Is 85C fine (safe) for CPU overclocking target?

[GoldenSun3DS]

https://pcpartpicker.com/user/GoldenSun3DS/saved/F76KZL

I have the i3-8350K and I've got it stable at 4.5GHz* 1.325V on Prime95 Blend for 3 hours before my PSU died (defective PSU most likely since a different PSU is able to provide power).

*Technically 4.8GHz with 3 AVX offset, but Prime95 Blend was using AVX and that caused it to run at 4.5GHz. I was going to test what stable non-AVX speed I could get after that before my PSU died.


It was reaching a peak of 95C 5-10 seconds at a time, with temperatures otherwise fluctuating between 82C to 92C while stress testing.
But that's just stress testing, so standard gaming should be more like 80-85C or less.

So, is it safe to run the temperature at 85C+ while gaming, or would this just cause it to die really quickly in like a month? I wouldn't really mind if the lifespan became just a few years instead of like 6 years.

On a side note, can disabling a core improve temperature at the same voltage? Or would I need to decrease voltage and run new stress tests to determine stability?

 

[Darkbreeze]

Ok I guess it was another moderator then sorry

No, it WAS I that said that in the other thread you are talking about. You are right about that.

However, I did NOT change my story in this thread. When I said, in THIS thread, 15-20 minutes, that was in regard to thermal testing, not in regard to stability testing. Again, those are two different factors that simply happen to have a relationship based on voltage. If your system is EVER going to overheat running Prime95 26.6, it WILL do it within 20 minutes. At least that is what I have found from MANY, MANY Prime runs on a variety of generations.

Stability testing though is an entirely different ballgame and in order to determine, as realistically as possible, that the system will be capable of being essentially stable for all types of uses, you need to run the gamut of FFT lengths in order to do so.

 

[Karadjgne]

DB, I don't find simultaneous testing a useless endeavor. I've tested ram at stable, tested cpu at stable, tested everything under the sun as stable, all individual tests. Then ran a simultaneous test and got errors. On my personal pc. Turned out to be some interaction between the ram OC and cpu OC which was only found by using linpack on OCCT in an otherwise stable individual environment.

The way I see simultaneous testing is its like nothing more than taking the heaviest, most punishing gaming engine, and adding weight to it.

And yes, it did take a lot of guesswork to find and fix as there was no stability issues to start with on individual OC. The fix was bumping current from 110% to 120%. Pc hasn't crashed in over 4 years now.

It's easy enough to drive on bald tires. It's easy enough to drive in the rain. Driving on bald tires in the rain is a different beast altogether. It introduces variables that don't really exist under individual stressing.

 

[Darkbreeze]

Well, as with most things, there is NOTHING that works 100% for every person in every situation. No doubt about that. But generally speaking the standard methodology works as it should in most cases. Still, those methods are only meant to be a baseline approach that delivers an acceptable result MOST of the time. There are always exceptions and like I always say, you can test everything for five days individually without issue and then end up with a problem anyhow from some unexpected source like a faulty or cheap O power supply. Or you could do as you've preferred and test using something like Realbench and still have an issue, then need to go test it all individually anyhow to figure out where the problem is, not find it, and have it turn out to be some other thing like power supply, drive problem, bad connection.

Everything is simply meant to reduce the probability of something being a problem but in reality we know there are never any guarantees of any kind, even if you don't overclock at all. So most the time I also like to do as you do, but after I've tested things individually. I still recommend that specific methodology though because in MORE cases than not, if there is a problem it will be found AND you can at least minimize the potential for issues once you do something like Realbench by tuning things down to a fairly close approximation of "good" before running it. To each their own though.

Realistically, if all you did was game and were ok with reinstalling every few months, you wouldn't actually need to do anything except some basic tests like you say at all anyhow because errors would never get the chance to accumulate to a critical level in the first place. I think it's dangerous though if there's any chance you'll save and reuse any of the data or files that were used during that period of time, after reinstalling though.

 

[CompuTronix]

Hi Guys,

Merry Christmas and Happy Holidays!

