Help! I cant even run prime 95 at stock speeds, i5 4690k.

[Tylermainia66]

I recently spent all of my money on a used I5 4690k on amazon and I put it in my rig and ran prime 95 small FFT for like 4 minutes before the temperature on core #0 spiked to a whopping 95 degrees! I thought I had applied to much thermal compound so I reapplied several times but I have had the same issue. My first core is running 15-20 degrees hotter than the others under load, so while the other 3 cores on the cpu are running just fine at low 70s under full load, the first core is getting hot enough to shut the whole system down. If I can't fix this then I wont even be able to think about an overclock as that would easily push the first core above the degree limit.
here is a photo of the aftermath of a temp spike (looking at max temps):

 

[aquielisunari]

You purchased a POS CPU. Maybe. This is the darkside of buying used electronics. Did you see it in action? Check its temps as you were testing it or was this basically a blind buy?

Which cooler are you using? I am using a 212 Evo with my 4690K.

 

[aquielisunari]

Sorry. I missed the Amazon. Bit tired over here. Not an excuse. Just a reason. I'd see about a refund or possibly an aftermarket cooler if you already aren't using one.

Temps during gaming?

 

[rhysiam]

Are you aware that Prime 95 (version 28 and later) utilise the AVX 2 instructions on Haswell and newer Intel processors. That generates a thermal load that is waaaaay beyond any vaguely normal CPU load. Newer Intel CPUs recognise this and have a feature called "AVX Offset" which will ensure the processor actually lowers the clockspeed and voltages under an AVX load. Your CPU doesn't have that feature however.

That explains the very high temperatures. The solution would be to use a different stress test like OCCT which presents a more realistic workload.

What this doesn't explain is why one core is so much hotter than the others. It is likely you lost the silicone lottery as @aquielisunari suggested. Still test it with a realistic workload and see what the temps look like.

Listing your CPU cooler would certainly help.

 

[CompuTronix]

Tylermainia66,

On behalf of Tom's Moderator Team, welcome aboard!

Although Digital Thermal Sensors (DTS) are factory calibrated by Intel, the specification for DTS accuracy is +/- 5°C. This means deviations between the highest and lowest Cores can be up to 10°C. If you're running at stock with Turbo Boost enabled, then deviations may exceed 10°C by a few more degrees, but certainly not by 15 to 20°C.

Regardless of whether you're running the stock cooler or a better aftermarket cooler, deviations between Cores of more than 10°C indicates that unfortunately, your particular 4690K sample most likely has a faulty factory application of Thermal Interface Material (TIM) between the Die and the Integrated Heat Spreader (IHS), which can only be corrected by "de-lidding".

Here's a YouTube - http://www.youtube.com/watch?v=XXs0I5kuoX4 - that shows before, how-to, and after, which perfectly describes the same high Core #0 temperature and deviation between Cores that you're experiencing while running Prime95.

If you're interesting in delidding but don't want to try it yourself, then you can send your 4690K in to Silicon Lottery - https://siliconlottery.com/collections/all/products/delid - which is a company that tests, bins and de-lids CPU's. Here's their figures on how much Core temperatures at 100% workload are improved by de-lidding:

7th Generation ... Kaby Lake - 12° to 25°C
6th Generation ... Skylake - 8°C to 18°C
5th Generation ... Broadwell - 8°C to 18°C
4th Generation ... Devil's Canyon - 7°C to 15°C
4th Generation ... Haswell - 10°C to 25°C
3rd Generation ... Ivy Bridge - 10°C to 25°C

De-lidding services are $49.99. They do a great job in a timely manner, and are reputable and highly recommended.

Here's the recommended operating range for Core temperature:

80°C Hot (100% Load)
75°C Warm
70°C Warm (Heavy Load)
60°C Norm
50°C Norm (Medium Load)
40°C Norm
30°C Cool (Idle)

Concerning Prime95, as rhysiam has pointed out, do not use versions of Prime95 later than 26.6 on 2nd through 7th Generation i3, i5 or i7 CPU's, which all have AVX (Advanced Vector Extension) Instruction Sets. More recent versions of such as 28.1 and 28.9 run AVX code on the CPU's Floating Point Unit (FPU) which causes unrealistic temperatures up to 20°C higher. The FPU test in the utility AIDA64 shows similar results.

