Question about overclocking stability
Question about overclocking stability
I'm currently working through a lengthy stability evaluation for my delidded and air-cooled (dark rock 3 pro) 8700k to determine the Vcore values for 5.0Ghz, 4.9Ghz, and 4.8Ghz. My aim is to create various overclocking profiles tailored to different scenarios, while planning the 4.9 or 4.8 settings to enable continuous overclocking during 24/7 sessions. I primarily engage in gaming and game development (using Unreal Engine, Visual Studio, and Blender). Occasionally, I run rendering tasks and build complex lighting setups that put a heavy load on the CPU. My objective is to find an overclock that avoids crashes and minimizes micro errors that could eventually damage my operating system or Unreal Projects.
In my stability testing routine, I follow these steps (thermals were validated with a 15-minute Prime95 v26.6 small fft test):
First, run a 1-hour real bench to identify the lowest stable Vcore before encountering errors.
Next, if the first was stable, proceed with an 8-hour real bench.
If the second remained stable, conduct a 2-hour Aida64 FPU stress test and another 2 hours covering CPU, cache, and memory.
Then, perform several Cinebench R15 and R20 runs.
Include a single Blender benchmark.
Run a 8-hour Prime95 blend v26.6 test.
I've observed something noteworthy. For instance, stress tests 1–5 passed smoothly at 5.0Ghz with a Vcore of 1.31 and LLC4. However, I experienced blue screens in test 6 during the Prime95 blend. To overcome this, I had to push up to 1.35V for an 8-hour Prime95 blend session without any issues.
Currently, I'm stress-testing at 4.9Ghz starting from 1.27V. After several tests, everything ran smoothly—no problems or errors detected. However, my 8-hour Prime95 blend test again resulted in a blue screen.
Should I adjust my testing approach and begin with the Prime95 blend from now on, as it appears to be the main constraint? Also, how much more demanding is this CPU stress compared to other tests? How concerned should I be about stability and micro errors?
By the way...in all the tests described, temperatures stayed in the 70s. I only begin to reach 80°C at 5.0Ghz during very intense benchmark runs. Thanks!
Hello, have you come across this stability guide? I rely on prime95 until there are no errors. I've tested it for 24 hours without any issues, then ran it again and encountered an error. To prevent this, I set both sizes to 8K, used FFTs in place, and kept the test at 15 minutes. This usually identifies problems within an hour if they're related to CPU voltage. When overclocking RAM, I try 800K, around 75% of the total memory, for 15 minutes without running FFTs, so we can better detect issues early. You might also want to boost the Loadline Calibration Control in your BIOS, adjust DRAM or PCH voltages slightly, or lower the RAM speed if the 800K test fails.
I currently have load line calibration on LLC 4 on my MSI board. It appears it's typically 0.005v to 0.01v higher under full load compared to the set Vcore. It seems stable with no spikes. If I adjust the lower or higher LLC settings, the vcore changes significantly.