How best proceed with overheating i7-4790K?

[REnso1]

I have an i7-4790K in a GA-Z97MX Gaming 5 mobo latest F4 BIOS.

I used a Noctua NH-L12 cooler rated at 95W for cooling the processor rated at 88W in a Lian Li PC V354 with 4 fans, 2 in 2 out, case closed and case open, ambient is 27-30°C.

Temperatures in BIOS and memtest86+ were high so I decided to try stress testing and in Prime95 small FFT cores 1&2 overheated to 100°C using Core Temp.

I tried reseating the heatsink and renewing the NT-H1 TIM and opening the case but it made no difference. I have a photo of the contact pattern here.

When I tested using the OCCT benchmark I was unable to complete a test due to the processor overheating so I underclocked the processor to 3.6 GHz, disabled turbo and manually set vCore to 1.1v.

With an underclocked processor I was able to get a heating and cooling curve using the OCCT auto capture, to enable me to study the problem.

Even when underclocked the processor was reaching high temperatures, rapid fluctuations in temperature with work load suggest a bottleneck in the thermal pathway. When I tested with the intel retail cooler which came with the CPU the cooling was much less effective than the NH-L12 (even when underclocked taking just over a minute of OCCT to reach the 85°C cut off point see below) indicating the NH-L12 was doing a good job of removing heat, which meant the processor was making the heat or the source of the bottleneck.

I have discussed it http://forums.hexus.net/cpus/327593-4790k-overheating-nh-l12.html elsewhere. Advice was to contact Intel due to an absence of information relating to my retailer's testing procedures. I have asked about these but am still waiting for a reply.

So my question is how should I proceed from here? Does this qualify for an RMA? If so is it possible to negotiate this with Intel direct or do I have to go through my retailer?

I have done my best to make sure I am not doing anything wrong and I would be grateful for any pointers to any mistakes I may be making.

[JBras1]

You can use the reference link 3zax gave you, I'd also make a couple of screenshots with CoreTemp under stress-load and provide them with your package, If you are using the stock-cooler I don't see how that'd be an issue as you are using the cooler THEY have provided with the original package. You might also give him a link to this thread as a reference, if that helps you out. If your problem is the same as mine then your cooler should blow like crazy for a few seconds as soon as you start up your browser or anything else, this is not normal for some high-end CPU.

However, if you are like me returning the CPU will not provide you the solution. I have returned my first CPU and this CPU is just slightly cooler than the one I had before.

[SElwa]

3zax

Thank you very much for the quick response, but I have a question, as I understand TCASE is different from Core temperature and also as I have read before on the web Core temperature tends to be higher than TCASE in the range of 10c to 15c or 20c, then can we consider the TCASE as an evidence? Please enlighten me

jeffrey0201

Yes, my core temp reaches 100c under any kind of stress test I threw at it, but while browsing (using chrome and opening 20+ tabs) that doesn't happen also while using photoshop or visual studio 2013 that also doesn't happen and while gaming (tried Hitman Absolution, Watchdogs and L.A Noire) guess what as soon as a game utilize 50%+ from the CPU after a while it reaches 100c and before updating my motherboard bios the PC would just hang but after the updating the CPU just throttles

[lpart]

Sorry for my english

the cores are on average higher than 10 to 15 ° to the temperature of the housing (Tcase)

Either way if the cpu throttle so that it runs in its frequency of origin, that the cooler is mounted correctly and good bios parameters then there is a problem.

I'll post tonight settings that I did in my bios (voltage, ...) to have a lower temperature than when everything was default.

[SElwa]

3zax

Don't worry about your English, it's not my first language either.

My CPU throttles to 3.9GHz - 3.2GHz and the cooler is mounted correctly (I checked four times) and I changed bios settings to no vain to the point that I set the VCore to a static value of 1.125v and set all cores frequencies to 4.2 (following that reference http://www.anandtech.com/show/8227/devils-canyon-review-intel-core-i7-4790k-and-i5-4690k/2 AnandTech | Devil's Canyon Review: Intel Core i7-4790K and i5-4690K), but the temperature reaches 100c under the stress test.

So if you can help me and point me out to a very trusted review for that CPU in which the reviewer uses ONLY the stock cooler it would be great as an evidence for my retail store cause I can't find any.

Also, I am waiting for your bios settings may be It could help, who knows!

[JVene1]

ShadiElwan, the best I can offer for the moment is a short technical specification on the CPU which shows that 100 C is the temperature where damage to the CPU will occur, and that maximum "operating" temperature is 72.72 C.

First, http://www.cpu-world.com/CPUs/Core_i7/Intel-Core%20i7-4790K.html http://www.cpu-world.com/CPUs/Core_i7/Intel-Core%20i7-4790K.html

Near the bottom of the details (about half way down the page) is the max operating temperature.

