i7-4790k heat issue

[you2]

I'm having a bit of a heat issue with my I7-4790K when stress testing. I am using intel stock cooler and seeing temps in the mid 80's (85C) with mprrime/prime95.
-
I'm fairly sure the cooler is installed correctly (could see pins on other side of the motherboard). Iniitally I was seeing 100c but under volted 0.050 vcore; that reduced things to 90c and a day of baking (mostly testing ram) and temp is settling around 84/85c.
-
The intel tool (link found in second link) indicates that the processor (at max usage - mprime) is using 82watts peak so my undervolting might not be neccessary for thermal specification (88watts) but is necessary for the temp. Also the thread (first link) suggest I might actually have a 'good' processor as others seem to run hotter.
-
From these threads (and others) I'm begining to think this is the norm and not an issue with my build:
https://communities.intel.com/thread/54032?start=315&tstart=0
https://communities.intel.com/docs/DOC-23517
--
A few questions:
Have others stressed test the I7-4790K with STOCK cooler and seem better results ?
I'm a bit confused from intel documentation and is it safe to run @ 85C for long periods ?
(thermal throttling kicks in at 100C and auto shutdown occurs at 130c/72c (130c is core and 72c is tcase(?))
-
Email to intel support has so far resulted in mix results. They won't comment if the current thermal behavior is (a) expected, (b) harmful but the thread linked above does have intel folks comment that there might be an issue here.

 

[daveybrat]

Your results are perfectly normal for that cpu with the stock cooler.

Unless you plan on only running stress-testing tools on your cpu, the stock cooler and temps are fine.

I'll still go on the record and say that an aftermarket cooling solution would still be your best bet.

 

[deanx0r]

Some stress tests such as Prime or Linpack can be unnecessary hard on the CPU, and commonly used real world applications such as rendering or video encoding rarely push CPU that hard.

I recommend Intel's XTU to stress test your system. It's a good all in one application that doesn't unnecessary overheat your CPU.

https://downloadcenter.intel.com/SearchResult.aspx?lang=eng&keyword=xtu

 

[ShintaiDK]

tcase is a useless metric for you.

The reading you do is against tjmax that is 100C.

85C under prime95 is perfectly fine and with a large headroom.

Intel guarantees you it can run at 100C for 3 years. But in reality it could run 10+ years with that.

 

[BonzaiDuck]

tcase is a useless metric for you.

The reading you do is against tjmax that is 100C.

85C under prime95 is perfectly fine and with a large headroom.

Intel guarantees you it can run at 100C for 3 years. But in reality it could run 10+ years with that.

Why do you say that . . "useless metric?" They decide what is a useful or useless metric: they've dropped other specs we might think more important but they keep publishing TCASE? It's been established that it is some number below the TJunction or core average; in C2D/C2Q days, it had been estimated to be a delta of 10C.

Now, we're not sure. But it was supposed to be a basis for case design intended, I think, to preserve the electronics.

If, on the other hand, you are correct, then someone looking to risk more thermal stress might still exceed voltages that increase electro-migration. The high clocks will need more voltage, the temperatures will increase, more power will be consumed, and more voltage would be needed just because of the temperature.

 

[Jovec]

Why do you say that . . "useless metric?"

To measure Tcase you have to mill the top of the IHS from one edge to the center and place a temperature prob there to sit between the IHS and heatsink base. Useless you do that, Tcase is a "useless metric."

 

[escrow4]

Never use a stock cooler. Worthless. Chuck it and buy an aftermarket. Mystery why Intel even issues them anymore. A basic Coolermaster Hyper or Noctua will drop temps massively, way better for longevity.

 

[ShintaiDK]

Why do you say that . . "useless metric?" They decide what is a useful or useless metric: they've dropped other specs we might think more important but they keep publishing TCASE? It's been established that it is some number below the TJunction or core average; in C2D/C2Q days, it had been estimated to be a delta of 10C.

Now, we're not sure. But it was supposed to be a basis for case design intended, I think, to preserve the electronics.

If, on the other hand, you are correct, then someone looking to risk more thermal stress might still exceed voltages that increase electro-migration. The high clocks will need more voltage, the temperatures will increase, more power will be consumed, and more voltage would be needed just because of the temperature.

Tcase is for OEMs/Cooler designers and Jovec explained how to measure it. So for you, its utterly useless.

And ark.intel.com is an OEM portal.

And your C2D/C2Q info is wrong as well.

 

[you2]

I did infact already test the chip with that tool and it does run 5C cooler; the reason this is surprising to me is that my sandbridge processors (with stock cooler and overclocked a little) never ran this hot. Of course this processor is already @ 4ghz but I had expected the new processors to run cooler. The thread I had linked @ intel is really a different issue - some MB's are configuring the processor with too much voltage and several people have reported that disabling xmp and running at pc 1600 significantly reduced their heat.

Some stress tests such as Prime or Linpack can be unnecessary hard on the CPU, and commonly used real world applications such as rendering or video encoding rarely push CPU that hard.

I recommend Intel's XTU to stress test your system. It's a good all in one application that doesn't unnecessary overheat your CPU.

https://downloadcenter.intel.com/SearchResult.aspx?lang=eng&keyword=xtu

 

[you2]

If it can run 10 years @ 85C then I'm more than happy. Other than initial testing of ram 99% of my usage will have the processor in the 50-60c range. I loved those heat pipes AMD used to give ya with their processors.

To measure Tcase you have to mill the top of the IHS from one edge to the center and place a temperature prob there to sit between the IHS and heatsink base. Useless you do that, Tcase is a "useless metric."

 

[BonzaiDuck]

ShintaiDK observed that the "C2D/C2Q" information I had was wrong. I only remember getting that info from a white-paper published at Tom's Hardware about seven years ago. Either the white-paper was wrong, or newer processors have superseded the validity of the delta between "TCASE" and the cores. Granted, it isn't easily measured. I have no gripe with the methodology and pictures posted.

With the Sandy cores, I found that increasing voltage above 10C beyond TCASE still resulted in voltages I didn't want to live with. And I'm sort of a sissy about voltages: I'm uncomfortable if the unloaded "turbo" high voltage exceeds 1.40, or the severely loaded voltage exceeds 1.38.

Everyone, including IDontCare, is going to tell me that I should twist it up higher!!:biggrin: Maybe I will, but I get decent clocks anyway with my self-imposed limit.

There isn't anymore a published spec for "maximum safe voltage," and I'd assume if there is one, it would be lower for the 22nm chips.

On the other hand, I find it useful to take note of the TCASE spec as I try my best to get temperatures in the 80's C reduced to 70C or lower. It may not be a "useful metric," but it seems to me a "useful objective."