F5F Stay Refreshed Power Users Overclocking Speed decreases gradually after about 30 seconds

Speed decreases gradually after about 30 seconds

Speed decreases gradually after about 30 seconds

Pages (2): 1 2 Next
D
DrBsTigger
Member
59
11-19-2017, 04:54 PM
#1
I recently upgraded from a 3570K to an 8700K. I used Conductonaut to delidded it, then RTV to relidded it before connecting it to my H110i with Kryonaut. During overclocking, it easily reached 5GHz. I set the voltage to "offset" at 0.005 because I didn’t want the CPU voltage to auto-adjust. Corsair Link showed around 1.3v. When running prime95 with small FFTs, it initially spiked to about 85°C before settling near 80°C. After relidding and adding shims to the backplane, it dropped to a peak of 80°C and stabilized at 75°C. Still quite high, so I’d like to talk about that with you. Could the RTV be creating a gap between the die and spreader? I used the Der8aur kit. The main thing I observed is that it performs well at 5GHz but consistently drops to around 3-4GHz after roughly 30 seconds. The HW monitor doesn’t indicate any thermal throttling or power throttling. I adjusted the CPU power settings in BIOS—long-term peak at 200W, short-term at 250W, disabled speedstep, and set VRM current limit to 140%. The result was the same: solid performance at 5GHz, but clocks were reduced afterward. As a newcomer to these later chips, I’m curious if there’s something I’m missing about their turbo behavior.
D
DrBsTigger
11-19-2017, 04:54 PM #1

I recently upgraded from a 3570K to an 8700K. I used Conductonaut to delidded it, then RTV to relidded it before connecting it to my H110i with Kryonaut. During overclocking, it easily reached 5GHz. I set the voltage to "offset" at 0.005 because I didn’t want the CPU voltage to auto-adjust. Corsair Link showed around 1.3v. When running prime95 with small FFTs, it initially spiked to about 85°C before settling near 80°C. After relidding and adding shims to the backplane, it dropped to a peak of 80°C and stabilized at 75°C. Still quite high, so I’d like to talk about that with you. Could the RTV be creating a gap between the die and spreader? I used the Der8aur kit. The main thing I observed is that it performs well at 5GHz but consistently drops to around 3-4GHz after roughly 30 seconds. The HW monitor doesn’t indicate any thermal throttling or power throttling. I adjusted the CPU power settings in BIOS—long-term peak at 200W, short-term at 250W, disabled speedstep, and set VRM current limit to 140%. The result was the same: solid performance at 5GHz, but clocks were reduced afterward. As a newcomer to these later chips, I’m curious if there’s something I’m missing about their turbo behavior.

E
Elliepls
Member
222
11-23-2017, 08:00 PM
#2
Thank you for the details. The latest Prime95 version (29.4) demands around 130% of capacity, which isn't advised. This is why your core temperatures are higher than normal.

The acceptable core temperature range is:
- Above 85°C: not recommended
- Below 80°C: ideal
- Core temps change with ambient conditions
- Different loads affect performance differently. Stress tests fall into two types:
- Stability tests with varying workloads
- Thermal tests with constant workloads

Intel evaluates processors under these conditions...
E
Elliepls
11-23-2017, 08:00 PM #2

Thank you for the details. The latest Prime95 version (29.4) demands around 130% of capacity, which isn't advised. This is why your core temperatures are higher than normal.

The acceptable core temperature range is:
- Above 85°C: not recommended
- Below 80°C: ideal
- Core temps change with ambient conditions
- Different loads affect performance differently. Stress tests fall into two types:
- Stability tests with varying workloads
- Thermal tests with constant workloads

Intel evaluates processors under these conditions...