Then, from this page:

http://ark.intel.com/products/family/75023/4th-Generation-Intel-Core-i7-Processors# @Desktop http://ark.intel.com/products/family/75023/4th-Generation-Intel-Core-i7-Processors# @Desktop

Then selecting for the 4790k, for example:

http://ark.intel.com/products/80807/Intel-Core-i7-4790K-Processor-8M-Cache-up-to-4_40-GHz ARK | Intel® Core™ i7-4790K Processor (8M Cache, up to 4.40 GHz)

The "Tcase" temperature, the maximum allowed at the IHS, is 74 C

Now, from this page:

http://www.intel.com/support/processors/sb/cs-033342.htm Processors — Processor operation temperature FAQ

We have the difference between the meaning of Tcase and Tjunction

What does Tcase Max and Tjunction Max mean?

Tcase Max is the maximum temperature that the Tcase sensor should reach. Both Tcase and the thermal specification information can be found on the Intel web site.

Tjunction Max is the maximum temperature the cores can reach before thermal throttling is activated. Thermal throttling happens when the processor exceeds the maximum temperature. The processor shuts itself off in order to prevent permanent damage.

So far, in the course of about 5 minutes, I can't seem to find the technical documentation from Intel which states the Tjunction temperature, but it SHOULD be available for the processor on it's spec sheet (I just don't see it).

In any event, find the spec sheet, show the above definition (especially the "in order to prevent permanent damage") part, and....note that on the spec sheet (which I did see) is specifies the "thermal solution specification" - which is the heatsink required to achieve normal operation in the specs (the stock heatsink).

That should be sufficient for any retailer competent enough to sell computer parts.

Emphasis on "should", because this process does not go smoothly enough considering the ubiquity of the problem.

[idata]

Intel support told me that the max is 75C

I have adjusted down from 80C peak. New Max CPU Current Limit settings for 75C is now 76 AMPS.

This gives still gives me short turboboost to 44x but under stress, the cpu is limited to 38x at a core voltage of 1.0340v

(Big $$$$ for a 38x peak processor. !!)

This is set in bios of the Asrock Z97 extreme6 , but I used the Intel extreme Tuning Utility to set and test for a lower peak temp.

The "Processor Current Limit" of 76 Amps is the only setting that I make.

[lpart]

Hello ShadiElwan

What are your temperatures at idle?

Are you sure 1.125V is static and that it does not increase when cpu load

If indeed the cpu goes up to 100 ° C with 1.125 and that everything is done properly then he has a real problem on this processor.

You can try a test with OCCT (it stops at 85 ° by default) and watch the graphics (in my pictures) to check for different voltages and frequencies in the cpu load.

These graphs can also be used for your vendor.

PS: between each disassembly, clean the cooler, the ihs and reapply thermal paste

[SElwa]

3zax

Hello 3zax, YES, I am 100% sure that 1.125v is static, my idle temp is between 38c and 41c considering my room temp is 25+c, also the CPU temp in the bios is between 51c to 64c and when I tested the same CPU in another motherboard the temp in the bios is 41c but under stress test both reaches 100c core temp.

My CPU is with the maintenance guy now, so I can't do any further testing right now.

[RKimb]

With reference to Tcase, one of the problems I have is, in Linux, to know which lm-sensors output is which. I have the following temperature info:-

acpitz-virtual-0

Adapter: Virtual device

temp1: +27.8°C (crit = +105.0°C)

temp2: +29.8°C (crit = +105.0°C)

coretemp-isa-0000

Adapter: ISA adapter

Physical id 0: +63.0°C (high = +80.0°C, crit = +100.0°C)

Core 0: +63.0°C (high = +80.0°C, crit = +100.0°C)

Core 1: +62.0°C (high = +80.0°C, crit = +100.0°C)

Core 2: +46.0°C (high = +80.0°C, crit = +100.0°C)

Core 3: +57.0°C (high = +80.0°C, crit = +100.0°C)

it8728-isa-0a30

Adapter: ISA adapter

temp1: +32.0°C (low = +127.0°C, high = +127.0°C) sensor = thermistortemp2: +39.0°C (low = +127.0°C, high = +127.0°C) sensor = thermistortemp3: +52.0°C (low = +127.0°C, high = +127.0°C) sensor = Intel PECI

Which one is Tcase?

[idata]

coretemp-isa-0000

this is TCase

[RKimb]

You mean Physical id 0 ?

This seems always to equal the currently highest core and varies just as quickly as the cores do. I would expect Tcase to vary more slowly.

[lpart]

hello all

At home I have seen (with the default bios and updated) when the cpu switch to turbo (4400 mhz) increased the vcore to 1.31V which had the effect of having temperatures close to 100 ° c.

