Overclocking Individual CPU Cores

[Z3R0 CHANC3]

So, I'm definitely an inexperienced overclocked, so I have a few questions.

 

I'm running Prime95 to test the cpu, and I close out everything in the taskbar as well as antivirus.

 

I am noticing that the same core consistently errors out ahead of the others.   So, I incrementally increase voltage and re-test.  I've had to up the voltage 3 times now and it LOOKS like I may have found the sweet spot.  

 

But....as voltage creates heat and lower voltage is better, should I decrease the frequency of that one core to enable lower voltage?  I haven't seen any overclocking videos where they do, but it would seem to make sense, although I could see that cores running at different frequencies could possibly be a bad thing.

 

I5-8600k, Asus Z370i, 17gb 3200 G.Skill Trident Z, 960pro 512gb nvme, EVGA 1080ti, EVGA 240 AIO.

 

I'm using XMP profile but haven't manually adjusted any memory settings.   Right now I'm playing with 4.5ghz, 1.200vcore.  LLC at 6 which is keeping it right at 1.200, ocassionally dipping to 1.184.  I can post up the full list of bios changes if needed.  Core temps are all staying under 60c.

 

Thanks.

[Jurrunio]

No you cant, the frequency of the entire Coffee Lake CPU is the same across the cores. The difference in frequency per core you might see is just an error.

[Bananasplit_00]

Just now, Jurrunio said:

No you cant, the frequency of the entire Coffee Lake CPU is the same across the cores. The difference in frequency per core you might see is just an error.

wait, you cant per core overclock coffee? when did that go away? not that it was an actiually usefull feature but it was kinda neet i guess  

 

1.2V is ok for 4.5GHz i guess

[Z3R0 CHANC3]

Yeah I needed around 1.39 for 5ghz but then I didn't like how hot it was getting (I'd prefer to stay under 80), plus that wasn't with long term Prime95 runs so I don't know if it was actually stable.

 

It would be interesting to de-lid this cpu but I have a feeling with the voltage it needs that it's not a silicon lottery winner.

 

I had noticed the bios had an option for individual cores but I didn't actually try to do it.   Those boxes may all be locked, and I know they are with "sync all cores" selected.

[Jurrunio]

19 minutes ago, Bananasplit_00 said:

wait, you cant per core overclock coffee? when did that go away? not that it was an actiually usefull feature but it was kinda neet i guess  

 

1.2V is ok for 4.5GHz i guess

You cant per core OC coffee lake. All you can do is set different multipliers for when X amount of cores are being utilized, but that doesnt specify the cores running that a specific speed. Mobile phone CPUs, on the other hand, allow different cores to have different clocks. At least my old Samsung Note 2 and Lenovo Phab can.

 

14 minutes ago, Z3R0 CHANC3 said:

Yeah I needed around 1.42 for 5ghz but then I didn't like how hot it was getting (I'd prefer to stay under 80), plus that wasn't with long term Prime95 runs so I don't know if it was actually stable.

 

It would be interesting to de-lid this cpu but I have a feeling with the voltage it needs that it's not a silicon lottery winner.

 

I had noticed the bios had an option for individual cores but I didn't actually try to do it.   Those boxes may all be locked, and I know they are with "sync all cores" selected.

the 'per core' thing doesnt specify the cores that will run at that speed. In other words, you cannot avoid the core that overclocked exceptionally awful with this option.

[Z3R0 CHANC3]

4 minutes ago, Jurrunio said:

 

the 'per core' thing doesnt specify the cores that will run at that speed. In other words, you cannot avoid the core that overclocked exceptionally awful with this option.

Yeah I decided to play with it.  All it would let me do is make each core the same or lower than the one preceding it.  Lol and from what you all are saying that wouldn't "take" anyway.

 

I'm playing with 4.7ghz now.  1.24v wasn't enough.  I bet 4.7 will require 1.26 or 1.27.

[Bananasplit_00]

40 minutes ago, Jurrunio said:

You cant per core OC coffee lake. All you can do is set different multipliers for when X amount of cores are being utilized, but that doesnt specify the cores running that a specific speed. Mobile phone CPUs, on the other hand, allow different cores to have different clocks. At least my old Samsung Note 2 and Lenovo Phab can.

Huh, I'm pretty sure I can put my cores into whatever stabile frequency I want on my I7 4790K, not that id want that but I could. Don't remember if it's a max frequency or just a locked, probably a max but I can set individual core multipliers which I guess coffee can't

[Jurrunio]

36 minutes ago, Bananasplit_00 said:

Huh, I'm pretty sure I can put my cores into whatever stabile frequency I want on my I7 4790K, not that id want that but I could. Don't remember if it's a max frequency or just a locked, probably a max but I can set individual core multipliers which I guess coffee can't

Your definition of 'per core' isnt what I'm talking about. I'm talking about each core getting its own multiplier. For example, Core 0 and 1 has many work, let it run 50x multiplier. Core 7 has nothing to do, let it run 8x. Core 4 overclock like turd, lock its max multiplier to 43x etc.

[Z3R0 CHANC3]

1 hour ago, Jurrunio said:

Your definition of 'per core' isnt what I'm talking about. I'm talking about each core getting its own multiplier. For example, Core 0 and 1 has many work, let it run 50x multiplier. Core 7 has nothing to do, let it run 8x. Core 4 overclock like turd, lock its max multiplier to 43x etc.

Yeah, that's exactly what I was asking.  And the answer is no.  Lol

 

Well I had to go to 1.29V for 4.7ghz.

1.36V seems to be working for 4.8ghz though I may have overshot.  Max temp got up to 89 on core 2 but now they're all back down to mid 60's for some reason.

 

I'm using kryonaut paste, flat application, not the rice ball method.

[Z3R0 CHANC3]

I guess different Prime95 data sets create more heat than others?  I should have just titled this i5-8600k overclocking.

[Z3R0 CHANC3]

Core 1 is consistently getting the hottest...  about 5c average.

[Jurrunio]

12 minutes ago, Z3R0 CHANC3 said:

Core 1 is consistently getting the hottest...  about 5c average.

that's negligible

 

15 minutes ago, Z3R0 CHANC3 said:

I'm using kryonaut paste, flat application, not the rice ball method.

I myself use the X method. Honestly flat (spread) method is the worst for newcomers because the spreading action favours air bubbles to form.

[Bananasplit_00]

6 hours ago, Jurrunio said:

Your definition of 'per core' isnt what I'm talking about. I'm talking about each core getting its own multiplier. For example, Core 0 and 1 has many work, let it run 50x multiplier. Core 7 has nothing to do, let it run 8x. Core 4 overclock like turd, lock its max multiplier to 43x etc.

That's pretty much what I ment, I could set the multiplier of core one and two to say 40x and core four and five to 51x

[Z3R0 CHANC3]

Well I'm happy with this so far.  This was at 4.8Ghz/1.300V and my 1080 ti with the OC auto calculated thing set up.  Asus just released a new bios the other day and I had to set up my OC profile again since I didn't think to save them to usb.  The system seemed to get more stable when I set my CPU Cache to 45 instead of auto.  Also, I switched to Prime95 v26.6 that doesn't use AVX apparently?  No more of that one pesky core error crap, if my voltage isn't set right the PC actually crashes.  I'm not sure if that's a good thing or a bad thing, but at least it's what I'd expect to happen.  

 

I tried to run Heaven and it went a ways past the night time scene and finally crashed.  I don't know if Heaven is a continuous stress test or if it provides a result.