Y
YkseyHD
Member
59
11-30-2017, 07:55 AM
#3
Jonathanese :
... 8700K ... delidded it using Conductonaut, applied Kryonaut to link it with my H110i, operates smoothly at 5GHz with minimal issues... voltage adjustments to "offset" and setting it to 0.005... Corsair Link shows around 1.3v...
... during execution of prime95 with small FFTs, performance drops around 80C and stabilizes near 75C after a short time...
... at 5GHz, clocks are then reduced arbitrarily.
Jonathanese
,
(1) Which Prime95 version is being used?
(2) Is the "AVX Offset" setting in BIOS configured to "Auto"?
(3) What is the room temperature?
CT
Y
YkseyHD
11-30-2017, 07:55 AM #3

Jonathanese :
... 8700K ... delidded it using Conductonaut, applied Kryonaut to link it with my H110i, operates smoothly at 5GHz with minimal issues... voltage adjustments to "offset" and setting it to 0.005... Corsair Link shows around 1.3v...
... during execution of prime95 with small FFTs, performance drops around 80C and stabilizes near 75C after a short time...
... at 5GHz, clocks are then reduced arbitrarily.
Jonathanese
,
(1) Which Prime95 version is being used?
(2) Is the "AVX Offset" setting in BIOS configured to "Auto"?
(3) What is the room temperature?
CT

_
_IceSlide_
Junior Member
20
11-30-2017, 08:36 AM
#4
I'm not sure what to call it, but the first download link on the main page of GIMPS caught my attention. I kept everything unchanged and ran small FFTs to minimize variables.
2. I'm planning to investigate this further. I'll assume it's a possibility.
3. The ambient temperature has been around 75°F/23.9°C.
I recently repositioned the heat spreader without silicone, just using the conductonaut and applying superglue once assembled to reduce gaps. It seems the pressure was on the die this time, which is a positive sign. Allowing it to cure fully. This will be my final delid attempt since resolving the thermal issue would require addressing something else entirely. Plus, I'm uncertain if gallium might oxidize, but I don't want to risk damaging the CPU.
Also, this setup uses the ASUS TUF Z370. I need to track temperatures carefully to avoid damaging the VRM, if that's feasible on this chip.
_
_IceSlide_
11-30-2017, 08:36 AM #4

I'm not sure what to call it, but the first download link on the main page of GIMPS caught my attention. I kept everything unchanged and ran small FFTs to minimize variables.
2. I'm planning to investigate this further. I'll assume it's a possibility.
3. The ambient temperature has been around 75°F/23.9°C.
I recently repositioned the heat spreader without silicone, just using the conductonaut and applying superglue once assembled to reduce gaps. It seems the pressure was on the die this time, which is a positive sign. Allowing it to cure fully. This will be my final delid attempt since resolving the thermal issue would require addressing something else entirely. Plus, I'm uncertain if gallium might oxidize, but I don't want to risk damaging the CPU.
Also, this setup uses the ASUS TUF Z370. I need to track temperatures carefully to avoid damaging the VRM, if that's feasible on this chip.

R
ReyCr4ft
Junior Member
45
12-11-2017, 01:28 PM
#5
Thank you for the details. The recent update to Prime95 (version 29.4) demands a significant 130% workload, which isn't advised. This is likely why your core temperatures are higher than anticipated. The standard operating limits for core temperature are as follows: - Temperatures exceeding 85°C should be avoided. - Values under 80°C are preferred. - Core temps change in response to ambient conditions. - Different types of loads affect performance; stress tests involve fluctuating demands, while thermal tests use consistent loads. Intel evaluates processors by applying a constant 100% TDP workload to confirm thermal specs. Prime95 version 26.6 offers Small FFT's for CPU thermal evaluation, providing a stable 100% load. This configuration closely mimics Intel’s testing environment, making it the best choice for accurate core temperature readings. Other tools that don’t stress or under-stress the processor will deliver reliable baseline data. Utilities that manage load effectively will help you obtain a trustworthy thermal reference. Below is a comparison of tools categorized by thermal and stability tests, based on TDP percentages across six processor generations at default settings, rounded to the nearest 5%: Higher TDP scenarios lead to increased core temperatures. All evaluations display 100% CPU utilization in Windows Task Manager, reflecting actual processor usage rather than TDP percentage. Although real power consumption (in Watts) changes with speed, voltage, and workload, Prime95 v26.6 Small FFT's consistently delivers a stable 100% load—whether you’re using stock or overclocked settings. Prime95 v26.6 - http://www.mersenneforum.org/showthread.php?t=15504 Note: Models from the 2nd to 8th generation i3, i5, and i7 feature AVX instruction sets. Versions of Prime95 after 26.6 incorporate AVX in their CPU Floating Point Unit (FPU), which can result in temperatures up to 20°C higher than expected under excessive TDP loads. Other high-TDP tools also yield similar outcomes. AVX functionality can be turned off in later Prime95 releases by adding "CpuSupportsAVX=0" to the local.txt file within Prime95’s directory. However, since core temperatures remain consistent with version 26.6, sticking with that release is simpler. AVX does not impact CPUs from the first generation, Core 2, Pentium, or Celeron lines, as they lack AVX support. If you’re running an overclocked setup and using applications like rendering or transcoding, consider lowering Vcore and core speed, or upgrading your cooling solution and case fans to prevent temperatures from reaching 85°C. Many 6th, 7th, and 8th generation boards resolve the AVX issue through BIOS offset adjustments—typically setting an offset of -2 (200 MHz) is sufficient. A helpful resource at the top of the CPU forum is recommended: Intel CPU Temperature Guide
R
ReyCr4ft
12-11-2017, 01:28 PM #5