I have therefore seek the optimal BIOS setting to lower the temperature at best.

I will give the most important if not ca could be used to other settings.

voltages after quote may vary from one processor to another and from one configuration to another.

my mainboard: Asus maximus 7 ranger

my cooler: Corsair H100i

blck frequency -> 100

cpu core ratio -> sync all cores

core ratio limit -> 44

min cpu cover ratio -> 43

max cpu cover ratio -> 43

cpu core voltage -> 1.228

Cache cpu voltage -> 1.198

Initial input voltage cpu -> 1.67

Eventual input voltage cpu -> 1.67

With these settings I get to 70 ° C max on the hottest core (test with 6h Occt and 2h Aida64)

On average I'm at 60-65 °

room temperature 22.5 °

[SElwa]

I have contacted Intel, and they replied with:

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Hello,

I understand that you are concerned about the processor reaching 100 degrees Celsius when running stress tests.

The processor should not reach above 74 degrees Celsius in full load, please make sure that you are reading the CPU temperature and not the cores temperature.

It is normal to see the cores of the processor reaching higher temperatures than the CPU itself.

Also there is a reported overheating issue with some of these processors

located here:

https://communities.intel.com/thread/54032?start=15 https://communities.intel.com/thread/54032?start=15<</span>(>&<)</span>>tstart=0

This overheating issue should be fixed by installing the latest BIOS version for your motherboard, however if the issue continues, then the processor should be replaced.

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------

So guys how can I read the CPU temp not Core temp? (Any softwares other than HWMonitor?)

Also, what is the exact distance between Core temp and CPU temp (TCASE) of that CPU? Or at least the maximum estimated average?

[idata]

I'm not a thermal engineer but...

As you can see here:

http://www.pugetsystems.com/labs/articles/Delidding-the-4790K-for-a-quick-look-at-the-TIM-573/ Delidding the 4790K for a quick look at the TIM - Puget Custom Computers

This rectangle component inside (die - core) that's core exactly.

It's covered by heatsink (TIM) which should conduct all the heat very well to IHS, then IHS should conduct heat to IHS heatsink (this one you put on cpu, or this one put at factory on box cooling), and then this heatsink conducting to radiator.

So Core->Internal Heatsink->IHS->External Heatsink->Radiator

If conduction parameters of Internal Heatsink is good (and Intel claims, they use NGSTIM in 4790k), then heat "loose" between core and IHS should be maybe 10 or 15%.

This might be exactly calculated, if Intel would release:

- Internal heatsink thermal conductivity

- IHS thermal conductivity

- Stock - external heatsink conductivity

Then,

P = power (our TDP is 88W)

d = thickness of internal heatsink layer

S = core surface

delta T = temperature diffrence between core and internal heatsink

For example:

Core "generating" 88W of TDP heat, internal heatsink conducting 84W (4W "stays" at core generating HEAT - our cause of problems) to IHS, IHS conducting 80W (4W "stays" at IHS), then external heatsink, then radiator which doing it's job quite well, beacuse of dimensions and airflow.

It's just approximations of course... but, when everything is working well (heatsinks, ihs) then answer is: diffrence between core and IHS, shouldn't be bigger than material conducting diffrences. Maybe I say 10 or 20%.

If i'm wrong, then let somebody correct me.

[idata]

Btw, i'm also planning to write software, which will:

1) Lookup your CPU ID

2) Check from internal database according to CPUID: designed TCase_Max, nominal frequency, turbo frequency

3) Put stress test on your CPU, for all cores

4) Monitor frequency and temperature for some time

If the temperature will reach 100c, or will cross TCase_Max greatly, CPU will be underclocked (under values sold by Intel), then such CPU might be marked as FAULTY and REPLACMENT needed. More interesting would be also submitting and building codebase of CPU ID and Batch ID. This way we can build full database of Intel faulty manufactured CPUs.

I think such software for diagnosing this problem might be very usefull, especially for average PC user. Just look how many ppls having problem with overheating CPU.

All of you know how to measure and test it, but honestly tell me - how many % of Intel customers could do that?

So making such software could help to identify real scale of problem.

If anybody is interested in placing such tests, let me know.

Also, keep in mind that underclocking CPU to 4GHz or disabling HT is only masking a problem, not solving it.

Same as placing better than stock cooling. It might keep CPU cooler, let say 80-85c and working OK, but if (possible) CPU internal heatsink problem would be solved, then you might get as low as 50 or 60c on same non-stock cooling. So still, just masking a problem

[lpart]

I think often the temperature data from the software that came with the mother board (ASUS Ai following example) gives the temperature of the cpu

To be confirmed.

