8700K Overclock

[WhiteKnuckle 11]

Hi,

I had my CPU professionally de-lidded and running at 5GHz with a massive 1.4 vcore!  I've run Realbench for 30 minutes with temps in mid 80s.  At 4.8GHz/1.28v I have no issues, with temps under 70.  I'm running RAM on XMP 3000 MHz with 1.36v.  I upped this from stock 1.35v to get Realbench running for more than a few minutes.

I don't understand why I have to throw so much vcore at this cpu.  Most BIOS settings I've left on auto and have LLC at 6.

Any suggestions for tweaks to try and bring the vcore down?

ROG Strix 370e/NZXT 240mm cooling/GTX 1060 3Gb

 

[aentwis 61]

Not all CPUs will reach the higher clocks that some can. Was that CPU you have pre-selected and tested for its capability to run at 5.0GHz? You would really have to go with a pre-selected CPU and have that delidded. I went with a i5-8600K that had been pretested to reach 5.1GHz at a Vcore a little lower than yours, I believe 1.39V, it is actually printed on the CPU's new polished heatsink. I run it at 5.0GHz and at a lower Vcore value, as that gives me a little headroom in terms of cooling and so on. The CPU runs really well at the settings I have.

To get the most out of the CPU, try following a specific overclocker's tutorial on how to overclock the CPU you have. If you are lucky, you may even find a tutorial for your motherboard/CPU combination...

But if the CPU really needs 1.4Vcore to reach 5.0GHz, there is not really much you will be able to do with other settings to make it more stable at a lower Vcore...

My suggestion is to find a mainboard-specific tutorial, leaving a lot of settings on Auto at such a high overclock is probably not the best idea...

A

[martin-w 3,235]

12 hours ago, WhiteKnuckle said:

Hi,

I had my CPU professionally de-lidded and running at 5GHz with a massive 1.4 vcore! 

 

 

 

Except that 1.4 isn't "massive" Tom. True, that my 8700K will do 5.2 at 1.4, but that's with HT off. Cant recall what it does at 5 GHz, something like 1.35 volts. 8700K is perfectly capable of handling even higher than 1.4 volts. 

The silicon lottery of course dictates that no two CPU's are the same, some clock higher, some lower, some higher volts, some lower volts.

 

  

Quote

 

I don't understand why I have to throw so much vcore at this cpu.  Most BIOS settings I've left on auto and have LLC at 6.

Any suggestions for tweaks to try and bring the vcore down?

 

 

Again, it's not a huge voltage. First thing to do is install the latest BIOS. Performance updates come thick and fast when new platforms are released, and if you aren't running with the latest you may be missing out on several updates. On three platforms now I have seen improvements in overclocking with newer BIOS. 

 

Quote

I've run Realbench for 30 minutes with temps in mid 80s

 

What kind of cooler, make and model? And what did they delid with? Hopefully liquid metal and not basic TIM. 

Edited January 27, 2019 by martin-w

[WhiteKnuckle 11]

Thanks for your response guys.  I didn't know such a thing existed as a pre-selected CPU, mine is just over the counter.

I made some adjustments and tested RB stable for 1 hour today with a max temp of 81C.  I should mention I'm in Melbourne and it's 35 outside today so the ambient is quite high!  I've had a good look across the forums, one mention of Intel recommended max vcore of 1.42, so thought I'd post here.

AVX Instruction Core Ratio Negative Offset: Auto to 3

BCLK Aware Adaptive Voltage:  Disabled to Auto

CPU VCCIO Voltage:  Auto to 1.15

CPU VRM Thermal Control:  Auto to Disabled (not sure this is relevant)

VCORE 1.4 to 1.405

RAM 1.36 to 1.365

Running latest BIOS

Hyperthreading On

Cooler is NZXT X52

The CPU has liquid metal

I'm thinking now to start lowering my vcore and re-test.  Making progress 😀

 

[martin-w 3,235]

9 hours ago, WhiteKnuckle said:

 

I made some adjustments and tested RB stable for 1 hour today with a max temp of 81C.  I should mention I'm in Melbourne and it's 35 outside today so the ambient is quite high!  I've had a good look across the forums, one mention of Intel recommended max vcore of 1.42, so thought I'd post here.

 

 

 

If you have a high ambient, then 81 is to be expected. Max CPU voltage for the 8700K is actually 1.5 volts. Although of course it would be very ill advised to set it that high without very exotic cooling. 

