My new Intel 8th Gen Core i7-8700K Processor

[marjamar]

The 'factory' overclock should have the -3 avx offset IIRC. So i think there its an offset, not throttling considering your temps dont hit 100c.

Not sure what your saying. To me, throttling is something designed to happen when the CPU reacts to something specifically that requires less clock. This avx offset is new for me as well, so I may be saying what you are saying, only using different (maybe wrong) wordage. You are right about the -3 avx offset and that is the reason the CPU falls back to 4.7GHz, but it only falls back (throttles back) when put under load -- By all accounts I have seen, any load whether or not is specific to avx programing. I am uncertain how it makes much sense to do this if this is how it works. Why not just overclock it to 4.7 without using the avx offset, if any load forces it to the same clock anyway. I suppose it must work with some software differently, but not that I have seen.

-Rodger

 

[wingman99]

I can do that. Are you wanting to see the factory overclock or my overclock? Also, there are different thermal responses to different "full processor" utilizations. I can't even run Prime95 Torture small FFT's with my 5GHz overclock. It will hit a 100 C in just a few seconds and that's too high to play with for me. I can run that however using the factory 5GHz setting. My overclock holds 5 GHz, the factory throttles to 4.7GHz. Best I could do with my overclock is Prime95 Torture Blend setting which will be higher everything most likely then the factory overclock running small FFT's.

-Rodger

I would like to see both overclocks.

 

[EarthDog]

The offset kicks in when avx instructions are used. Throttling is a protective function to me. I suppose they are interchangeable, but id stick with more common terms for what they are...its called an avx offset in the bios, and thermal/vrm/temperature 'throttling' are common/correct.

I would like to see factory overclock and utilization so the CPU won't go above 95c.

It would be helpful to explain WHY you are asking for these things so those you are helping can see the plan/process/thinking.

 

[wingman99]

I don't know what he is doing so if he shows the screenshots I can help to tell.

 

[Johan45]

Not sure what your saying. To me, throttling is something designed to happen when the CPU reacts to something specifically that requires less clock. This avx offset is new for me as well, so I may be saying what you are saying, only using different (maybe wrong) wordage. You are right about the -3 avx offset and that is the reason the CPU falls back to 4.7GHz, but it only falls back (throttles back) when put under load -- By all accounts I have seen, any load whether or not is specific to avx programing. I am uncertain how it makes much sense to do this if this is how it works. Why not just overclock it to 4.7 without using the avx offset, if any load forces it to the same clock anyway. I suppose it must work with some software differently, but not that I have seen.

-Rodger

The offset kicks in ONLY when the system encounters AVX instructions. Maybe the SW you use all has some sort of AVX instruction but not everything does. Take gaming for instance where high CPU speed can give a few more FPS, Cinebench R15 doesn't use those intructions either. So it's not "heavy" load that activayes the offset it's AVX,AVX2 that do if the CPU detects them.

 

[marjamar]

The offset kicks in ONLY when the system encounters AVX instructions. Maybe the SW you use all has some sort of AVX instruction but not everything does. Take gaming for instance where high CPU speed can give a few more FPS, Cinebench R15 doesn't use those intructions either. So it's not "heavy" load that activayes the offset it's AVX,AVX2 that do if the CPU detects them.

This is what I thought also. Thing is, it's still clocking down for some reason, even on Cinebench. In the case of Cinebench, it's individual cores, sometimes 2 or 3 at a time drop to 4.7GHz from the factory extreme 5GHz overclock. I suspect there must be some reason for this, other then it just happens sometimes.

Anyway, seems I actually cannot run prime95 blend for more then 5 1/2 minutes or so without dropping a worker, or crashing with any of the 5GHz overclocks I have been trying. The factory extreme 5GHz will run prime95 blend fine, but of course that's with the -3 avx offset kicking in to 4.7Ghz.

-Rodger

 

[Johan45]

I could be wrong about CB15, you can try WPrime benchmark instead. But it seems that you're still not quite stable.
My approach would be to set a voltage of 1.35V and see how far that gets you heat/stability wise with P95. 4.6 , 4.7 or wherever it stops. Then raise multi and test NON AVX workloads for stability for max clocks.

