8700k overclocking: What are safe Vcores? How long to run stress tests?

Toggle sidebar Toggle sidebar
M

mjbn1977

Distinguished
Aug 20, 2015
944
116
19,490
Hi, I am currently in the process of overclocking my 8700k. They goal is to find a super stable 24/7 overclock at 4.8Ghz and one "stable enough" for gaming (if extra fps are required) that will not be used 24/7 at 5.0Ghz (maybe with an AVX offset of -2). I am using the following stress tests: prime95 current version, prime95 26.6, Intel Extreme Tuning Utility, and Cinebench.

The questions I have
1. what are Vcores for the 8700k would be considered good and safe for 24/7 OC?
2. what Vcores for the 8700k would you consider safe for occasional OC?
3. what Vcores would be considered unsafe and should not be applied during gaming sessions?
4. how long should I run stress tests for a) first evaluation of stability and b) final last test before using it as 24/7 overclock
5. how do the different LLC settings affect the overclock stability. should I try more or less?
6. which temperature is the temp I actually should monitor during the stress test in HWmonitor?

What I figured out so far:

It seems that my system is stable at 4.8Ghz on all cores with Vcore 1.27 and LLC 4. Can run all above benchmarks 30 mins without any errors or problems . At Vcores below 1.27 I will have some workers stop in the latest version of prime95 small FFP test. QUESTION 1: will adjusting the LLC doing anything against that?

I can run Cinebench and IET at 5.0Ghz and Vcore 1.35 and LLC 4 (no AVX offset) without problems. I can even run both test together. Prime95 26.6 runs also fine for at least 30 min. But current version prime95 stops workers and the temperatures are insane (mid to high 90s). prime95 26.6 runs at mid to high 80s. Would adding an AVX offset of -2 make those settings considered "game stable"?

Thank you!
 
I think you scored a very good chip.
Here are some stats from silicon lottery:
as of 11/11/2017
What % of I7-8700k chips can oc
at a agressive vcore of 1.4 or so and delidded
4.9 100%
5.0 81%
5.1 58%
5.2 30%
5.3 6%

I would leave your OC as is. 1.27 is excellent, You will not notice a extra multiplier or two.

If you implement speedstep, the vcore and multiplier will drop when the cpu has little to do; that is a good thing.
 


Thank you. What setting in MSI bios/uefi would be the speedstep. Right now my the clock does go down in idle and lower loads, but the vcore stays at 1.27. I wouldn't mind the vcore to go down on lower loads.

Also, were did you get this numbers/percentages from?
 


Thank you, very good article. But this doesn't answer question 4 and 5. The article clearly states that the described test are for temp testing not stress testing. But still a very good and interesting read.
 
Silicon lottery is a outfit that bins processors and sells the better ones for a premium.
That is my source. Note that these numbers are for delidded chips. Delidding voids your intel warranty, but lets the chips run cooler.

I think you are looking in the bios for something called adaptive voltage.
It may be called something else on your motherboard.
 
For stability i like doing an evaluation with a single run of Cinebench.
As for the final test: one program can be stable 100% stable while another program can crash in 5 seconds after starting.
i'd suggest just running Cinebench few times, few mins of Cpu-z stress test and few mins of Aida 64 stress test. If those are stable then try some games and if there's no crashes withing 2 days, then you have a pretty stable oc.
 
mjbn1977,

Stability testing has always been a highly controversial and hotly debated topic. Opinions vary greatly regardless of the utility in question. Silicon Lottery, mentioned by geofelt, uses Asus RealBench for stability testing. They had previously stated on their website that they run it for a minimum of 1 hour, however, anything anyone claims will be countered with an argument supporting longer time durations up to 24 hours, and / or a completely different approach. I firmly endorse and highly recommend Asus RealBench:

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

Finstar,

Here’s a sample of test utilities shown according to % of TDP at Default BIOS settings:

