Coffee Lake OC, AVX Offset and Stress Testing

[Sigep46]

Just completed the build listed in my signature, been at least 6 years since last complete build and OC. So, I have been having to research and relearn. I am needing some guidance on how to proceed with my overclock. Below I have listed current BIOS setting for my OC. My OC procedure was to slowly increase the Core Ratio (wanted 4.9 or 5GHz), reboot, run Cinebench till PC froze/bluescreen, reboot increase Vcore rinse and repeat till was stable with Cinebench. Did this till I got to Core Ratio I was shooting for then ran Prime95 V29.5 Custom with Min/Max FFT Size 1344, Run FFTs in Place, and time to run each FFT set to 15 minutes. If OC made it 30 minutes of custom test, I would then run Prime95 Small FFT, otherwise would adjust VCORE and try again. I could never make it past 90 minutes with settings below running small fft in Prime95. I was seeing temperatures between 68-80C during small FFT. It felt like I would have to keep increasing my VCORE to make it 2+ hours with Prime95 small FFT which I am concerned would be close to 1.4 VCORE or more.

I have read that Prime95 version I have installed uses AVX2 which really loads and heats up the CPU more than necessary. This PC is primarily for gaming with some web browsing, picture editing, music and movies, so limited to no AVX usage. Should I put an AVX Offset in and rerun Prime95 knowing I am really stressing OC at a lower frequency or run pre-AVX version of Prime95? Is 90 minutes of Prime95 small FFT with AVX2 sufficient and move on to running AIDA64 and Realbench for more real world stress testing? How long should I run each to consider OC stable as time seem to vary from all forums I read?

Also, CPU was re-lidded using Thermal Grizzly Conductonaut. I have been using Coretemp and HWInfo for monitoring.

AI Overclock Tuner - MANUAL
BCLK Freq - 100
AVX Instruction Core Ratio Negative Offset - 0
CPU Core Ratio - SYNC ALL CORES
Core Ratio Limit - 49
IA AC Load Line - .01
IA DC Load Line - .01
CPU Core/Cache Voltage - ADAPTIVE MODE
Additional Turbo Mode CPU Voltage - 1.33
Load Line Calculation - 5

 

[J☼E]

I've had my 8700k for a couple of weeks now. I'm at 5ghz 100x50, 1.35v what motherboard are you using?

 

[mackerel]

Kinda going through similar with my i3-8350k at the moment. Given non-AVX workload is the focus, you have a known voltage you're happy with for whatever clock in that condition. What I'd suggest is now finding the highest clock that is prime95 stable at that voltage. Use the AVX offset to find the highest stable. If it is any indication, my 8350k seems stable at 5.0 GHz for non-AVX workloads. At the same voltage required to do that, I have to drop to 4.7 GHz to be Prime95 stable, so I set an AVX offset of 3 in my case. You might also get further gains from increasing cache speed also. My 8350k defaults to 3.7 GHz stock, it seems stable with P95 at 4.4 GHz, and I have run it at 4.6 GHz for non-AVX, but there's no offset ratios here so I have to pick the lower stable setting.

BTW does P95 29.5 even exist? Latest release is 29.4 as far as I'm aware, although I don't think for stress testing there is anything different compared to 29.3, even then only very minor differences compared to 28.10.

How long to run... it is a matter of confidence. Of course you don't have to be at the computer all the time when doing stress tests so I'd aim to give it a couple hours each of prime95 and aida64, and typically an hour of realbench. More never hurt. Just leave it running while you do other things elsewhere.

 

[Sigep46]

Using Asus Maximus X Hero. I was able to run 5GHz at 1.35 vcore pass Cinebench and 30 minute custom Prime95, but failed Prime95 small fft nearly as soon as the test started. Ended up increasing vcore to 1.40 and still blue screened small fft. Decided to move back down to 4.9 GHz.

 

[Sigep46]

Your correct, Prime95 V29.4 b5.

 

[Tír na nÓg]

It should do 5GHz in the 1.35v range max.