 

Quote

CPU VRM Thermal Control:  Auto to Disabled (not sure this is relevant)

 

CPU Power Thermal control is generally left on Auto so that the safe operating margins are not breached. If throttling is experienced, the usual advice is to direct a fan at the VRM, rather than disabling this setting. For all normal overclocking you shouldn't need to alter this. You do have a high ambient, so make sure your case is directing enough air at the VRM.

 

Quote

Cooler is NZXT X52

 

Make sure the fan curve is properly configured so that under high loads the fans are at sufficient RPM. 

 

Good luck Tom. 😊

[pgde 571]

If you don't need HT for other applications, I would try turning it off to lower temps..

[WhiteKnuckle 11]

I've read some more on AVX, not sure if setting it to anything other than 0 represents a true overclock but it does allow me to reduce my vcore.  At the moment I'm testing AVX at 1 and my vcore is 1.39.  I'll test around this and see how it goes.  I prefer to keep HT for video processing and reading the replies here I can relax about my vcore if I want to push it up a little, so long as my temps are acceptable.  I'll set CPU VRM back to auto. The de-lid makes a huge difference and FSX performance is awesome, although VAS still gives me issues.  I'm seriously thinking to move to Prepare3D but that's for another post.

[martin-w 3,235]

9 hours ago, WhiteKnuckle said:

I've read some more on AVX, not sure if setting it to anything other than 0 represents a true overclock but it does allow me to reduce my vcore.  At the moment I'm testing AVX at 1 and my vcore is 1.39.  I'll test around this and see how it goes.  

 

I don't believe FSX uses AVX, so you will get your full overclock with FSX. It will be only software that uses AVX that will clock back, when those AVX instruction sets run. P3D does utilise AVX I believe, not sure as to what extent. 2 or 3 is the usual AVX offset most use. Worth remembering that one hundred megahertz or two hundred megahertz makes very little difference to frame rate. 1 to 2 frames per second at 30 frames per second. 

 

Quote

The de-lid makes a huge difference and FSX performance is awesome

 

Yep, 15 degrees when I delidded my daughters 7700K and over 10 degrees for my 8700K, although I suspect I could have improved that temp drop. I suspect I may have applied too much pressure when I reattached the IHS. Many don't reattach the IHS of course, they just let it float. I tend to use a VERY small smear of adhesive just at the corners and reattach. 

[WhiteKnuckle 11]

21 hours ago, martin-w said:

 

I don't believe FSX uses AVX, so you will get your full overclock with FSX. It will be only software that uses AVX that will clock back, when those AVX instruction sets run. P3D does utilise AVX I believe, not sure as to what extent. 2 or 3 is the usual AVX offset most use. Worth remembering that one hundred megahertz or two hundred megahertz makes very little difference to frame rate. 1 to 2 frames per second at 30 frames per second. 

As I suspected but I need to do some more testing to determine the affect of increasing the RAM voltage to 1.365 (XMP is 1.35).

I've pretty much decided it's time to move to P3D and that requires some planning, not to mention the time and cost.  In FSX I'm at 30 fps most of the time with the NGX, although FTX Airports and UTX have some impact.  I have both of those wound back a fair bit.  I have a nightmare with VAS.

 

21 hours ago, martin-w said:

Yep, 15 degrees when I delidded my daughters 7700K and over 10 degrees for my 8700K, although I suspect I could have improved that temp drop. I suspect I may have applied too much pressure when I reattached the IHS. Many don't reattach the IHS of course, they just let it float. I tend to use a VERY small smear of adhesive just at the corners and reattach. 

I guess for me 5GHz is the holy grail, I've been trying for many years but only recently, by accident found someone who can de-lid.  Is the IHS the lid?  He basically split it open with a special tool, removed the thermal paste, used nail polish to protect the small three contacts, applied with care liquid metal to the chip, put the lid back, waited 15 minutes and it went back on the MB.  Really glad I didn't attempt myself after seeing him do it 🙂  Saw temps come down 15-20, so I figure I should be able to get 5GHz easy.  Setting the AVX seems to make a big difference on vcore.  For 4.8GHz, 1.28v is stable, temps in low 70s, so 1.4v for 5GHz seems like a huge jump really.

[martin-w 3,235]

Quote

As I suspected but I need to do some more testing to determine the affect of increasing the RAM voltage to 1.365 (XMP is 1.35).

 

Why have you increased your RAM voltage Tom? Not necessary unless you are trying to overclock your RAM. I would stick to the XMP setting. The RAM's rated speed and voltage. 

 

Quote

Is the IHS the lid?  He basically split it open with a special tool, removed the thermal paste, used nail polish to protect the small three contacts, applied with care liquid metal to the chip, put the lid back, waited 15 minutes and it went back on the MB. 

 

Yes, the silver cap on top is the Integrated Heat Spreader, does just what it says, the die itself is rather small, the spreader just spreads the heat over a larger surface area and protects the die. The special tool isn't that special, there are plenty on the market, I use the Delid Die mate 2. I don't use nail polish, it's not required really. Pop the top, clean off the old glue, apply liguid metal, smear of adhesive on the corners, top back on... done. It's easy enough. 

 

Quote

Setting the AVX seems to make a big difference on vcore.  For 4.8GHz, 1.28v is stable, temps in low 70s, so 1.4v for 5GHz seems like a huge jump really.

 

Are you using the same AVX offset for both 4.8 GHz and 5 GHz? Or do you mean AVX offset for 4.8 GHz and not for 5 GHz? 

If you have AVX offset set, and then run RealBench, RealBench runs AVX as part of it's Handbrake test, so it will be automatically be clocking down in accordance with the offset you set in the BIOS, so it will be a lower frequency. However, if no AVX offset it set for 5 GHz it will run at 5 GHz and thus require higher voltage. 

1.4 for 5 GHz with HT on, is higher than average, but then that's the silicon lottery for you.  I would say keep all overclocking as simple as possible, just a few settings in the BIOS need to be considered, don't get carried away. 🙂 

 

By the way, I just noticed you changed CPU VCCIO Voltage. Personally I wouldn't, I would leave it on auto and only modify is I was pushing for high overclocks and needed tweaking. It's a very fussy voltage to mess with. 

[WhiteKnuckle 11]

On 2/4/2019 at 9:02 PM, martin-w said:

Why have you increased your RAM voltage Tom? Not necessary unless you are trying to overclock your RAM. I would stick to the XMP setting. The RAM's rated speed and voltage. 

I found it provided some stability running Realbench.  I've tested for 1 hour and when I had it at 1.35 I couldn't get 15 mins.  CPU VCCIO could be coming into play here too.

On 2/4/2019 at 9:02 PM, martin-w said:

Yes, the silver cap on top is the Integrated Heat Spreader, does just what it says, the die itself is rather small, the spreader just spreads the heat over a larger surface area and protects the die. The special tool isn't that special, there are plenty on the market, I use the Delid Die mate 2. I don't use nail polish, it's not required really. Pop the top, clean off the old glue, apply liguid metal, smear of adhesive on the corners, top back on... done. It's easy enough. 

The guy explained that he puts nail polish over the contacts just in case there's any movement of the liquid metal before it's fully set.  An abundance of caution you might say.  I have a couple of photos but don't remember how to upload them here.

On 2/4/2019 at 9:02 PM, martin-w said:

Are you using the same AVX offset for both 4.8 GHz and 5 GHz? Or do you mean AVX offset for 4.8 GHz and not for 5 GHz? 

I had no AVX set for 4.8 GHz and had this overclock before the de-lid, very stabled at vcore 1.28

On 2/4/2019 at 9:02 PM, martin-w said:

1.4 for 5 GHz with HT on, is higher than average, but then that's the silicon lottery for you.  I would say keep all overclocking as simple as possible, just a few settings in the BIOS need to be considered, don't get carried away. 🙂

 

This is the real clincher for me and your post puts my mind at rest to go a little higher on the vcore.  With no AVX the temps were creeping up to 88 but it has been very hot here, cooler now.  At AVX = 1 then I don't anticipate any issues and can keep the temps to low 80s which I'm more combatable with.

On 2/4/2019 at 9:02 PM, martin-w said:

By the way, I just noticed you changed CPU VCCIO Voltage. Personally I wouldn't, I would leave it on auto and only modify is I was pushing for high overclocks and needed tweaking. It's a very fussy voltage to mess with. 

So probably what I need to do is set this to auto and run RB for an hour.  All being well I can drop the RAM back to 1.35 and test again.  If all good then I'm pretty much done.  If first test fails I set it back and drop the RAM and re-test.

[martin-w 3,235]

Quote

The guy explained that he puts nail polish over the contacts just in case there's any movement of the liquid metal before it's fully set.  An abundance of caution you might say.  I have a couple of photos but don't remember how to upload them here.

 

Yes that's right, some people do. Iv'e not come across anyone who has had an issue not doing that though. Guess it's okay to be safe rather than sorry.