For the record, there is nothing wrong with Prime95 version 26.6. Intel tests their processors under carefully controlled conditions at 100% Thermal Design Power (TDP). Prime95 Version 26.6 Small FFT's is the standard 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 at stock settings. No other utility so closely replicates Intel's 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

If you run AVX apps, then Asus RealBench runs a realistic AVX workload typically within +/- a few % of TDP at stock settings, however, it’s a fluctuating workload for stability testing, which isn’t very well suited for CPU thermal testing, so you need to observe "maximum" Core temperatures.

• Asus RealBench - http://rog.asus.com/rog-pro/realbench-v2-leaderboard/

Also, you might want to read this Sticky: Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

CT

 

[Tylermainia66]

You purchased a POS CPU. Maybe. This is the darkside of buying used electronics. Did you see it in action? Check its temps as you were testing it or was this basically a blind buy?

Which cooler are you using? I am using a 212 Evo with my 4690K.

Im using a 212 evo aswell. I don't know what a POS cpu is. but I bought it off of amazon with the knowledge that if it was "broken" i would be able to return it because amazon requires used sellers to have a return policy equal or greater than that of tehre own. it was a blind buy.

 

[Tylermainia66]

Are you aware that Prime 95 (version 28 and later) utilise the AVX 2 instructions on Haswell and newer Intel processors. That generates a thermal load that is waaaaay beyond any vaguely normal CPU load. Newer Intel CPUs recognise this and have a feature called "AVX Offset" which will ensure the processor actually lowers the clockspeed and voltages under an AVX load. Your CPU doesn't have that feature however.

That explains the very high temperatures. The solution would be to use a different stress test like OCCT which presents a more realistic workload.

What this doesn't explain is why one core is so much hotter than the others. It is likely you lost the silicone lottery as @aquielisunari suggested. Still test it with a realistic workload and see what the temps look like.

Listing your CPU cooler would certainly help.

When I try other workloads like cinibench my temps are great, like in the mid 50s. but if I cant be using prime 95 then how am I supposed to dial in my overclock? is OCCT just as good and reliable for stability testing? (hyper 212 evo is my cooler btw)
thanks.

 

[Tylermainia66]

Tylermainia66,

On behalf of Tom's Moderator Team, welcome aboard!

Although Digital Thermal Sensors (DTS) are factory calibrated by Intel, the specification for DTS accuracy is +/- 5°C. This means deviations between the highest and lowest Cores can be up to 10°C. If you're running at stock with Turbo Boost enabled, then deviations may exceed 10°C by a few more degrees, but certainly not by 15 to 20°C.

Regardless of whether you're running the stock cooler or a better aftermarket cooler, deviations between Cores of more than 10°C indicates that unfortunately, your particular 4690K sample most likely has a faulty factory application of Thermal Interface Material (TIM) between the Die and the Integrated Heat Spreader (IHS), which can only be corrected by "de-lidding".

Here's a YouTube - http://www.youtube.com/watch?v=XXs0I5kuoX4 - that shows before, how-to, and after, which perfectly describes the same high Core #0 temperature and deviation between Cores that you're experiencing while running Prime95.

If you're interesting in delidding but don't want to try it yourself, then you can send your 4690K in to Silicon Lottery - https://siliconlottery.com/collections/all/products/delid - which is a company that tests, bins and de-lids CPU's. Here's their figures on how much Core temperatures at 100% workload are improved by de-lidding:

7th Generation ... Kaby Lake - 12° to 25°C
6th Generation ... Skylake - 8°C to 18°C
5th Generation ... Broadwell - 8°C to 18°C
4th Generation ... Devil's Canyon - 7°C to 15°C
4th Generation ... Haswell - 10°C to 25°C
3rd Generation ... Ivy Bridge - 10°C to 25°C

De-lidding services are $49.99. They do a great job in a timely manner, and are reputable and highly recommended.

Here's the recommended operating range for Core temperature:

80°C Hot (100% Load)
75°C Warm
70°C Warm (Heavy Load)
60°C Norm
50°C Norm (Medium Load)
40°C Norm
30°C Cool (Idle)

Concerning Prime95, as rhysiam has pointed out, do not use versions of Prime95 later than 26.6 on 2nd through 7th Generation i3, i5 or i7 CPU's, which all have AVX (Advanced Vector Extension) Instruction Sets. More recent versions of such as 28.1 and 28.9 run AVX code on the CPU's Floating Point Unit (FPU) which causes unrealistic temperatures up to 20°C higher. The FPU test in the utility AIDA64 shows similar results.