TDP … Thermal Test - Steady Workload

129% … Prime95 v29.3 - Small FFT’s (AVX, No Offset)
101% <-- Prime95 v26.6 - Small FFT’s
89% … HeavyLoad v3.4.0.234 - Stress CPU
87% … FurMark v1.19.1.0 - CPU Burner
78% … CPU-Z v1.82.0 - Bench - Stress CPU
66% … AIDA64 v5.95.4500 - System Stability Test - Stress CPU
55% … Intel Processor Diagnostic Tool v4.0 - CPU Load

TDP … Stability Test - Fluctuating Workload (Peak)

123% … OCCT v4.5.1 - CPU: OCCT (AVX, No Offset)
118% … LinX v0.6.5 - Default
116% … IntelBurn Test v2.54 - High
113% … OCCT v4.5.1 - CPU: Linpack (AVX, No Offset)
110% … AIDA64 v5.95.4500 - System Stability Test - Stress FPU
99% <-- Asus RealBench v2.56 - Stress Test (AVX, No Offset)
94% … Sandra 2017.09.24.41 - Burn in - Processor Tests
92% … CineBench v15.0 - CPU - Render Test
79% … Intel Extreme Tuning Utility v6.4.1.15 - CPU Stress Test

All tests will show 100% CPU Utilization in Windows Task Manager, regardless of actual Workload. Higher TDP produces higher Core temperatures. Results will vary according to Microarchitecture, Core count, Cache, Core speed, Turbo Boost, Core voltage, Hyperthreading, Instruction Sets, BIOS versions and CPU microcode.

Nevertheless, as you can see above, the CPU Render Test in CineBench is a lower TDP workload than Asus RealBench, as is the Bench Stress CPU Test in CPU-Z, which is only ~78% workload. That's a relatively low workload, which gives the user a false sense of security due to correspondingly low Core temperatures. Prime95 version 26.6 Small FFT's is a correct and proper 100% workload for thermal testing. 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

CT :sol:
 


And yet an oc that 1h of prime 95 showed to be fully stable crashed within 5 mins of Aida 64? (i5 6600k)
The same cpu also reached higher temps within a single run of Cinebench than 10 mins of Real Temp stress.
But sure, that could just be coincidence so let's look at my old i5 4690k.
I ran 1h of prime 95 on it without crashes but still it would crash in Dota 2 every couple of hours with only 60% cpu load.

You claim to be a specialist but you fail to see how different architectures have different weak points.
Sure prime 95 is good for testing max temps and the motherboards power delivery but it's practically useless when determining if a cpu is reaching it's physical limits.
You should use as many different stress tests as possible to test as many weak points as possible.
 
Sorry, but you appear to be missing the point.

I made it clear that Prime95 v26.6 Small FFT's is ideal for THERMAL testing, and Asus RealBench is ideal for STABILITY testing.

I completely agree with you that there are better choices for stability testing than Prime95 v26.6 Small FFT's. I have also stated in the Intel Temperature Guide that a combination of stress tests, apps or games must be run to verify CPU stability.

Respectfully, your statement that I "claim to be a specialist" implies that I am incompetent, which I find inconsiderate, disrespectful and insulting. So, if you intend to debate this topic with me, then know this; I have been continually working on this topic for nearly 11 years. During that time I have invested over 5,500 hours of painstaking research and meticulous hands-on testing to support my "claim". I am also the author of the Intel Temperature Guide, which is a fact; not a "claim".

The Guide is a Sticky near the top of the CPU's Forum, which is also indexed by Google. Our Forum Rules asks that Members read the Stickies before offering comments. If you were unaware of the Guide and haven't yet read it, then please take this opportunity to do so, as it's possible that you might find some useful information within. There's a link in my signature.

CT :sol:
 


I'm not doubting your knowledge of thermals, it's just that thermals are just an unwanted byproduct caused by the friction of electrons moving in the silicon.