I'd like to weigh in on this Thread and offer some clarifications, insights, opinions and references to processor temperatures and test methods. We can all agree these are long standing controversial and complex topics that involve Intel's specifications, which can be about as clear as mud. Often, applying a little common sense is prudent, so it's extremely important when discussing these topics to be very specific.

1st, users introduce confusion and conflicting information concerning stress testing by not differentiating between stability testing and thermal testing.

2nd, users often proceed on false assumptions when ambient temperature is neither stated nor asked for. People write into our Forums from all over the planet during any seasons and climatic environments who are operating desktop computers in temperatures ranging from 10°C (40°F) to 40°C (104°F). Threads might exceed 20 posts before it occurs to anyone that ambient temperature hasn't yet been established, when Standard or "normal" is 22°C (72°F). Ambient temperature is a HUGE variable which needs to be established at the beginning of the Thread, along with hardware specifications, test software and monitoring utilities.

3rd, users introduce further confusion and conflicting information concerning Prime95 by failing to specifically mention and differentiate between which VERSION is being used. This divides into two groups; VERSIONS up through 26.6 (Pre-AVX) and VERSIONS later than 26.6 (AVX). Further, many users fail to mention or don't know that Small FFT's is a steady 100% workload used for CPU testing, and the default test, Blend, is a fluctuating workload used for memory testing, thus the need to be very clear and specific.

Central to this discussion is when it's appropriate to run tests that use Advanced Vector Extensions (AVX), which is code that's used for accelerating number crunching such as in rendering and transcoding software ... AND ... Prime95 VERSIONS LATER THAN 26.6, which once again needs to be very clearly and specifically stated, as feelinfroggy777 neglected to mention, as well as countless others when answering Threads on our Forums.

• This topic involves standards and specifications. I can not overemphasize how critical it is to be extremely clear and specific at all times, so as to minimize confusion in an already complex and confusing topic.

Here's some background information concerning AVX:

Intel introduced AVX with 2nd Generation Sandy Bridge processors, which increases the workload on the processor's Floating Point Unit (FPU) or number cruncher, thereby increasing Core temperatures by ~4°C. 3rd Generation Ivy Bridge processors also support AVX, but showed more increase in Core temperatures, since 3rd and later Generations of mainstream processors no longer have a soldered Integrated Heat Spreader (IHS), but instead use Thermal Interface Material (TIM or "thermal compound" or "paste") between the Die and the IHS.

Core temperature characteristics changed with 4th Generation Haswell and Devil's Canyon which introduced AVX2. This dramatically increased number crunching capabilities, and Core temperatures by ~20°C. Remember all the 4770K and 4790K high Core temperature Threads from that era? 5th Generation quickly and quietly came and went when 6th Generation was launched just 2 months later, however, the AVX problem persists throughout today's 7th and 8th Generation processors.

The differences are that AVX / Core temperature problems are most prevalent in higher TDP processors with Hyperthreading, such as i7's. i5's are less affected as they don't have hyperthreading, and i3's are lower TDP which are much easier to cool. However, the i3 8350K is the same processor as the i5 7600K, which needs delidding and big air or liquid to run cool when highly overclocked. AVX doesn't affect Core i 1st Generation, Core 2, Pentium or Celeron processors since they don't have AVX Instruction Sets.

Following the Core temperature issues with 4th Generation processors, Intel responded to the AVX2 problem by having motherboard manufacturers implement "AVX Offset" adjustments into BIOS with 6th Generation and later processors. However, some AVX Offset adjustments don't work, or work properly, or need a BIOS update to work properly. Regardless, many users are unaware of what AVX offset is for, or how to configure it, so the Prime95 VERSION issue persists.

AVX offset allows gamers to run a higher overclock, which will automatically downclock during AVX workloads to maintain Core temperatures and to ultimately prevent thermal Throttling, especially if your cooling solution is marginal in a high ambient environment. Typically a -2 or -3 (-200 or -300MHz) offset is adequate to maintain reasonable Core temperatures.