Either way a temperature of 100 ° on the cores probably give a higher temprerature at 74.04 °

[idata]

In monitor mode, the Intel Extreme Tuning Utility (XTU) will show the CPU and individual core temps. It will also let you change many of the operating voltages,wattages, and amperages on the fly. Here's the link: https://www-ssl.intel.com/content/www/us/en/motherboards/desktop-motherboards/desktop-boards-software-extreme-tuning-utility.html Intel® Extreme Tuning Utility (Intel® XTU)

[JVene1]

karwos wrote:

I'm not a thermal engineer but...

It's just approximations of course... but, when everything is working well (heatsinks, ihs) then answer is: diffrence between core and IHS, shouldn't be bigger than material conducting diffrences. Maybe I say 10 or 20%.

If i'm wrong, then let somebody correct me.

I don't have a correction, just underlining one of your qualifiers - when the IHS is working well, or correctly mounted at Intel.

Consider this thread: http://forums.anandtech.com/showthread.php?t=2261855 http://forums.anandtech.com/showthread.php?t=2261855

In particular, look at his list of links for the "http://forums.anandtech.com/showpost.php?p=34053183&postcount=570 Proof that the benefit from Delidding is entirely due to reducing the CPU-to-IHS gap"

This latter discussion involves comparing TIM compounds, and in that list he also shows the effect of running the CPU without the IHS (fitted with an adapter to support the HS).

The bottom line is that the primary issue is a random manufacturing flaw where the IHS isn't making contact.

So, in your equations one must account for the lack of contact, and how much resistance to thermal transfer is associated with it.

In reality thermal compounds do NOT conduct heat well on their own. When there's a gap between the core and the IHS, the compound can't transport sufficient heat away from the die. When that flaw isn't present, either because the fabrication at Intel doesn't lift the IHS off the die, or some brave enthusiast "delids" their processor, scrapes off the glue lifting the IHS and thus actually corrects the flaw, the IHS does a fairly good job.

The thread I linked claims a 19.8 C drop in full load temperature compared to the stock IHS without other interventions, when overclocked to 4.7 Ghz.

He shows a 4.7 Ghz overclock where the stock IHS (before delidding) on water (Corsair H100) hit 97.6 C, but the "corrected" IHS on the same setup maxed at 77.8 C.

He clearly shows this is independent of the TIM compound used, give or take 1 or 2 C, and entirely associated with the IHS mounting at fabrication.

It's also interesting to note his chart on bare mounting the HS, removing the IHS entirely.

For his cpu, stock speed is 3.5 Ghz. At 3.5 Ghz under full load, the stock IHS temp rose to 64.8 C, hit 56.4 C with the "corrected" IHS (after delidding), hit 55.6 without the IHS.

In other words, while temps dropped over 10 C just correcting the IHS, there was only a 0.8 C difference between the correct IHS and the bare HS mount.

At 4.7 Ghz the difference between corrected IHS and bare mounting the HS was 4.2 C. The IHS increased the highest overclock temp by 4.2 C, but the gap in the stock IHS increased that by 19.8 C.

Total potential temperature drop, from a flawed stock IHS to a bare HS mount (no IHS), overclocking to 4.7 Ghz on water (Corsair H100) was 24 C.

There's an interesting implication from the shim test described at the link above. Using a caliper, he measured the stock CPU as 4.21 mm thick, and observed that when the glue was removed under the IHS there was a reduction of 0.06 mm in thickness. His shim is 0.06 mm thick for his tests.

The implication is that we MIGHT gain knowledge about a particular CPU by measuring it's thickness and comparing that measurement in a database. My own hypothesis is that CPU's exhibiting temperature anomalies will be closer to the observed 4.21 mm than the 4.15 mm of the delidded version (meaning a 0.06mm gap was due to the glue).

[idata]

You are 100% right. However, situation with bad contact between conductors is "sure" situation when things will stop working... so yes, bad contact might be here in account as well..

ANYWAY, SMALL UPDATE .....

Intel asked me to do IPDT tests. So i grabbed IPDT Fedora Image.

Installed i7z in meanwhile, and here is very interesting results:

http://youtu.be/vRQdKtBmjx0 http://youtu.be/vRQdKtBmjx0

CPU is melting at 100c, speed drops, but IPDT tests are being passed and you can see big green "pass" window....

Funny, isn't it?

[JVene1]

karwos wrote:

Funny, isn't it?

Yes, in that ironic, tragic, peculiar sort of way.

A typical owner probably wouldn't realize there's an actual problem in common usage, precisely for the reason this test passes:

The CPU is throttling down to keep from self destruction without crashing or interrupting the test, though obviously at reduced throughput.

Please tell me they're going to replace these on a warranty exchange!

[lpart]

Intel thinks the results of this test ?