Crashes while overclocking are usually caused by not enough voltage existing to deliver a signal to the next register.
More voltage is needed when there's defects in the silicon creating more resistance.
Let's say there's a defect over a area of the cpu that handles a certain instruction, that instruction will require more voltage to complete making it a weak point. This is why it's important to test with as many programs as possible to locate these weak points and to make sure they can operate as they should.
 
The purpose of Load / Line Calibration (LLC) is to minimize voltage overshoot and undershoot (surges and sags) as CPU load changes. For example, if you measure the voltage in your house at the wall socket, when a heavy load switches on such as an air conditioning compressor, you'll see a momentary but significant voltage sag due to "inrush current" which quickly settles at a slightly lower voltage than you initially measured. When the load switches off, the opposite occurs. This is normal and expected.

Since your CPU responds instantly to changes in load, a significant sag in Core voltage, sometimes called "Vdroop", represents a potential software crash point (BSOD), especially on an overclocked computer. LLC attempts to smooth out power delivery and Core voltage / load transitions as much as possible. If LLC is set too low, there's less safeguard against crashes. If LLC is set too high, the processor is subjected to voltage "spikes", which also can slightly affect Core temperatures in an adverse manner.

Further, the more power phases your motherboard has, the smoother the power delivery to the CPU. Then factor in SpeedStep and Speed Shift to the equation, and LLC becomes heavily tasked, as Core voltage fluctuates with random changes in load, in as little as 15 milliseconds, from about 0.700 to as high as 1.400 or so, and Core speed transitions from 0.800GHz to your overclock at 4.8GHz.

This applies to 8th Generation processors as well.

Also, Core voltage corresponds with microarchitecture, yet some users are under the misconception that one-size-fits-all. As a rule, CPU's become more susceptible to Electromigration (degradation) with each Die-shrink, so Core voltages must decrease accordingly. However, the most notable exception is Intel's 14 nanometer microarchitecture, where advances in FinFET technology have improved voltage tolerance.

As such, 14 nanometer silicon should typically not exceed 1.4 Vcore, which some hardcore overclockers may consider to be somewhat conservative, especially when running a high-end custom loop with comfortably low Core temperatures. Nevertheless, for the majority of gamers, transcoders and power users, 1.4 Vcore remains a reasonable limit for 14 nanometer processors.

Here’s a list of maximum recommended Vcore settings over the last decade of Intel processors:

Core

8th Generation 14 nanometer ... 1.400 Vcore
7th Generation 14 nanometer ... 1.400 Vcore
6th Generation 14 nanometer ... 1.400 Vcore
5th Generation 14 nanometer ... 1.400 Vcore
4th Generation 22 nanometer ... 1.300 Vcore

Legacy Core

3rd Generation 22 nanometer ... 1.300 Vcore
2nd Generation 32 nanometer ... 1.400 Vcore
1st Generation 45 nanometer ... 1.400 Vcore

Core 2 45 nanometer ... 1.400 Vcore
Core 2 65 nanometer ... 1.500 Vcore

Concerning AVX, if you game and don't use apps that run AVX, then there's no point in experimenting with offsets for AVX stability, unless you plan on using AVX sometime in the future. However, a setting of -2 or -3 (200 to 300MHz) is usually sufficient for AVX stability, as well as controlling Core temperatures during AVX.

On a final note, since the Digital Thermal Sensors (DTS) are located near the transistor Junctions at the heat sources within the Cores where temperatures are highest, the most critical processor temperatures to monitor are, of course, Core temperatures. "Package" temperature is typically the hottest Core, as is "CPU" temperature, which is often shown on a motherboard's two or three digit debug display.

CT :sol:
 


So, I am currently using on all my different overclocking test runs LLC 4 on my MSI gaming pro carbon. I notice that on most stress test the vcore is about 0.01 to 0.01V higher than what I set the Vcore in Bios/Uefi. For example: I set vcore to 1.24V with LLC 4 in adaptive mode @4.8 Ghz. During RealBench the Vcore is 1.264V. I this caused by my LLC 4 setting? In regards to LLC, I can set it to lower (let's say 5-8 LLC) for stable less aggressive OCs, and set it higher between 1-4 with more unstable and more aggressive OCs?

