8700k overclocking: Which Vcores are secure? How extended should stress tests last?
8700k overclocking: Which Vcores are secure? How extended should stress tests last?
CompuTronix :
Sorry, but your message seems off track.
I wanted to emphasize that Prime95 v26.6 Small FFT is best suited for THERMAL testing, while Asus RealBench is more appropriate for STABILITY testing.
I fully understand your point that there are superior options for stability testing compared to Prime95 v26.6 Small FFT. I have also mentioned in the Intel Temperature Guide that a mix of stress tests, applications, or games should be conducted to confirm CPU stability.
Respectfully, your claim that I "claim to be a specialist" suggests you may not fully grasp the subject, which I find disrespectful and inappropriate. If you wish to discuss this further, please be aware that I have dedicated nearly 11 years to this area. During that time, I have spent over 5,500 hours on thorough research and detailed testing to support my expertise. I am also the author of the Intel Temperature Guide, which is a verified resource—not just a statement.
The Guide is prominently displayed near the top of the CPU's Forum and is indexed by Google. Our Forum Rules require members to read the Stickies before commenting. If you are unfamiliar with the Guide and have not yet reviewed it, I encourage you to do so, as it may contain valuable insights. A link to it is included in my signature.
CT
I don’t question your understanding of thermals; what I mean by that is that they are often an unintended consequence of electron friction within the silicon.
Crashes during overclocking typically result from insufficient voltage to send a signal to the next register.
Increased voltage is necessary when there are defects in the silicon that introduce more resistance.
Suppose a defect exists in a specific area of the CPU handling a particular instruction. That instruction will demand higher voltage to execute, making it a potential weak point. Therefore, comprehensive testing with various programs is essential to identify and address these vulnerabilities, ensuring optimal performance.
Wow, thanks for the feedback. This lively conversation mostly addressed my concerns. The only thing left is... anyone interested in trying question 5?
Mjbn1977:
Wow, thank you all. This lively conversation has addressed most of my queries. The only one left is... anyone interested in trying it on question 5?
Thanks!
The goal of Load / Line Calibration (LLC) is to reduce voltage overshoot and undershoot—those sudden spikes and dips—in response to changes in CPU load. For instance, measuring the voltage at a wall socket in your home reveals a brief but noticeable dip when a heavy device like an air conditioner turns on, caused by the "inrush current" that quickly stabilizes to a slightly lower level. Conversely, when the device powers down, the opposite happens. This behavior is normal.
Given how quickly your CPU reacts to load variations, a significant drop in core voltage—known as "Vdroop"—can signal a risk of software crashes, particularly on overclocked systems. LLC works to stabilize power delivery and manage core voltage shifts as much as possible. Setting LLC too low weakens the protection against crashes, while setting it too high exposes the processor to voltage spikes that may also raise core temperatures negatively.
Moreover, the greater the power phases your motherboard supports, the more consistent the power supply becomes for the CPU. When combined with features like SpeedStep and Speed Shift, LLC plays a crucial role in maintaining stable core voltage and speed transitions, which can occur rapidly—sometimes within 15 milliseconds—as core voltage swings from around 0.700 to as high as 1.400 or more, while core speed jumps from 0.800GHz to your overclocked 4.8GHz. This applies to all 8th Generation processors.
It's worth noting that core voltage is tied to the microarchitecture, yet some users mistakenly think a universal setting works for everyone. Generally, CPUs are more prone to issues like electromigration with each die-shrink, so voltages should be adjusted accordingly. Notably, Intel’s 14nm design benefits from advanced FinFET tech, allowing better voltage tolerance.
For most gamers, transcoders, and power users, a core voltage of around 1.4V is reasonable for 14nm chips. Here’s a summary of recommended Vcore limits over the past decade:
Core
8th Gen 14nm... ≤1.400 Vcore
7th Gen 14nm... ≤1.400 Vcore
6th Gen 14nm... ≤1.400 Vcore
5th Gen 14nm... ≤1.400 Vcore
4th Gen 22nm... ≤1.300 Vcore
Legacy Core
3rd Gen 22nm... ≤1.300 Vcore
2nd Gen 32nm... ≤1.400 Vcore
1st Gen 45nm... ≤1.400 Vcore
Core 2 45nm... ≤1.400 Vcore
Core 2 65nm... ≤1.500 Vcore
Regarding AVX, if you primarily play games and avoid apps that use AVX, experimenting with offsets for stability isn’t necessary unless you anticipate using AVX later. A setting of -2 or -3 (200–300MHz) usually suffices for stability and temperature control.
