Reaching 90-100+ C in prime95 when OC'd?

[william.x]

Hi, I'm new to the forum. I recently built my first rig from the ground up. It's 5820K with Asus X99-Pro, I overclocked it around 4.28Ghz with the onboard Asus AI suite's auto overclocking.

When I ran the prime95 while staying OC'd, the CPU core temperatures instantly shot up to 90 within 30 seconds, then 100+ in less than 2 minutes. It scared the crap out of me, I turned it off very quickly. Later I figured it might be the overclocking, so I set the clock back to factory default and then ran it again. The temperature then started from 40s and only crept up to low 60s after a whole hour under full load. So I'm wondering if it's my loops incompetence or the bad application of the thermal gel? Bad contact between waterblock and CPU? Or stress test while OC'd is something that I'm not supposed to do?

It is fully watercooled, although it only has a 650ti I bought off a friend, since I'm waiting for the new flagships to come out.
It has 2 radiators, one 280mm alphacool monsta rad that's 80mm thick in the front and some no name super thin 360mm on top. I kind of cheaped out during the building process and decided to not get all the fans needed. So now there's only 2 original Fractal 140mm case fans in the front, and they are not high static, so I can't even feel the wind when I put my hand behind the 80mm rad which it should push through. Then there's the 2 high static pressure 120mm corsair fans I had before, which I mounted on the top radiator. So the rig only has 4 fans in total, lacking one on top and one in rear exhaust.

 

[Jovec]

Excessive voltage is probably the most likely factor, given that you used software / automatic overclocking.

Two main schools of thought here. 1) P95 will place an artificially heavy load on your CPU. Watch your temps under your normal workloads, and you'll likely find your temps lower (even for heavy load apps like video encoding, but excluding DC apps). If you are okay with this, then you can continue to use the Auto/Software OC.

2) Hand tune the BIOS settings for the lowest voltage and temps for your desired / max stable OC under P95 or other stress tester. This will cap your maximum temps to a known level you can be comfortable with, and again real world apps will be lower. If you run DC apps you'll want to do this, since they can hit the same temps at stress testers.

 

[Essence_of_War]

Modern versions of P95 use AVX2 instruction sets and causes the CPU to overvolt itself on top of whatever voltage you're set to, generating a ton of extra heat (and dramatically improving throughput!).

Since your numbers w/o overclocking seem reasonable, my guess is that it isn't bad thermal contact.

I would suggest not using AI suite's auto settings and OC'ing manually. Auto OC'ing is probably overvolting your CPU for the clock setting you're getting.

If you haven't tried it before, Intel's Extreme Tuning Utility is a pretty handy way to figure out your OC settings on the fly without having to boot into your bios everytime you're trying to change something.

 

[schmuckley]

Dat's tew hawt.
Lowur yur veecoar.
Manyewally if yew haf to.

Chek the mowent on yoor waterkewling.
redux it if need be.
Yew hav puhlentee of raads.
I just put one together and am getting 73c max @ max possible air/water OC for this chip.
It runs warmer than others from what I've seen.
That's 73c running prime95 with AVX as per XTU...@ 4722Mhz/1.323v

 

[flexy]

>>
Or stress test while OC'd is something that I'm not supposed to do?
>>

Nowadays, tests like OCCT, Prime etc. are really not useful anymore EXCEPT for "testing" how hot your CPU could possibly get under extreme situations which in "real life" would never occur. (Unless of course you plan to build a machine while would only run Prime...)

As others pointed out, the latest versions of Prime and some other "stress tests" are now also using AVX2 instructions which add another 0.1V to your Vcore voltage, driving temps up into insane ranges.

This is especially problematic if you use "adaptive voltage" on your board (which could very well be set as default) rather than setting voltages manually. "Adaptive" would just add voltage on top of what you already have depending on your CPU clock..and of course then STILL adds the 0.1 V once a stress-test uses AVX instructions such as the latest prime.

For more "real life" testing I recommend Asus "Real Bench 2" or "X264 Benchmark". TBH I am at a point where I say that any other test except the above is relatively pointless.

Just guessing here...it could be you get up to 100C with the latest Prime but might never exceed 85C with Real Bench V2 or X264 Benchmark...and then in "real real life", such as gaming etc. you will likely never see temps higher than 70s. So no point to worry about 100C in Prime...my $0.02

 

[william.x]

Modern versions of P95 use AVX2 instruction sets and causes the CPU to overvolt itself on top of whatever voltage you're set to, generating a ton of extra heat (and dramatically improving throughput!).

Since your numbers w/o overclocking seem reasonable, my guess is that it isn't bad thermal contact.

I would suggest not using AI suite's auto settings and OC'ing manually. Auto OC'ing is probably overvolting your CPU for the clock setting you're getting.

If you haven't tried it before, Intel's Extreme Tuning Utility is a pretty handy way to figure out your OC settings on the fly without having to boot into your bios everytime you're trying to change something.

Thanks, the Intel ETU is really helpful.

 

[monkeydelmagico]

another option is to use version 26.6 of Prime95

 

[Stuka87]

Yeah use version 26.6. Its a better tool for testing overclocks than the current version.

That or use OCCT with AVX turned off.

 

[UaVaj]

for true "100%" stability. aka stock stability.
there is absolutely no reason why you cannot run prime95 v28.5 AVX indefinitely.
if your overclock fails this, then clearly is it NOT 100%.


for "ENOUGH" stability. for everyday typical tasks.
then by all mean skip prime95 v28.5 AVX or OCCT AVX.
as long as your overclock can 1) runs the other lesser stressor fine and 2) does not crash in your gaming session.
then that can be consider good enough.



as for WHAT is good enough? everyone has to decide that for themselves.

 

[Techhog]

for true "100%" stability. aka stock stability.
there is absolutely no reason why you cannot run prime95 v28.5 AVX indefinitely.
if your overclock fails this, then clearly is it NOT 100%.


for "ENOUGH" stability. for everyday typical tasks.
then by all mean skip prime95 v28.5 AVX or OCCT AVX.
as long as your overclock can 1) runs the other lesser stressor fine and 2) does not crash in your gaming session.
then that can be consider good enough.



as for WHAT is good enough? everyone has to decide that for themselves.

But the current version is hindered by a bug in the CPU. Hardly anyone without very high-end cooling will be able to have "100% stability" on
Haswell under those conditions since it pushes voltage above the level actually being tested.