Set a -200MHz AVX offset and you should be good to go.

 

[SPL Tech]

I would leave AVX offset at 0. If you set it to 2, you're going to drop the clockrate by 200 Mhz which in effect means you're not really testing the set clock speed of your system. If you're at 49x multi with AVX offset, you're actually testing 4700 Mhz at your current voltage which could result in a stable condition that is not stable on non-AVX programs.

Use Intel Burn-In test for stability testing. It's more relevant and easier. If you can run one loop on the "high" setting, you're going to be stable enough for gaming. Prime 95 is excessive. My computer is not Prime 95 Stable, but it is stable in Intel IBT and I've never once had the computer BOSD or lock up during gaming and I've had it for more than four months.

 

[mackerel]

The whole point of AVX offset is you can be Prime95 stable and everything else stable too. If you are not Prime95 stable, you are unstable. Just because you haven't seen a crash yet doesn't make you stable. At best "stable enough" to do certain things, like we use the term "bench stable". AVX offset wasn't an option until relatively recently, so in the old days you had no choice.

 

[SPL Tech]

Just because you haven't seen a crash yet doesn't make you stable.

I can run Cinebench on loop, Superposition on loop and game all day and night and it wont crash, not even after several months. If I can do that it's as stable as I will ever need it to be. However, I also run AVX offset of 0. If I ran it at 1, I would pass P95 with my current clock speed.

 

[EarthDog]

I would leave AVX offset at 0. If you set it to 2, you're going to drop the clockrate by 200 Mhz which in effect means you're not really testing the set clock speed of your system. If you're at 49x multi with AVX offset, you're actually testing 4700 Mhz at your current voltage which could result in a stable condition that is not stable on non-AVX programs.

Use Intel Burn-In test for stability testing. It's more relevant and easier. If you can run one loop on the "high" setting, you're going to be stable enough for gaming. Prime 95 is excessive. My computer is not Prime 95 Stable, but it is stable in Intel IBT and I've never once had the computer BOSD or lock up during gaming and I've had it for more than four months.

it would only drop for avx instructions. So you are correct when you say you are only testing 4.7, but only for avx instructions. Is p95 avx all the time???

 

[mackerel]

Is p95 avx all the time???

The heavy lifting code in P95 stress test uses AVX where it makes sense to do so. For pretty much all Intel CPUs that support AVX (introduced with Sandy Bridge) that is all the time. If you want to, you can force it off by editing options. It also uses AVX on Ryzen providing you have a recent 29.x verison. 28.x didn't detect Ryzen properly so used older transforms, although due to the intentionally weak FPU in Ryzen AVX doesn't give a big performance boost like you see on Intel. As such it isn't as good a stress test on Ryzen. Less work done, less power, less stress.

Remember that P95 was originally written to find large prime numbers, and performance is optimised where possible. It's use as a stress test spun off from that. Still waiting for AVX-512 support though.

 

[Sigep46]

Well, I tried 5GHz at 1.35 VCORE and downloaded Prime95 V26.6, which does not have AVX. Went to BSOD pretty quickly, probably can get 5GHz if I want to keep increasing voltage, but 100MHz is not that important to me. Dropped ratio down to 49 and Prime95 V26.6 has been running for 12+hrs. I know at 1.33 VCORE and Ratio 49 I can run small FFT Prime95 V29.4 for 90 minutes, betting I can get 2+ hrs at 1.35 VCORE. Going to run Realbench and maybe AIDA64 for few hours just to be sure.

 

[aaawew]

I've had my 8700k for a couple of weeks now. I'm at 5ghz 100x50, 1.35v what motherboard are you using?

What are your stress testing temps? I'm hitting 90 max with 1.3v 4.7Ghz on prime blend with a H115i

 

[gsrcrxsi]

i run AVX -1 on my 5GHz gaming rig (in sig)

would i prefer to not use the offset and have 5GHz all the time? sure. but since theres no option for different voltage levels with AVX offset, and the voltage required and resulting temperatures to support 5.0GHz with AVX was unreasonable.