9700k overclocking - what is with the temps in prime95 blend?

[Ace McPlane]

So, my overclock is stable and my temps are good in every other test, but when I run the newest version of prime95 with the blend test, I'm seeing temperatures as high as 97 degrees!

About 10 degrees higher than everything else.

 

I followed Der8auer's 9th gen overclocking guide when I started my overclocking where he said to use an older version of prime95 and to use a custom test with a min of 12 and a max of 12. It does fine in that test. I wanted to test my memory overclock with blend though and it only took like half an hour to see these crazy temps. If I do the same test in the version Der8auer suggests, it always fails at a certain point even when I remove the memory overclock and I can see that it can't even detect what CPU I have.

 

So, this has to come down to the reason why Der8auer suggested doing this specific test instead of anything else which I'm completely lost on other than avoiding having the CPU scale down by the AVX offset in the small FFT's test, but that should mean lower temps not higher ones.

 

It doesn't get this hot in anything else. Not the test Der8auer suggested to do, not Aida64, etc. Just prime95 blend. 

Should I just not being using prime 95 blend or should I be rolling back my overclock? I'm only at 1.3v on the vcore for god sakes and I have a 360mm aio so it shouldn't be that hot.

I mean, it's acting fine everywhere else. I feel like prime95 is just acting crazy with this CPU and is completely inconsistent with the rest of my tests. 

 

If you know anything about what's different with 9th gen overclocking, please let me know what I'm missing here.

 

9700k, asrock z390 taichi, 16gb corsair vengeance RGB pro 3200mhz, corsair ax760, fractal design s36, 500gb samsung 970 evo, etc. 

 

This is the guide I'm talking about: 

 

[Cereal5]

IIRC, P95 is known to run hotter than pretty much all other stress tests. Probably just the instruction set it uses and the processes it tasks the CPU with. If you're at good temps in AIDA64 or Realbench then you should be good. And what really matters is actual real world temps you see. Even some CPU intensive tasks like video editing don't always use 100% of a CPU, and gaming definitely won't, so you likely won't see those kinds of temps anyways.

[Ace McPlane]

3 minutes ago, Cereal5 said:

IIRC, P95 is known to run hotter than pretty much all other stress tests. Probably just the instruction set it uses and the processes it tasks the CPU with. If you're at good temps in AIDA64 or Realbench then you should be good. And what really matters is actual real world temps you see. Even some CPU intensive tasks like video editing don't always use 100% of a CPU, and gaming definitely won't, so you likely won't see those kinds of temps anyways.

Do you think I should be concerned about the memory overclocking if the blend test failed in the older version of prime95 though?  The ram passed memtest86 and the aida64 ram test. It's worth nothing that the prime95 blend test failed with the new version even after removing the memory overclocking and putting it on the stock XMP profile.

[Cereal5]

Just now, stateofpsychosis said:

Do you think I should be concerned about the memory overclocking if the blend test failed in the older version of prime95 though?  The ram passed memtest86 and the aida64 ram test. It's worth nothing that the prime95 blend test failed with the new version even after removing the memory overclocking and putting it on the stock XMP profile.

I would interpret that as CPU instability, not RAM instability. If you're worried about it, you can just bump up the RAM voltage a touch. I wouldn't bother though if it passed memtest and AIDA64. Like I said, P95 is known to be excessively heavy on temps. There was even a version that killed some CPUs lol (reason why I don't use it).

[Ace McPlane]

21 minutes ago, Cereal5 said:

I would interpret that as CPU instability, not RAM instability. If you're worried about it, you can just bump up the RAM voltage a touch. I wouldn't bother though if it passed memtest and AIDA64. Like I said, P95 is known to be excessively heavy on temps. There was even a version that killed some CPUs lol (reason why I don't use it).

Cool, yea I don't think the CPU is unstable. I tend to notice little things like unexpected shutdowns in the reliability history when you have even a tiny bit of instability and I know certain programs I run completely bluescreen if you're not 100% stable so I really think it's just that blend test with that specific version failing for another reason related to the software. 

 

I'm going to leave this question open either way though as there is something about overclocking 9th gen that throws off stress tests that I would like to know more about so maybe someone will know the brass tacks on that. 

Thanks though!

[Ace McPlane]

Update: I lowered my cache ratio multiplier from 46 to 44, am running the blend test in the older version of Prime95 Der8auer suggested and I think it's good now. It doesn't seem to be crashing anymore. I'll have to wait longer to be sure, but I think it's good now. I can probably lower my vcore now too and get those temps down some more.

[Ace McPlane]

Another update: That last stress test failed after all.

Since I was at my temperature limit, I went back to the drawing board and started over with lower LLC. I used level 3 which is right in the middle, I'm at 1.34 vcore now with lower temps, the cache ratio is back up to 4.6, I reinstated my memory overclock and it passed 16 hours of Prime95 blend (the 26.6 version).

So, looks like I'm finally stable.

 

It seems high amounts of LLC aren't worth it here since they increase the temps too much with not as much return in terms of stability as just using a higher voltage at a more mid range LLC level.