Finally, because Digital Thermal Sensors (DTS) are placed near transistor junctions in hot spots, the most important temps to watch are core temperatures—often the "package" temp—and CPU temperature, which is typically displayed on a motherboard’s debug screen.
CompuTronix :
mjbn1977 :
Thank you all for your input. This lively conversation has addressed most of my queries. The only remaining question is whether anyone would like to try it on question 5?
Appreciate the support!
The goal of Load / Line Calibration (LLC) is to reduce voltage overshoot and undershoot—those sudden spikes and dips in voltage as CPU load shifts. For instance, when a heavy appliance like an air conditioner turns on, you’ll notice a brief but noticeable drop in voltage due to the "inrush current," which quickly stabilizes at a slightly lower level than before. When the device powers down, the opposite happens. This behavior is typical and normal.
Given how quickly your CPU reacts to load changes, a significant dip in core voltage—often referred to as "Vdroop"—can signal a risk of software crashes, particularly on overclocked systems. LLC works to stabilize power delivery and manage core voltage shifts as much as possible. If LLC is set too low, the protection against crashes is weakened. Conversely, if it’s set too high, the processor faces voltage spikes that can also influence core temperatures negatively.
Moreover, the greater the power phases your motherboard supports, the more consistent the power supply becomes for the CPU. When combined with features like SpeedStep and Speed Shift, LLC takes on a heavier role, as core voltage can swing rapidly—sometimes by up to 15 milliseconds—from around 0.700 to 1.400 volts, while core speed jumps from 0.800GHz to overclocked levels at 4.8GHz. This effect is especially relevant for the 8th Generation processors.
It’s worth noting that core voltage is tied to the microarchitecture, yet some users mistakenly assume a universal setting. Generally, CPUs are more vulnerable to electromigration with each die-shrink, so voltages should be reduced accordingly. However, Intel’s 14nm design stands out due to its FinFET technology, offering better tolerance.
Typically, 14nm silicon should stay below 1.4Vcore, which some advanced overclockers might view as a cautious limit—especially when maintaining low core temps in custom builds. For most gamers, power users, and transcoders, 1.4Vcore is still a sensible upper bound for 14nm chips.
Here’s a summary of recommended Vcore ranges over the past decade for Intel processors:
CT
I’m currently using LLC 4 on my MSI Gaming Pro Carbon during various overclocking tests. Most stress tests show a vcore around 0.01–0.01V higher than what I set in BIOS/UEFI. For example, with LLC 4 at adaptive mode and 4.8GHz, the measured vcore was 1.264V. Could this be due to my LLC setting? Regarding LLC, I can lower it (e.g., 5–8) for more stable, less aggressive builds, or raise it slightly (1–4) for more unstable, aggressive overclocking.
I’m still searching for the ideal 4.8GHz vcore for my system. I’ve noticed benchmark results varying around 4.8GHz with different voltage settings. Is this normal? Are a few percent of variation typical? The RealBench Encoding benchmark is particularly inconsistent, sometimes showing 170 vs. 163. Is this acceptable?
Should I expect benchmark performance differences when overclocking with adaptive power settings versus without them? I prefer adaptive for continuous overclocking.
The Vcore example you provided increased by 24 millivolts, from 1.240 to 1.264. This change remains within an acceptable range. The aim is to avoid any shift, though a slight positive adjustment is preferable; negative shifts are undesirable.
Note: Intel relocated the voltage regulators onto the processor package for 4th Generation chips, which was also adopted in 5th Generation models compatible with socket 1151.
The benefits include a stable Vcore. The BIOS settings you configure directly determine the outcome, regardless of stress tests, applications, or games executed—resulting in minimal to no fluctuation, ideally near zero.
The drawback lies in the fact that placing the regulators close to the die tends to raise core temperatures slightly, similar to a higher TDP processor. Consequently, for 6th Generation and newer chips, Intel moved these regulators back onto the motherboard.
Benchmark variations are typical and acceptable. Reviewers generally perform multiple passes—often three or more—for each test, then average the results, which is reflected in the published scores. For instance, many reviewers detail their approach in sections like "How We Test," including factors such as pass count, ambient temperature, and relevant standards.
Variations between 170,000 and 163,000 represent only about a 4% difference, which falls within normal expectations. Remember that RealBench evaluates the entire system, so differences aren't solely due to the processor itself.
Changes in voltage settings (Adaptive, Fixed, Offset, Offset + Adaptive, or Auto) usually have little impact on benchmarks, provided the processor remains stable and doesn’t throttle.
rofl OP should definitely feel quite stressed with all this. The process of overclocking the 8700k is different for everyone, and there are many components to think about. Some recommend a 1.35, others 1.45, some mention an AVX offset of 12, while others prefer a lower setting or the Adaptive offset mode. Some like enabling Turbo Enhanced MCE on Z370 boards, and some prefer leaving it off.