• AVX can be disabled in Prime95 versions later than 26.6 by simply inserting "CpuSupportsAVX=0" into the "local.txt" file in Prime95's folder. However, since Core temperatures will be the same as 26.6, it's easier to just use 26.6.

If you game and don't use AVX apps, then there's no point in testing for it, unless you want your rig to be "AVX stable" if you choose to run AVX apps at some point in the future. If you DO run AVX apps, then by all means you must test for it, but keep in mind that being AVX stable means that you must be willing to accept a lower overclock to match the Core temperatures of a non-AVX rig. This is where AVX offset comes into use if you're running a 6th through 8th Generation processor.

If you're running a 4th Generation Processor, there are no AVX offsets in BIOS. If you game and run AVX workloads, you must configure your overclock to accommodate AVX. This can be accomplished by creating different BIOS Profiles for gaming and AVX apps, but requires rebooting to switch between Profiles, which is a bit cumbersome and inconvenient. For 3rd and 2nd Generation, AVX has much less impact on Core temperatures.

Here's an Intel link which discusses Prime95. Although somewhat vague concerning versions, the information still applies to the latest processors: Troubleshooting Intel® Core™ i7-4790K / i5-4690K overheating - https://communities.intel.com/docs/DOC-23517

Q: So what's too hot? Here's the Core temperature scale from my Intel Temperature Guide:

Core temperatures above 85°C aren't recommended.

As shown above, 80°C is hot, so I personally don't like see any Core temperatures in the 80's.

As a footnote, keep in mind that Intel's "Tcase" Thermal Specification is NOT Core temperature, which applies to 6th Generation and earlier processors. Tcase is a factory only measurement performed on the external surface of the IHS on Engineering Samples for developing specifications using the stock cooler. Tcase is NOT a thermal limit, and is a misleading specification. Intel's "Tjunction" Thermal Specification, which is "Tj Max" or "Throttle" temperature, is instead the limiting specification, and is most commonly a Core temperature of 100°C.

Here's a rare and revealing video interview between Tom's Hardware and Intel concerning Core temperatures related to overclocking on 4th Generation processors, which targets 80°C. This generally holds true for prior, as well as later Generations of silicon: Intel Discusses i7 4790K Core Temperatures and Overclocking - https://www.youtube.com/watch?v=BGTnJkuqlbo

For everyone's benefit, 100% CPU Utilization seldom equals 100% Workload, which is 100% Thermal Design Power (TDP). Intel tests their processors using Intel motherboards, on an open bench without a case, under carefully controlled conditions at 100% TDP, to develop and validate thermal specifications that are consistent and repeatable. So when thermal testing your rig, the goal is to not exceed 80°C when running at a steady 100% TDP with case covers removed, and all fans (and pump if liquid cooled) at 100% RPM.

For the record, there is nothing wrong with Prime95 VERSION 26.6, which is the latest Pre-AVX version. Although there are other utilities which are better suited for stability testing, Prime95 Small FFT's is ideal for CPU thermal testing, because it's a steady 100% workload with steady Core temperatures that typically runs Core i variants with Hyperthreading and Core 2 processors within +/- a few % of TDP. No other utility so closely replicates Intel's proprietary test conditions. This is also the utility that Real Temp uses to test Core temperature sensors.

• Prime95 v26.6 - http://www.mersenneforum.org/showthread.php?t=15504

To elaborate on a quote mentioned by darkbreeze from a statement I made a few years ago, "How can anyone extrapolate accurate Core temperatures from workloads that fluctuate like a bad day on the Stock Market?"

Take a look:

Prime95's default test, Blend, is a fluctuating workload for testing memory stability, and Large FFT's combines CPU and memory tests. As such, Blend and Large FFT's both have fluctuating workloads which aren’t well suited for CPU thermal testing.

Other stability tests such as Linpack and Intel Burn Test have cycles that peak at 120% workload, which again aren’t well suited for CPU thermal testing. The test utility OCCT runs elements of Linpack and Prime95, which will terminate the CPU tests at 85°C.

The "Charts" in SpeedFan span 13 minutes, and show how each test creates distinct thermal signatures.