Thank you for the details. The recent update to Prime95 (version 29.4) demands a significant 130% workload, which isn't advised. This is likely why your core temperatures are higher than anticipated. The standard operating limits for core temperature are as follows: - Temperatures exceeding 85°C should be avoided. - Values under 80°C are preferred. - Core temps change in response to ambient conditions. - Different types of loads affect performance; stress tests involve fluctuating demands, while thermal tests use consistent loads. Intel evaluates processors by applying a constant 100% TDP workload to confirm thermal specs. Prime95 version 26.6 offers Small FFT's for CPU thermal evaluation, providing a stable 100% load. This configuration closely mimics Intel’s testing environment, making it the best choice for accurate core temperature readings. Other tools that don’t stress or under-stress the processor will deliver reliable baseline data. Utilities that manage load effectively will help you obtain a trustworthy thermal reference. Below is a comparison of tools categorized by thermal and stability tests, based on TDP percentages across six processor generations at default settings, rounded to the nearest 5%: Higher TDP scenarios lead to increased core temperatures. All evaluations display 100% CPU utilization in Windows Task Manager, reflecting actual processor usage rather than TDP percentage. Although real power consumption (in Watts) changes with speed, voltage, and workload, Prime95 v26.6 Small FFT's consistently delivers a stable 100% load—whether you’re using stock or overclocked settings. Prime95 v26.6 - http://www.mersenneforum.org/showthread.php?t=15504 Note: Models from the 2nd to 8th generation i3, i5, and i7 feature AVX instruction sets. Versions of Prime95 after 26.6 incorporate AVX in their CPU Floating Point Unit (FPU), which can result in temperatures up to 20°C higher than expected under excessive TDP loads. Other high-TDP tools also yield similar outcomes. AVX functionality can be turned off in later Prime95 releases by adding "CpuSupportsAVX=0" to the local.txt file within Prime95’s directory. However, since core temperatures remain consistent with version 26.6, sticking with that release is simpler. AVX does not impact CPUs from the first generation, Core 2, Pentium, or Celeron lines, as they lack AVX support. If you’re running an overclocked setup and using applications like rendering or transcoding, consider lowering Vcore and core speed, or upgrading your cooling solution and case fans to prevent temperatures from reaching 85°C. Many 6th, 7th, and 8th generation boards resolve the AVX issue through BIOS offset adjustments—typically setting an offset of -2 (200 MHz) is sufficient. A helpful resource at the top of the CPU forum is recommended: Intel CPU Temperature Guide

W
War91
Member
186
12-17-2017, 02:26 PM
#6
Check if there are any adjustments you can make concerning the slowdowns. Are these issues connected to overload caused by prime?
W
War91
12-17-2017, 02:26 PM #6

Check if there are any adjustments you can make concerning the slowdowns. Are these issues connected to overload caused by prime?