CompuTronix :
mjbn1977,
Your Vcore example above increased by 24 millivolts (1.240 vs 1.264). This change is still within an acceptable range. The aim is to eliminate any shift, but any variation should be slightly upward; not downward.
Note: Intel relocated the voltage regulators onto the processor package for 4th Generation chips, which was also adopted in 5th Generation processors compatible with socket 1151.
The benefit is a rock-solid Vcore. What you configure in BIOS determines the outcome, no matter what stress tests, applications or games are executed, maintaining +/- zero or nearly zero fluctuations. The downside is that bringing the regulators closer to the die raises core temperatures slightly, as if the processor had a higher TDP. Consequently, for 6th Generation and beyond, Intel moved the regulators back onto the motherboard.
(2) Vcore remains the most important factor influencing stability. All other adjustments have a much smaller impact.
(3) Benchmark variations are normal and anticipated. That’s why reviewers, such as those here at Tom's, usually perform at least three passes per test, then compute the average, which is the figure reported in their review. For instance, many reviewers detail their methods—number of passes, ambient temperature, relevant standards—and state this in sections like "How We Test."
(4) The gap between 170,000 and 163,000 is only about 4%, which falls within typical and expected differences. Remember that RealBench represents a full system test, so these variations aren’t solely due to the processor.
(5) Variations in voltage settings (Adaptive, Fixed, Offset, Offset + Adaptive or Auto) generally have minimal effect on benchmarks, provided the processor remains stable.
CT
Thank you for your detailed response. I gained valuable insights from your explanations. I now follow the established testing procedures when attempting a new overclocking configuration:
1. I run CPU-Z benchmarking.
2. I execute 3 instances of Cinebench.
3. I perform a 60-minute RealBench session.
4. I use prime95 V26.6 for 60 minutes.
Can I reasonably conclude that if my overclock settings succeed without errors in all these tests, my system is quite stable?
In my initial post, a breakdown of utilities by TDP percentage was provided. It's worth noting that CPU-Z's performance test is relatively light, so it should be omitted, as it's unlikely to reveal any issues and isn't suitable for thermal analysis. Instead, focus on running utilities near 100% TDP.
Once you're content with your O/C configurations, during final stability checks you can include AIDA64's CPU, FPU cache, and memory tests individually, as well as in combination with the most demanding applications and/or games. Additional useful tests from Futuremark, such as 3DMark, 3DMark Vantage, and 3DMark 11, are also recommended—they tend to cause unstable CPUs to crash during physics evaluations. Unigine's tools like Heaven, Valley, and Superposition will similarly trigger crashes in an unstable CPU.
Be aware that if your GPU and memory are also overclocked, make sure to revert them to factory settings afterward. This helps avoid introducing further complications when trying to identify CPU issues. A slight rise in temperature can expose hidden instabilities, so monitoring ambient conditions is important.
Stability can be affected by temperature changes. CPUs that performed well during colder months might show intermittent problems in warmer seasons, even if the room is kept at a consistent 20°C in winter and 25°C in summer. A mere 5°C increase can make previously undetectable issues appear.
For overclocking, it's advised to test CPU, RAM, and GPU separately for stability before combining them, requiring four separate sessions. Achieving stable system overclocking demands effort, but treating it as a process of building a dependable setup is worthwhile. If settings stay consistent and hardware remains functional, you only need to invest time once—then you can relax knowing your build is solid.
I tested my 2600k at 1.38v for five and a half years, pushing it to the brink of stability. It would occasionally drop to 0x124 about once or twice yearly, regardless of how much I played around with it. No matter what I did, I consistently ended up with the same vcore around 4.5 and 4.6ghz, which matched what the chip was designed for. Stability feels like an illusion.
What are the recommended Vcores for the 8700k to ensure stable 24/7 overclocking? One point of view suggests 1.35v as a good starting level for 5.0ghz. For occasional overclocking, the suitable Vcores depend on the specific CPU and require personal adjustment. Somewhat unsafe levels are above 1.4v, especially for beginners. The stress test duration should be around 10 rounds initially using Intel Burn Test to assess stability, followed by a longer Prime95 run and additional Intel Burn Test sessions. The LLC settings can influence stability; experimenting with them may help, but load line calibration is advised to minimize voltage drops. Keep an eye on multiple temperature readings during testing, not just the package temp, as uneven core temperatures might indicate issues with thermal paste or installation.