Shown above from left to right: Small FFT's, Blend, Linpack and Intel Burn Test.

Note the steady thermal signature of Small FFT's, which allows accurate measurements of Core temperatures. A steady 100% workload is crucial for thermal testing so CPU, cooler, socket, motherboard, VRM's and power delivery components can heat-soak and thermally stabilize.

Shown above from left to right: Small FFT's, Intel Extreme Tuning Utility CPU Test, and AIDA64 CPU Test.

Intel Extreme Tuning Utility is also a fluctuating workload at ~90% TDP for overclocking and system tuning. Although AIDA64's CPU test is a steady workload, it's far below TDP at ~60%, which is insufficient for thermal testing. All other AIDA64 CPU test combinations are fluctuating workloads, which again aren't well suited for thermal testing. Also, AIDA64 is not Freeware, so the Trial version expires.

Further, I'm not a big fan of AIDA64 for thermal testing for several other reasons. Although AIDA64 is good for stability testing, the many different test combinations of fluctuating CPU, FPU, Cache and Memory workloads don't provide any conclusive results with respect to thermal testing. Additionally, the FPU test when run individually, will instantly slam your Cores to Throttle temperature.

Tj Max specifications may vary with TDP. Certain low TDP processors Throttle below 85°C (185°F) but many 3rd Generation processors Throttle at 105°C (221°F). Low TDP Core i 6th, 7th and 8th Generation CPU's have Configurable TDP (cTDP) and Scenario Design Power (SDP) which can trigger Throttling below Tj Max.

Here's where a little common sense applies:

Although most processors Throttle at 100°C (212°F), it’s not advisable to run your CPU near the thermal limit, just as you wouldn't run a vehicle with the temperature gauge pegged in the red "hot" zone. If your hottest Core is near it's specified Tj Max Throttle temperature, then your CPU is already too hot.

The consensus among highly experienced and well informed system builders and overclockers, is that cooler is better for ultimate stability, performance and longevity. Experts agree it's prudent to observe a reasonable thermal margin below Tj Max. So regardless of environmental conditions, hardware configurations, workloads or any other variables, Core temperatures above 85°C aren't recommended.

feelinfroggy777, although you've made some astute observations and also provided some good advice on several Threads I've read, some of your inputs are based upon misconceptions, misinformation and incomplete or partial information, which others find conflicting, confusing and misleading, especially concerning your Prime95 blanket statements without reference to VERSION.

Here's another example of misinformation:

You've stated that the Throttle temperature for the OP's i3 8350K is 95°C, which is incorrect.

At 95C, your CPU is thermal throttling.

The 8350K has the same TDP and Thermal Specifications as the i5 7600K, both of which throttle at 100°C:

i3 8350K - https://ark.intel.com/products/126689/Intel-Core-i3-8350K-Processor-8M-Cache-4_00-GHz
i5 7600K - https://ark.intel.com/products/97144/Intel-Core-i5-7600K-Processor-6M-Cache-up-to-4_20-GHz

You're not the first Member to make these mistakes, so please don't think you're being singled out. However, and with respect, if you're not sure, then spend some time to become better informed, and look it up before you offer further comments on future Threads. Please keep in mind that we at Tom's endeavour to provide the most complete and accurate information for our Members and readers.

Misinformation spreads like wildfire and is difficult to dispel. Our Forum Rules ask that Members read the Stickies, which are presented for everyone's knowledge and benefit. Regardless, many Members don't take the time to read the information which some of us have invested thousands of hours of painstaking research and hands-on testing over the course many years, to provide and maintain accurate and updated content for our Members and readers.

If you'd really like to get yourself up to speed on this topic, then please read this Sticky: Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

CT

 

[Darkbreeze]

So, that's one of the three people I was talking about. Might want to defer to his good judgement.

Also, I added a couple of edits to one of my earlier posts, in a nod to statements made by Karadjne which in retrospect have some relevance and merit, but also because I simply should have included those additional statements to begin with.

