[SOLVED] i7-8700K Overclocking Stability Testing Help

ak2000spooky

Distinguished
Mar 14, 2011
41
0
18,530
0
Hoping to get my hands on a RTX 3080 at some point... decided to buy a new 1TB SSD, 16GB more RAM, a new sound card, format my SSD, install a fresh copy of Windows, update my MB bios, and OC my CPU/GPU. However, I'm running into an issue when stability testing my CPU.

I went through the following order:
  1. Set memory to XMP profile (3200)
  2. Test memory with Memtest86 (2 full passes)
  3. Set CPU core ratio (all core)
  4. Set CPU core voltage (fixed)
  5. Set CPU LLC (Level 1)
  6. Run Prime95 Small FFT (AVX & AVX2 off) for 15min
  7. Run Realbench stress test (up to 16GB memory) for 8 hours
  8. Run Prime95 Blend (AVX & AVX2 off) for 8 hours
I passed the memory test with the XMP profile. Eventually I ended up at a 48 multiplier, a core fixed voltage of 1.28v, and a LLC Level 1... I was able to "pass" the Prime95 Small FFT test (~77-80 degrees Celsius temps), then I "passed" the Realbench test, but during the Prime 95 Blend test one of the workers (#4) stopped due to an error (rounding was 0.5, expected less than 0.4) after about 3 hours. I bumped up the CPU core voltage to 1.29v and ran the Prime95 Blend test again and had the same issue around the 2 hour mark. I'm running the Memtest86 again for the full 4 passes... if there are no errors, does it mean I need to bump up my CPU core voltage in order to pass the Prime95 Bend test?

After I get a stable CPU OC, I plan to work on the GTX 1080 OC. I forgot to mention that I only use this PC for gaming.

CPU: i7-8700K
MB: ASRock z370 Extreme4
RAM: 4x8GB G.Skill TridentZ DDR4 (3200)
GPU: Gigabyte GTX 1080 G1
M.2 SSD: 1TB Samsung 970 EVO Plus & 500GB Samsung 960 Pro
PS: 850w EVGA SuperNOVA G2 (80+ GOLD)
SC: Soundblaster AE-5 Plus
Case: Nanoxia Deepsilence 5
Cooling: Custom EK Loop
 

Darkbreeze

Retired Mod
If you read the entire guide you'll see that I recommend you should configure your CPU overclock BEFORE you set the memory to it's XMP profile configuration or manually overclock the memory. That way you can be sure that any problems that arise are from the CPU overclock and not from the memory configuration. Your XMP profile or manual overclock of the memory should happen AFTER you have established that your CPU overclock is stable.

Also, this could be helpful to some degree as well.


I wouldn't bother running a "long" Small FFT test. Just run it for 15 minutes with all forms of AVX disabled. If it's within thermal spec, then that part is done. Then you move on to the stability tests and the stability tests don't generally involve Small FFT except perhaps as a supplemental addition at the tail end of things, if you really want to. I'd just run Realbench as outlined. You don't even need to run the Blend test until after you've established the CPU overclock is stable and have enabled XMP or manually overclocked the memory. If you can pass Memtest86 for 4 passes, 8 hours of Realbench AND the CUSTOM Prime95 configuration, then you are about as stable as you can ever expect to be at the end of it all.

If you have not done so, reading the Intel temperature guide is a good place to gain some additional understanding of things too.

 
Last edited:

Darkbreeze

Retired Mod
Testing with Memtest is pointless if you don't run four full passes. I've seen MANY configurations not fail until some point in the fourth pass. As a quick test during actively configuring memory configurations, fine, but for determining actual stability, four passes, or more. (Paid version)

Obviously, everybody has some different idea or a different take on the same idea when it comes to various types of testing, and you should not make the (Somewhat common) mistake of confusing thermal testing and stability testing because they are entirely different things, but here is my take on testing which is derived from a rather lengthy years long refinement process that includes feedback from some of the most respected members here and elsewhere, as well as input from various engineers and reviewers.

You will need to scroll down towards the end of the guide for the sections on testing.

 
Reactions: CompuTronix

ak2000spooky

Distinguished
Mar 14, 2011
41
0
18,530
0
Testing with Memtest is pointless if you don't run four full passes. I've seen MANY configurations not fail until some point in the fourth pass. As a quick test during actively configuring memory configurations, fine, but for determining actual stability, four passes, or more. (Paid version)

Obviously, everybody has some different idea or a different take on the same idea when it comes to various types of testing, and you should not make the (Somewhat common) mistake of confusing thermal testing and stability testing because they are entirely different things, but here is my take on testing which is derived from a rather lengthy years long refinement process that includes feedback from some of the most respected members here and elsewhere, as well as input from various engineers and reviewers.

You will need to scroll down towards the end of the guide for the sections on testing.

Yes, I used that guide as a base. I’m running all 4 passes now. I assume that if no errors pop-up that the issue would be with my CPU OC and the next step would be to increase the voltage, right? The blend test is the only one giving me trouble so far... note, I haven’t ran a long prime95 small fft test yet.
 

Darkbreeze

Retired Mod
If you read the entire guide you'll see that I recommend you should configure your CPU overclock BEFORE you set the memory to it's XMP profile configuration or manually overclock the memory. That way you can be sure that any problems that arise are from the CPU overclock and not from the memory configuration. Your XMP profile or manual overclock of the memory should happen AFTER you have established that your CPU overclock is stable.

Also, this could be helpful to some degree as well.


I wouldn't bother running a "long" Small FFT test. Just run it for 15 minutes with all forms of AVX disabled. If it's within thermal spec, then that part is done. Then you move on to the stability tests and the stability tests don't generally involve Small FFT except perhaps as a supplemental addition at the tail end of things, if you really want to. I'd just run Realbench as outlined. You don't even need to run the Blend test until after you've established the CPU overclock is stable and have enabled XMP or manually overclocked the memory. If you can pass Memtest86 for 4 passes, 8 hours of Realbench AND the CUSTOM Prime95 configuration, then you are about as stable as you can ever expect to be at the end of it all.

If you have not done so, reading the Intel temperature guide is a good place to gain some additional understanding of things too.

 
Last edited:

ASK THE COMMUNITY

TRENDING THREADS