For the record, there is nothing wrong with Prime95 version 26.6. Intel tests their processors under carefully controlled conditions at 100% Thermal Design Power (TDP). Prime95 Version 26.6 Small FFT's is the standard 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 at stock settings. No other utility so closely replicates Intel's 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

If you run AVX apps, then Asus RealBench runs a realistic AVX workload typically within +/- a few % of TDP at stock settings, however, it’s a fluctuating workload for stability testing, which isn’t very well suited for CPU thermal testing, so you need to observe "maximum" Core temperatures.

• Asus RealBench - http://rog.asus.com/rog-pro/realbench-v2-leaderboard/

Also, you might want to read this Sticky: Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

CT

im going to download v26.6 and see if I have the same issues or not. thank you.

 

[rhysiam]

When I try other workloads like cinibench my temps are great, like in the mid 50s. but if I cant be using prime 95 then how am I supposed to dial in my overclock? is OCCT just as good and reliable for stability testing? (hyper 212 evo is my cooler btw) thanks.

Well that's great news because it suggests your CPU is likely fine and ready to dial in an OC. It's a little strange that an Prime95 AVX2 workload pushes one core so much higher than the others, but given you'll probably never encounter such a workload in any real-world scenario it's not worth losing sleep over.

Lots of people have ditched Prime95 for stability testing primarily because of the unrealistic temperatures it results in. OCCT is another choice, though I've heard some (with more experience than me) say anecdotally that Prime95 tends to expose stability issues much quicker that other stress tests. An early P95 version is a good option. There aren't any hard-and-fast rules here, people have their own preferences for stress testing.

 

[CompuTronix]

It's a little strange that an Prime95 AVX2 workload pushes one core so much higher than the others ...

Although Digital Thermal Sensors (DTS) are factory calibrated by Intel, the specification for DTS accuracy is +/- 5°C. This means deviations between the highest and lowest Cores can be up to 10°C. If you're running at stock with Turbo Boost enabled, then deviations may exceed 10°C by a few more degrees, but certainly not by 15 to 20°C.

Regardless of whether you're running the stock cooler or a better aftermarket cooler, deviations between Cores of more than 10°C indicates that unfortunately, your particular 4690K sample most likely has a faulty factory application of Thermal Interface Material (TIM) between the Die and the Integrated Heat Spreader (IHS), which can only be corrected by "de-lidding".

Here's a YouTube - http://www.youtube.com/watch?v=XXs0I5kuoX4 - that shows before, how-to, and after, which perfectly describes the same high Core #0 temperature and deviation between Cores that you're experiencing while running Prime95.

Lots of people have ditched Prime95 for stability testing primarily because of the unrealistic temperatures it results in. An early P95 version is a good option.

Concerning Prime95, as rhysiam has pointed out, do not use versions of Prime95 later than 26.6 on 2nd through 7th Generation i3, i5 or i7 CPU's, which all have AVX (Advanced Vector Extension) Instruction Sets. More recent versions of such as 28.1 and 28.9 run AVX code on the CPU's Floating Point Unit (FPU) which causes unrealistic temperatures up to 20°C higher. The FPU test in the utility AIDA64 shows similar results.

For the record, there is nothing wrong with Prime95 version 26.6. Intel tests their processors under carefully controlled conditions at 100% Thermal Design Power (TDP). Prime95 Version 26.6 Small FFT's is the standard 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 at stock settings. No other utility so closely replicates Intel's 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

If you run AVX apps, then Asus RealBench runs a realistic AVX workload typically within +/- a few % of TDP at stock settings, however, it’s a fluctuating workload for stability testing, which isn’t very well suited for CPU thermal testing, so you need to observe "maximum" Core temperatures.

• Asus RealBench - http://rog.asus.com/rog-pro/realbench-v2-leaderboard/

There aren't any hard-and-fast rules here, people have their own preferences for stress testing.

rhysiam,

Our modern techno-world runs 24/7/365 on gadgets based on engineering standards and design specifications, so when discussing the topic of processor temperatures, it's extremely important to be very specific. We all remember science class where one of the guiding principles for conducting a controlled experiment, is that it's critical to set up the same conditions and follow the same procedures every time. This minimizes variables so results will be consistent and repeatable.

