results compare to OCCT standard and without AVX or another test program altogether? Probably more important than that, what kinds of temps are you getting during your encoding? You very well may make use of the AVX instructions in your work, but Haswell + adaptive voltage + AVX may pose an issue.

 

On another front, I noticed in the video your rear fan is bring air into the case leaving the H100i as the only exhaust point in the system. I don't think this has anything to do with your OCCT test high temperatures, but I am wondering how that is working for you in real application use. Trying to move all your exhaust through the radiator is always going to create a choke point at that level and it will certainly have an adverse effect on CPU temps when a moderate or higher GPU load is present. I know that configuration doesn't work well on a 540 Air and that was with a pair of measly 970's. The 980 Ti is whole different creature. If you have nominal GPU loads, then it may not matter too much, but I suspect you have Ti x 2 for a reason.

[damien c]

Thanks for the reply.

 

I have been awake since 5am (UK Time) testing and I have found that the cpu is a garbage cpu and will not run for longer than, 10 minutes at anything below 1.3v with or without AVX enabled.

 

The temps at that voltage are just stupid, peaking at 71c without AVX enabled and 91c with AVX enabled and that is at the stock 4.4Ghz.

 

The voltage is fixed in the bios not set to adaptive or offset.

 

The temps when rendering a video or using Aida64 stress test are the same as running when running OCCT with AVX not enabled.

 

Using Intel Burn Test or Prime95, the temps are the same as using OCCT with AVX enabled.

 

The rear fan is set to intake air, the bottom fans are now set to exhaust air now, the front fans are set to intake and the fans on the H100i are set to exhaust.

 

The CPU temps remain the same regardless of which way round the fans are set.

 

The 2 980Ti's are in use because I game at 4K, and the temps are hitting 79c on the cards in this case where as they were hitting 84c in the Air 540 case.

 

I am just going to get the custom waterloop sorted asap and then when I can I am going to rip this cpu and mobo out and just use them in a media server and get something different, probably go to X99 unless AMD's Zen based CPU's are any good.

[c-attack]

Well, it does tell you something. If the temperatures had not decreased without the AVX + Linpack, then you would know to start looking for a hardware issue. As it did decrease by a reasonable margin, it does seem probable the issue is the rigorous test protocols and the chip itself. Haswell is always really hot on IBT, Prime, and Linpack-OCCT. Those results don't surprise me too much, but the 70-ish AIDA64 run does strike me as high. I would have expected better. However, since I skipped over that chip, the generalizations about what it should or should not be are better left to someone else. Hopefully, the a larger reservoir and stronger pump will knock those numbers down. I don't know about the IBT and Prime, but I would think it will take your encoding runs into the lower 60's.

 

For comparison, my 5820k running at 4.4/1.285v adaptive, 4.0 cache/1.20v, Corasir H110 (prior Asetek) hit a steady 75C after the 1 min warm up on Linpack + AVX. I only ran for a few minutes, but I would expect my two hot cores to climb to 80C as the test continued. On AIDA64 full CPU/Cache/FPU/Memory run my two hot cores will get into the low to mid 60's, with the remaining 4 a full 9C lower. However, this is at 17C room temp and I have my system well dialed in at this point. I would say at best, I have an a slightly below average chip. I need 1.33-1.34V to run 4.5 completely stable. It is also fairly common to have the #3 & 5 cores (0-5) run considerably above the others on 5820k and 5930. That is always going to be the limiting factor.

[damien c]

I am getting close as well to de-lidding the cpu but I don't know yet.

[damien c]

Well after talking to my dad I ended up popping the chip in a vice and popping the lid off, applied some Artic MX4 and the temps have dropped but only by 2c on each core but the temps are closer across the cores now.

 

Will apply the TIM that is coming on Tuesday and see what happens but currently I am testing voltages to see what I can get it stable at.

[snapper69]

I like the Akasa Apache Black fans, but they aren't the right fans for cooling a radiator, they're best at cooling cases. You need SP fans for a radiator, I used to use Noctua fans on an H100i. They probably aren't the main problem, but they probably aren't helping.

[Kippsy]

Just for comparison - Here's a screen shot of my temps after a 5 min run through Aida64 (Stress CPU/FPU/Cache)

 

http://i.imgur.com/OLYlb2R.jpg?1

 

(Link to full size pic here - http://i.imgur.com/CpJuFN8.jpg

 

Im also a 4790k (Although running at 4.6ghz at the shown voltage) + H100i

 

Looking at your results, I would agree with you comment about the chip being abit of a dog, your temps are not that much higher than mine when using Adia64 though. And the 4790k is well known for being a hot cpu

[damien c]

Well update time.

 

I managed to get the money together to build a custom loop using the rads that I already had.

 

I de-lidded the cpu and found that I could run the cpu now at a much lower voltage, so I can now run at 1.160v in the bios and it runs for 7+ hours with OCCT running with AVX enabled, where as prior to de-lidding I could only run the cpu at 1.25v for less than an hour but it needed 1.3v and up to remain stable for longer than 1 hour.