 

[EarthDog]

Marj... if you want to figure out why its throttling, leave hwinfo open or xtu and run whatever application it is and see...weve mentioned this before... (this process is droning on and on, lol...).

My 8700k i have must be awesome... 1.2v 4.8ghz no avx offset...

 

[marjamar]

Marj... if you want to figure out why its throttling, leave hwinfo open or xtu and run whatever application it is and see...weve mentioned this before... (this process is droning on and on, lol...).

My 8700k i have must be awesome... 1.2v 4.8ghz no avx offset...

That's part of what I do when testing an overclock to monitor what's happening. That's how I know Cinebench was being affected by this avx offset. I'll post back here when I have more figured out and what I have done to get my system up and working how I'd like it to.

-Rodger

 

[wingman99]

Marj... if you want to figure out why its throttling, leave hwinfo open or xtu and run whatever application it is and see...weve mentioned this before... (this process is droning on and on, lol...).

My 8700k i have must be awesome... 1.2v 4.8ghz no avx offset...

My 8600k is even better 1.160v 4.8Ghz no avx offset. Temp is high with hyper 212.

 

[marjamar]

Well, both of you can just send me the plans to follow then, HA!

 

[EarthDog]

I think you are doing a good job of following your own.

 

[marjamar]

Still can't get 5 GHz stable with no avx offset. Shifting gears, I can get 4.8 stable with no avx offset @ 1.36. Lots of untried settings and I'll need to research many of them. I did get the new Kryonaut TIM and took the CPU waterblock off, drained and blew out the water from the loop, took the waterblock apart and cleaned it as there was some minor debris in the loop and the block. I did all of this before doing these upgrades but must have been some bits and pieces left in the rad I suppose. Anyway, seeing an improvement with temps a few C lower and actually getting about .6 to .7 C difference now on inlet/outlet temps too -- Most likely because the loop and water block are more clean. I don't really think I'll see much better without delidding, so I just have to give more thought to what my goal is here.

-Rodger

 

[EarthDog]

Werent you seeing .7C prior as well?

 

[marjamar]

Werent you seeing .7C prior as well?

I thought I was, but it must have only happened once and that maybe was in my mind. Now it's always at .6C and bumps up for a bit to .7C. Does this up and down ever minute or less, and it never did that earlier. There was a enough debris in the waterblock that a few of those milled channels may have been blocked or restrictive enough to account for this.

I suppose I will need to delid this CPU. For some reason or another, it bothers me that I can't run a solid 5GHz if I want to. I guess I'm just a obsessive-compulsive and have to make things work the way I think they're suppose to. Good for a business person (makes money), not so good for a family person (makes wife upset at times), HA.

-Rodger

 

[EarthDog]

Not sure what you are accounting for, lol. Those were around the same values you mentioned earlier (twice) in the thread. Both times it was in your head ? Like I said, also twice before I think, temperatures inside a properly radded and flowing loop differ VERY little.


LMK if you want to borrow a delid tool instead of buying one. I do not have the relid kit. Would just cost you shipping. PM me.

 

[marjamar]

Not sure what you are accounting for, lol. Those were around the same values you mentioned earlier (twice) in the thread. Both times it was in your head ? Like I said, also twice before I think, temperatures inside a properly radded and flowing loop differ VERY little.


LMK if you want to borrow a delid tool instead of buying one. I do not have the relid kit. Would just cost you shipping. PM me.

Thanks EarthDog for the loan, but I have already ordered all I needed for this delidding adventure! Been watching a few videos, and I can see how much better thermal transfers will be if this is done right, with the right TIM and liquid metals.

-Rodger

 

[EarthDog]

Temps will drop on the CPU for sure. Make sure you do some baseline testing first, then after so we can see the difference.

 

[marjamar]

Temps will drop on the CPU for sure. Make sure you do some baseline testing first, then after so we can see the difference.

Will do.

-Rodger

 

[wingman99]

When your doing the delidding temperature comparison try to note the ambient temperature before and after delidding.