Here's some reasons why users find processor temperatures confusing:

• Terminology and specifications.
• Abundance of misinformation.
• Inconsistent test procedures.

Since many users take a haphazard approach to testing their rigs with "W" hardware at "X" Ambient temperatures using "Y" stress software with "Z" monitoring utilities resulting in "CPU" or "Package" or "Core" temperatures, it's extremely difficult to compare apples to apples, especially when these variables are not listed in detail.

“Load” or “full load” are popular but misleading user terms which are undefined variables that could mean anything. Gaming, applications, rendering, transcoding, virus scanning and web surfing are fluctuating workloads with fluctuating temperatures which aren’t suitable for testing thermal performance, or for accurately comparing Core temperatures. Also, 100% CPU utilization doesn’t always equal 100% CPU workload, which is 100% TDP.

Prime95's default test, Blend, is also 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 suitable 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 suitable 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 different 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 critical for thermal testing.

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. Although AIDA64's CPU test is a steady workload, it’s well below TDP, which is insufficient for thermal testing. All other AIDA64 CPU test combinations are fluctuating workloads, which again aren't suitable for thermal testing.

Please read this Sticky: Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

CT

 

[Tylermainia66]

Just Installed and ran prime 95 V26.6 and my temps have dropped from 95 on the hottest core to just around 60! Thanks so much to everyone who posted and gave me information regarding this topic.

 

[rhysiam]

@CT: I have great respect for your knowledge on this topic and fully concede I have only a fraction of your experience in this area. I work in research (different field), and in fact am about to start a new position lecturing university students in Introduction to Research, so I'm well aware of the scientific method.

What I would state, from my perspective, is that "stress-testing" for the purposes of consumer level overclocking on a gaming computer does not need the rigour of a scientific experiment. In fact, such rigour is nonsense in this context, because in a few months time the room might be a vastly different temperature, dust may have accumulated and the machine might be running different workloads.

Stress testing for the purpose of overclocking is ultimately trying to answer the following question: is my current overclock likely to be safe (temps okay) and stable (no errors/crashes/BSODs) for my system on the workloads that I use it for.

If, for example, someone has a 240hz monitor and only ever plays CS:Go on their PC, you could argue that the best stress-test for that person would be running CS:Go. A "stable" overclock for that person means it's stable for them 24/7. Now perhaps with that same overclock Prime95, OCCT, Aida64 and other games all crash within the first few seconds, but you could still argue that this person achieved "stable" overclock because it works on their workloads.

Different overclocking experts (and I'm absolutely not including myself in this category) recommend different stress-testing regimes. If I understand your post correctly, you're making a helpful distinction between "stability" stress testing and "thermal" stress testing, and pointing out that the tests which produce a steady thermal load are critical for thermal stress testing. That's helpful and all makes sense. In the end of the day though, surely whatever test/series of tests your decide to use, all you're trying to do is approximate, as quickly as possible, a worst-case workload (or series of workloads) for your system. The logic of the stress-test regime being: if it works in these worst-case scenarios it's **probably** going to work everywhere else too.
By that logic, a "good" stress test regime is simply one that closely resembles the worst-case scanario for that system.
A "bad" stress test would be one that presents an unrealistic worst-case load for that system: that could be too stressful a load (like the AVX2 Prime95 loads), or one that isn't stressful enough -> you mention Aida64 as an example here.

I suppose my questions are:
1) If OP's Core0 temps remain fine under fluctuating workloads (further testing would be required to confirm this - of course), and you list "gaming, applications, rendering, transcoding, virus scanning and web surfing" as fluctuating workloads... then is there actually a problem? i.e. what workloads are OP's system going to be realistically subjected to that involve steady thermal signatures?
2) If "steady thermal signatures" is likely in the real-world workloads, then what thermal stress test would you recommend OP use to explore the temperature difference between cores? (I'm guessing Prime95 FTT 26.6 - is that correct?)

 

[CompuTronix]

1) If OP's Core0 temps remain fine under fluctuating workloads (further testing would be required to confirm this - of course), and you list "gaming, applications, rendering, transcoding, virus scanning and web surfing" as fluctuating workloads... then is there actually a problem? i.e. what workloads are OP's system going to be realistically subjected to that involve steady thermal signatures?