 

[HOkay]

I'd personally stay a little lower on temps, go for 85C max for stress tests. I run Cinebench 10 times or so & if that's ok on temps & doesn't crash you're probably game stable.
For you, I'd suggest coming down to 4.7 GHz or maybe 4.6 GHz as I suspect you'll be able to reduce the voltage a bit & get those temps under control, & you'll not notice any difference between 4.6 & 4.8 anyway. Also, a cheeky delid does wonders for the temps (20-25C reduction on my 8700k, & I could then reduce voltage a little for the same OC too!), but that's not for everyone though.

 

[Darkbreeze]

That sounds great, but doesn't entirely make sense. Delidding a CPU and replacing the thermal compound automatically reduces temperatures, but it does not create a more stable CPU or overclock, so it would not inherently allow you to "reduce voltage a little for the same OC".

The whole point is that you will be able to run the CPU at the voltage that is necessary to keep it stable WITHOUT having to reduce voltage, because reducing voltage increases instability most of the time, and up to a certain point.

The fact that you could mistakenly make such a monumentally opposing comment cannot help but create doubts in my mind as to the validity of your above statement, but since I am unfamiliar with you I'll simply leave it at that.

 

[william p]

What was the original question?
Is 85*C. a safe temp?
The general consensus is - not really. But workable when doing actual thermal testing. For this CPU.
Is that what the OP can take away from all this?

 

[Karadjgne]

If he has too. Generally you'll want to keep thermal maximums like under p95 @70ish,which puts heavy game loads @55ish. If max thermals are @85ish, op'll be heavy gaming @65ish for extended periods of time. While it won't hurt the cpu to run at those temps, it's more wear and tear on fan bearings, more power used overall, and a good way to keep your toes warm. Many gpus with semi fanless modes don't even kick in their fans till the gpu hits @65ish, so it's possible op'll get constant on/offs, which can get annoying, not to mention the added fan noise at higher temps. Dropping cpu temps even 10°C can significantly lower case temps, which only helps with airflow, gpu temps, hdd/ssd temps from radiated heat etc.

85°C is a workable temp max, just not a good one to have.

 

[william p]

If you drop CPU temps doesn't the heat go from the CPU into the case? That would raise case air temps, and require faster fan speed to remove the hot air. Wasting electricity and heating up the room. Causing noise and wear on the fan bearings.

 

[Darkbreeze]

NO William, if you DROP CPU temps, then there is less heat to begin with, therefore less heat to have to dissipate and remove, period. Sheesh.

I guess it depends on what your definition of "drop" is. If you are "dropping" by increasing the CPU or case fan profile to something that runs faster all the time, then yes, there might be increased wear and tear on a few things. If you "drop" CPU temps by reducing your overclock or voltage, if that can be done while still being able to maintain acceptable stability, OR by adding hardware that can do a BETTER job, then you REDUCE the total amount of heat and you INCREASE the life of those components because there is LESS heat that has to be removed.

 

[USAFRet]

If the CPU is doing more work (higher GHz or whatever), it is generating more heat.

Either that heat stays in the CPU 70-80-85-90...and still rising.
or, you do something better with your cooling situation, and move more of that off the CPU. 90-80-70...

The goal is to have the CPU run at whatever performance you desire (and that it can manage), and at a temperature that will not kill it.

But whatever happens, that heat is generated, and must move out into the environment.

 

[cranmerman]

Basing your system performance on a program like PRIME is a short sighted answer.
It is like jetting your carburetors for 8.000 feet when you never drive above sea level.
A complete waste of headroom..
Esp since gaming you are actually sat there (its not going to blow up!)
Yes, I will run prime and laugh.
Intel state 1.35 v and 72C for most recent chips,but you want to skew your own testing with prime.
Be my guest.

 

[Darkbreeze]

I guess you don't realize that moderators can still see your deleted posts.

 

[Karadjgne]

Go ahead and laugh. P95 is worst case scenario ceiling. I wanna know if when I do an hour long render on my i7 that it'll still be within acceptable temps. Knowing where your ceiling actually resides is far better than playing guessing games.