 

Running 1.160v with the H100i the temps were peaking on the hottest core at 80c, so I managed to get a 1c drop when using the H100i with a de-lidded cpu running at a lower voltage.

 

On the custom loop with the same voltage I was getting a max temp on the hottest core of 70c.

 

TIM used on both those runs was Artic MX4 between the DIE and the IHS, with Noctua NT-H1 between the IHS and the H100i waterblock and then on the custom loop between the IHS and the EK waterblock.

 

Custom loop is setup with a 360x30mm rad and a 240x60mm rad, with 6 Akasa Apache Blacks on the 360 rad and 4 Corsair SP120 Quiet Editions on the 240 rad.

 

I have received the Coollaboratory Liquid Pro and applied some to the Die, and used some Clear Nail Polish over the little bits next to the die, I cannot remember the name of them I think it's power regulators or something like that.

 

I then used some Noctua NT-H1 on the IHS between that and the EK Waterblock.

 

Booted up the PC and loaded OCCT with CPU-Z, Core Temp and Real Temp and turned on OCCT with AVX enabled and the max Core Temps are now 60c so a full 10c drop, just by using the Coollaboratory Liquid Pro on the Die compared, to Artic MX4 on the Die and Noctua NT-H1 between the IHS and the block.

 

Going to try the H100i on my media/lan pc which has a I5 4440 in it, as I know that reached on the stock air cooler 65c on the hottest core, but I have not played with the voltages etc on that, because 65c I am happy with but I might as well try the H100i on it.

 

Happy with the temps now, and I can look at overclocking soon.

[c-attack]

That's substantial progress. Interesting about the voltage drop after the de-lid. I'm not a big fan of the Cool-lab stuff, but once you take the top off the risks no longer seem as relevant. It's possible you could take a little more off the top end on long runs with a higher rpm fans, however that benefit may be diminished with the higher flow (presumably) and larger tank custom loop. It also might be a bit annoying if you stay at the desk, but there is always the option of limiting it to 1500 rpm or so with the top end for emergencies.

 

I still have a couple of reservations about the H100i. I felt like it should have done better on the AIDA run, but perhaps the peak values were misleading and the result of the physical chip properties. If the steady state averages were a bit lower, that might tie things off. I guess you'll get another shot at it on the I-5.

[damien c]

Yeah I have some stupidly fast fans that I could put on but I want the pc to be quiet although, it's not to quiet now it's quieter than what it would be if I put the Scythe Kazi Jyuni fans on the rads.

 

I am currently running OCCT with AVX enabled, at 1.256v with a 4.6Ghz clock speed and the max temp currently is 68c after 12 minutes on the hottest core.

 

I will see how far I can push it with the temps not going above 75c as that is where I will be comfortable with a overclock, but I wasn't comfortable with those temps at stock speed.

 

[ame]

[/ame]

[c-attack]

Sounds like you are right and extra fan speed is unnecessary. The short term CPU temps under Linpack/AVX are even more impressive back at the 4.4/1.16 setting.

[damien c]

Well after 40 minutes at 4.6Ghz on the CPU and Ram at 2400Mhz with 1.25v for CPU and the ram at it's XMP Profile Voltage of 1.65v the max core temp is still only 69c on the hottest core.

 

I know it's not a high overclock and 40 minutes is not enough to confirm it being stable but it's looking like things might be decent, going to let it run for 1 hour and then try 4.8Ghz at 1.25v and see what happens.

[Kippsy]

Well after 40 minutes at 4.6Ghz on the CPU and Ram at 2400Mhz with 1.25v for CPU and the ram at it's XMP Profile Voltage of 1.65v the max core temp is still only 69c on the hottest core.

 

I know it's not a high overclock and 40 minutes is not enough to confirm it being stable but it's looking like things might be decent, going to let it run for 1 hour and then try 4.8Ghz at 1.25v and see what happens.

 

 

Mines runs stable at 4.6ghz on 1.254 volts (Using adaptive voltage is does over volt occasionally upto ~ 1.27 volts)

 

I then need 1.31 volts to run 4.8ghz, but the temps are too high for my liking using the H100i - Would expect you to be able to at least push 1.35 volts ~ 5.0ghz with custom loop cooling

[damien c]

Any voltage above 1.25v puts the temps up quite abit, but at the moment the cpu will not go above a 48 multi regardless of voltage.

 

I am going to get a direct die mount for the block and see what the temps are like then and at that point see where I can push the chip.

 

As for the H100i, I put it on my I5 4440 and the temps dropped when rendering from 65c to 45c, so the H100i seems to be working fine but it just couldn't cope with the 4790k, even when it was undervolted.

 

There is one curious thing though, which I am currently testing and that is when I press the power button on the I5 system it won't boot, but as soon as I loosen the retaining bolts off on the cpu block for the H100i it boots, but as I say I need to do some more testing.