Also, right now I am still working to find the perfect 4.8Ghz vcore for my system. I noticed that benchmark results fluctuating around on the same 4.8Ghz with different voltage settings. Is this normal fluctuations? are a few % changes in benchmarks normal expectation? The one benchmark with the biggest inconsistency is the RealBench Encoding benchmark. sometimes I get 170.000, sometimes one 163.000. Is this expected?

Should I expect differences in benchmark performance when overclocking with adaptive power settings vs overclocking without adaptive powersettings? I like the idea of using adaptive for my 24/7 overclock.

 
mjbn1977,

(1) Your Vcore example above increased by 24 millivolts (1.240 vs 1.264). 24 millivolts is still reasonable. The goal is zero change, but any change should be in a slightly positive direction; not negative.

Note: Intel moved the voltage regulators onto the processor package for 4th Generation processors, which carried over to socket 1151 compatible 5th Generation processors.

The advantage is rock-steady Vcore. What you set in BIOS is what you get, regardless of whatever stress tests, apps or games are run, with +/- zero, or near zero millivolts fluctuation. The disadvantage is that the proximity of the regulators to the die, unfortunately has the effect of increasing Core temperatures by a few degrees, just as if the processor actually has a higher TDP. So for 6th Generation and later processors, Intel moved the regulators back onto the motherboard.

(2) Vcore is by far the most critical value governing stability. All other tweaks have a much less pronounced effect.

(3) Fluctuations in benchmarks are normal and expected. That's why reviewers, like those here at Tom's, will typically run at least 3 passes on each specific test, then calculate the average, which is the value shown in the review. For example, many reviewers state in their review in a section like "How We Test", their methodologies such as the number of passes that were run, ambient temperature, and any standards and specifications which apply.

(4) The difference between 170,000 and 163,000 is only ~4%, which is again within normal and expected variances. Keep in mind that RealBench is a full system test, so variances are not exclusively attributable to the processor.

(5) Differences in voltage settings using Adaptive, Fixed, Offset, Offset + Adaptive or Auto typically have a negligible effect on benchmarks, as long as the processor is not throttling and is stable.

CT :sol:
 
rofl OP must feel really overwhelmed with all this :pt1cable:
well the way you overclock the 8700k is differnt from each ones preference. and the motherboard and the cooler. alot of factors to consider. some say 1.35. some say 1.45. some have AVX offset of 12. some prefer it to 1. some like Adaptive offset mode. some prefer to turn on the Turbo Enhanced\MCE on the z370 boards. some prefer off.
 


Thank you for taking the time to answer in this much detail. I learned a lot from you answers. I know established following testing routine when I try a new overclocking setting:

1. I run the CPU-Z benchmark
2. I run 3x Cinebench
3. I run RealBench for 60 min
4. I run prime95 V26.6 for 60 min

Can I assume that if my OC config passes all those runs with out errors, that my system is pretty stable?
 
mjbn1977,

In my 1st post, there's a listing of utilities according to % of TDP. Note that CPU-Z's test is a relatively light workload, so I would skip it, as it's least likely to expose any instabilities, nor is it useful for thermal testing. I would instead focus on running utilities closer to 100% TDP.

When you're satisfied with your O/C settings, for your final stability tests you can add to your list AIDA64's CPU, FPU cache and memory tests run separately, then simultaneously, as well as your most demanding apps and / or games. Also useful, but not shown in the % of TDP list, are Futuremark's tests such as 3DMark, 3DMark Vantage and 3DMark 11, which will typically crash an unstable CPU during the Physics tests. Unigine's utilities such as Heaven, Valley and Superposition will also crash an unstable CPU.