From our Intel Temperature Guide, Section 12, Note 1: " ... a combination of stress tests, apps or games must be run to verify CPU stability." Respectfully, have you read our Sticky?

I understand the nature of the OP's problem, which is a common issue, and provided information as well as links for him to reach a solution. Most repetitive temperature questions posted on our Forums have answers which have already been addressed in our Sticky. In countless instances I've provided detailed answers to help others learn and understand, knowing full well that if they read the Stickies, their questions would be unnecessary.

The debate concerning stability testing, and what's considered warranted is highly controversial. I've extensively tested and analyzed every utility available. I know which procedures work, what's appropriate, what's excessive, and why. Nevertheless, there comes a point where regardless of the materials and proof presented, users either accept the recommendations, or just make it up as they go.

What I would state, from my perspective, is that "stress-testing" for the purposes of consumer level overclocking on a gaming computer does not need the rigour of a scientific experiment. In fact, such rigour is nonsense in this context, because in a few months time the room might be a vastly different temperature, dust may have accumulated and the machine might be running different workloads.

From my perspective, I for one subscribe to, and strongly advise others to follow a methodical approach, which as an engineer, I would never describe as "nonsense". I'm also aware of the effects of variables such as ambient temperature and workloads, which are covered in detail in our Temperature Guide, as well as proven methods for establishing valid thermal baselines.

2) If "steady thermal signatures" is likely in the real-world workloads, then what thermal stress test would you recommend OP use to explore the temperature difference between cores? (I'm guessing Prime95 FTT 26.6 - is that correct?)

The OP's answer:

Just Installed and ran prime 95 V26.6 and my temps have dropped from 95 on the hottest core to just around 60!

Intel's statement: Troubleshooting Intel® Core™ i7-4790K / i5-4690K overheating - https://communities.intel.com/docs/DOC-23517

"Overheating occurs very quickly on motherboard products with certain BIOS settings when running the Small FFT test that is part of the Prime95* application, version 28.5 ... Overheating is not as severe with earlier versions of Prime95 ... "

Even if users run versions of Prime95 later than 26.6, such as 27.7, 27.9, 28.1, 28.5 or 28.9, AVX can be disable by inserting "CpuSupportsAVX=0" in Prime95's "local.txt" file which is located in Prime95's Windows folder. However, temperatures will be exactly the same as version 26.6, so it's just easier to use 26.6.

If users run or plan to run AVX apps, then test with Asus RealBench (linked above in my 1st post), which provides a realistic but fluctuating AVX workload. Peak Core temperatures will be just a few degrees higher than P95 v26.6, but still much lower than later versions of P95. If users overclock, being AVX stable requires higher Vcore at lower Core speed to maintain stability with similar Core temperatures. This means the highest stable overclock will sacrifice at least 200 MHz compared to being non-AVX stable.

So if a user is an overclocker and a gamer, this can be a very relevant consideration. As you pointed out in your 1st post, later Generations have AVX offset adjustments in BIOS, but for those with 4th Generation processors, making use of different BIOS configurations saved in Profiles provides a work-around solution. If users overclock but don't run AVX apps, then there's no point in testing for it, just to end up settling for a lower overclock. Either way, systems need to be configured accordingly so sustained Core temperatures aren't in the 80's.

CT

 

[Tylermainia66]

,

Concerning Prime95, as rhysiam has pointed out, do not use versions of Prime95 later than 26.6 on 2nd through 7th Generation i3, i5 or i7 CPU's, which all have AVX (Advanced Vector Extension) Instruction Sets. More recent versions of such as 28.1 and 28.9 run AVX code on the CPU's Floating Point Unit (FPU) which causes unrealistic temperatures up to 20°C higher. The FPU test in the utility AIDA64 shows similar results.

CT

Any chance that the later versions of prime 95 have that same effect on the 4th gen G3258 pentium? I had one before the i5 and never realy had any overheating issues with it running prime, (most likely because its a cold dual core), but that doesnt mean it wasent running hotter than it should have been right? because I was never aware of what that chip was supposed to be running at, could it have been running 20degrees hotter than it should have been while stressing and I just didn't notice?

 

[CompuTronix]

Tylermainia66,

Core i Pentium and Celeron variants, as well as all Core i Previous (1st) Generation and Core 2 processors aren’t affected by the Prime95 AVX version issue since they don’t have AVX Instruction Sets.

CT