X
197
12-19-2017, 10:38 AM
#7
I own an 8700K at 5.0, currently paired with a front-mounted H105 in my personal setup, which I had delidded with conductonaut. My stable Vcore stands at 1.344, meaning my CPU needs a bit more power than yours. At 22°C ambient temperature, the core temperature is around 69°C with P95 v26.6 Small FFTs. It will hit approximately 150 Watts... while later versions (v29.4) will surpass 190 Watts, marking a significant jump, all thanks to AVX.

AVX demands substantial hardware resources and calls for a lower frequency and Vcore to avoid excessive power draw and core temperatures, ensuring stability. When activated in BIOS, AVX Offset achieves just that; it reduces clock speeds across your rig, whether it's running at default or overclocked, whenever AVX code is active. Other tasks without AVX support—such as gaming and applications—will still operate at full stock or overclocked levels in BIOS.

The extent of the downclock varies based on your configuration; options like Auto, 0, 1, 2, etc., are typical. A setting of 2 (200MHz) generally works well for common AVX tasks like rendering and transcoding.

Read the Guide... the details are there.
X
x_HmOoDy_KsA_x
12-19-2017, 10:38 AM #7

I own an 8700K at 5.0, currently paired with a front-mounted H105 in my personal setup, which I had delidded with conductonaut. My stable Vcore stands at 1.344, meaning my CPU needs a bit more power than yours. At 22°C ambient temperature, the core temperature is around 69°C with P95 v26.6 Small FFTs. It will hit approximately 150 Watts... while later versions (v29.4) will surpass 190 Watts, marking a significant jump, all thanks to AVX.

AVX demands substantial hardware resources and calls for a lower frequency and Vcore to avoid excessive power draw and core temperatures, ensuring stability. When activated in BIOS, AVX Offset achieves just that; it reduces clock speeds across your rig, whether it's running at default or overclocked, whenever AVX code is active. Other tasks without AVX support—such as gaming and applications—will still operate at full stock or overclocked levels in BIOS.

The extent of the downclock varies based on your configuration; options like Auto, 0, 1, 2, etc., are typical. A setting of 2 (200MHz) generally works well for common AVX tasks like rendering and transcoding.

Read the Guide... the details are there.

S
Stratin_OG
Member
232
12-21-2017, 10:20 AM
#8
I have a 8700k running at 5ghz @ 1.27v. I haven't done any tuning. Just using a 240mm liquid cooler. Cinebench r15 temps are under 80c. All AAA games are running at 5ghz with under 70c. I haven't tried prime95 or any other AVX workloads because gaming won't impact that much.
S
Stratin_OG
12-21-2017, 10:20 AM #8

I have a 8700k running at 5ghz @ 1.27v. I haven't done any tuning. Just using a 240mm liquid cooler. Cinebench r15 temps are under 80c. All AAA games are running at 5ghz with under 70c. I haven't tried prime95 or any other AVX workloads because gaming won't impact that much.

F
Flames4Evr
Junior Member
14
12-23-2017, 03:21 AM
#9
That's another example of how you don't do things. How do you know your system is stable? What because it boots into windows and you can fire up some games?
The danger of untested overclock that if there is hidden instability then you will pile up errors overtime which will result in things starting to crash eventually.
F
Flames4Evr
12-23-2017, 03:21 AM #9

That's another example of how you don't do things. How do you know your system is stable? What because it boots into windows and you can fire up some games?
The danger of untested overclock that if there is hidden instability then you will pile up errors overtime which will result in things starting to crash eventually.

P
PimQ1
Member
163
12-23-2017, 05:51 AM
#10
I avoid CPU heavy work. I just play games. Everything is working well. There are some hidden issues in the games, but we can adjust the settings if needed.
P
PimQ1
12-23-2017, 05:51 AM #10

I avoid CPU heavy work. I just play games. Everything is working well. There are some hidden issues in the games, but we can adjust the settings if needed.

Pages (2): 1 2 Next