Keep in mind that when running these utilities, if your GPU and memory are also overclocked, remember to restore them back to stock settings. In the event that they're not completely stable, they won't become yet another set of variables. This will save you from needlessly chasing your tail while attempting to expose, troubleshoot and isolate CPU instabilities, intermittent or otherwise.

Also, there's a correlation between ambient temperature and stability. Processors which were overclocked and stable during winter months when ambient temperatures are lower, may yet reveal intermittent instabilities during summer months when ambient temperatures increase, even in air conditioned environments set at 20°C in winter and 25°C in summer. An increase of just 5°C can reveal instabilities previously undetected.

Basic overclocking techniques strongly recommend that CPU, RAM and GPU overclocks be tested separately for stability, then tested simultaneously, which means four test sessions. Establishing system overclock stability requires a bit of work, but consider it as building a reliable infrastructure. If settings remain unchanged and there are no hardware failures, you only have to invest the time once. Then you can sit back and enjoy your rig with peace of mind, knowing the job was well done.

CT :sol:
 
i ran my 2600k@1.38v for 5 1/2 years on the edge of stability. it would 0x124 about once or twice a year and no matter how much i toyed with it. i always ended up at the same vcore for 4.5 and 4.6ghz as that was what my chip wanted. stability is just an illusion.
 
1. what are Vcores for the 8700k would be considered good and safe for 24/7 OC? 1.35v is what der8aur recommends as a starting point for 5.0ghz
2. what Vcores for the 8700k would you consider safe for occasional OC? Depends on the CPU. There's really only a general range. You'll have to fine tune it yourself.
3. what Vcores would be considered unsafe and should not be applied during gaming sessions? Don't go over 1.4v if you're new to overclocking.
4. how long should I run stress tests for a) first evaluation of stability and b) final last test before using it as 24/7 overclock. I generally do quick 10 round runs with Intel Burn Test to get a rough idea of what works and what doesn't. Then I do a final run Prime95 for a couple hours, and Intel Burn Test at 50 or so runs.
5. how do the different LLC settings affect the overclock stability. should I try more or less? Load Line Calibration can improve stability by reducing drops in voltage. refer to this article https://www.msi.com/blog/why-llc-is-your-friend-when-overclocking
6. which temperature is the temp I actually should monitor during the stress test in HWmonitor? You should consider several. The package temp is one. You should also monitor the core temps to be sure they aren't drastically different from each other. This can be a symptom of bad thermal paste application, or bad paste underneath the IHS.
 


Thanks for the detailed answer. What would be considered drastic in terms of temperature difference between different cores? When I run prime95 with AVX I have a max temperature difference between my core 1 (which runs the coolest of all 6) and core 2 (which runs the hottest) of about 8C - 10C. would the be considered drastic?

 


That would be perfectly reasonable, especially if your CPU has not been delid. If one or two cores were throttling, or borderline TJ Max, that would be cause for concern.
 
mjbn1977,

This is from Section 4 in the Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

" ... Intel’s specification for Digital Thermal Sensor (DTS) accuracy is +/- 5°C. Although sensors are factory calibrated, this means deviations between the highest and lowest Cores can be up to 10°C. Deviations on processors using Turbo Boost might exceed 10°C by a few degrees. Sensors are typically more accurate at high temperatures to protect against thermal damage, but due to calibration issues such as linearity, slope and range, idle temperatures may not be very accurate. ... "

Also, see Section 5.1.5.2.1, Digital Thermal Sensor Accuracy (Taccuracy), 8th Gen (S-platform) Intel® Processor Family Datasheet Vol.1 -
https://www.intel.com/content/www/us/en/processors/core/8th-gen-processor-family-s-platform-datasheet-vol-1.html

" ... The error associated with DTS measurements will not exceed +/-5°C within the entire operating range. ... "

If your highest and lowest Cores are within 10°C of one another, then your processor is in spec. If it's not, then you have grounds for an RMA.

CT :sol:
 


Thanks Man! I've read your excellent temp guide twice and still finding new things! Awesome!
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom