Skylake i7 Temps

[[DHT]Osiris]

All,

Looking for some anecdotal information regarding Skylake i7 temps. Have mine @4.5, 1.325v, with a thermaltake 3.0 ultimate cooler (the 360mm one). First foray into water cooling, and it's been about half a decade since I built a new system, much less tinkered with OC stuff/benchmarking/stress testing.

After a little research, I prodded at it with prime95, 8 threads @ 4K threw it to 100c so halted that immediately, 4 threads had it hovering at 90c at 1.35v, pulled down to 1.325v and it brought it to somewhere around 87c under load. 'Normal' stress testing (from like, CPU-Z's stresser) is more around 73-75c, from my understanding prime95 runs unusually hot due to the tests performed.

Idle temps are about 5c over ambient (which I found friggin fantastic, coming from a gen1 i7 and a cinderblock heatsink), and gaming doesn't turn my room into an oven, so I feel like I'm doing good for now.

Finally, is 1.325v roughly normal for a 4.5 OC on this chip? I tried to read on it, most of what I saw was 4.4 @ that voltage, but mine has seemingly been stable (though I haven't gone through the whole '12 hours of stress testing' thing yet).

Thank you!

 

[Techhog]

Newer versions of Prime95 throw AVX instructions on the CPU, which causes voltage to go far beyond what you set when using adaptive voltage. You should always used fixed voltage when using stress tests that go beyond real-world load.

 

[myocardia]

It has definitely become the norm in the OCing community since the debut of Haswell (the 4000 series, IOW) to no longer use Prime95 to check stability, except for those people who are overclocking to make their Prime95 performance better, ie the ones who run Prime95 24/7/365, not just people like you or I who would be using it only to check for stability.

All of the newer versions of Prime95 use the newest version of the AVX software. Like SSE before it, it adds massive amounts of performance, like enough so that a stock-clocked CPU would outperform/perform as well when using AVX2 as it would highly overclocked, when not using AVX2.

The problem is it stresses the CPU a considerable amount extra. Enough so that the first CPU which supported AVX2, the Haswell generation, the processors all needed an extra .1V, to remain stable. And they don't just require it when running Prime95, they require it when running any AVX2, which is still quite rare. As far as I can tell, Prime95 is still the only non-business software that uses it up until this point.

So, what is an overclocker to do? Well, use a program which will use the most stressful instruction set that is widely used (AVX1), and also will use as many cores/threads as your CPU happens to have. It's better when it's free, of course. That program today is Handrake. It's an H.264 video encoder. It works as well in 2015 for stability testing overclocks as the earlier versions of Prime95 did, and it's also available free of charge. Good luck.

 

[MongGrel]

Interesting, wasn't aware of this with Prime95 and the newer chips.

But is what this site is for I guess

 

[[DHT]Osiris]

Awesome information. At least it means I wasn't going crazy in regards to my temps. I'll check out handbrake, though I suspect it'll be about the same as the other stress testers I've used.

Sticking below 80c at full load a solid target with skylake? Haven't had problems in years with my core2quad or gen1 i7 using that logic. Pushing more for reliability than big numbers. Just wanted to push the envelope a smidge.

Thank you again!

 

[myocardia]

Awesome information. At least it means I wasn't going crazy in regards to my temps. I'll check out handbrake, though I suspect it'll be about the same as the other stress testers I've used.

Sticking below 80c at full load a solid target with skylake? Haven't had problems in years with my core2quad or gen1 i7 using that logic. Pushing more for reliability than big numbers. Just wanted to push the envelope a smidge.

Thank you again!

If you want to try a different stress test, whatever the latest version of Intel's 'Intel BurnTest' software also works fine, along with a few others that are similar, but whose names escape me at the moment!

As far as temps go, that's completely up to you. Anything under 100°C is technically safe for the CPU, as long as it isn't throttling. I like low numbers myself, but I doubt you'd find anyone who knows what they're talking about that would say up to 80°C is safe 24/7/365. <<----edit: Oops, I accidentally read this again, and that sentence there was supposed to end in "who would not agree that 80°C is safe 24/7/365"

 

[LTC8K6]

The tests in Intel XTU are really all you need, imo. If it passes those, I wouldn't worry about it.

 

[Dufus]

The problem is it stresses the CPU a considerable amount extra. Enough so that the first CPU which supported AVX2, the Haswell generation, the processors all needed an extra .1V, to remain stable.

Not all, see here.

And they don't just require it when running Prime95, they require it when running any AVX2, which is still quite rare.

That doesn't make sense as any time an AVX2 instruction was encountered would mean a stall of many thousands of cycles while the voltage was ramped up not only negating performance but actually degrading it by a huge amount.

 

[myocardia]

Not all, see here.

You're right, your laptop CPU only went up by 90% of .10V. Congrats. Of course, when AVX2 is active, your CPU power doubles, and your temps go up by 22°C, so I'm honestly not sure what your point happens to be.

 

[Dufus]

Your not reading the graph correctly. The idle adaptive voltage was 1.036V and it never went above that while running AVX2, instead voltage dropped a little, less than 10mV. BTW 10mV is 0.01V, have no idea where you got 90% of 0.1V (90mV).

 

[myocardia]

Your not reading the graph correctly. The idle adaptive voltage was 1.036V and it never went above that while running AVX2, instead voltage dropped a little, less than 10mV. BTW 10mV is 0.01V, have no idea where you got 90% of 0.1V (90mV).

Haha sorry, my brain doesn't operate in the laptop world. I saw the 1.027v and 1.036v, and my brain translated that into the desktop equivalent, 1.27v and 1.36v. As a matter of fact, when I opened your screenshot back up today, I had to look at it twice, even though I knew what it said, after having read your post.