Cant Get i7-9700k Stable at 5Ghz or Above

Zer0CoolX · Jan 4, 2019

bmgjet said:
Are you going to tell them your returning it because it doesnt overclock high enough?

I may be stupid but I'm not an idiot

Zer0CoolX · Jan 5, 2019

Exchanged CPU, seems much better (making it at least average overall).

Testing 50x CPU, 47x ring, 1.365v, LLC 3, Fixed voltage, enhanced turbo off.

Prime95 as recommended in first 2 responses with avx and fma3 disabled, going on 35 mins. If I get to 2 hours I will consider it stable and then try with adaptive voltage and stress test again. Max temps are 80-82c.

Noticed right away a big difference. Old CPU couldn't hold 4.9Ghz with more voltage (tried up to 1.4v I think) and higher LLC regardless of ring ratio.

In testing old CPU Prime95 and Intel XTU would crash almost instantly when failing. New one ran my last attempt with lower voltage for 30mins before crashing.

Pretty confident 5.0 stable is now possible and if I get it I am not even going to push it harder, I'll be content.

Thanks all for the help, I can finally join the 5Ghz club!

Zer0CoolX · Jan 5, 2019

Actually still need advise.

As per above 1.365v LLC 3 In fixed mode proved to be stable.

I changed from fixed to adaptive and the test failed after about 10 mins. I noticed vcore readings on adaptive never went even close to 1.365v... The max was 1.336v under adaptive for vcore.

So how can I have it scale down voltages when not at full load but have it hit the same voltages as fixed when at full load?

Is this just a matter of using a higher LLC or is it likely something else?

Thanks

jfriend00 · Jan 5, 2019

Zer0CoolX said:
Actually still need advise.

As per above 1.365v LLC 3 In fixed mode proved to be stable.

I changed from fixed to adaptive and the test failed after about 10 mins. I noticed vcore readings on adaptive never went even close to 1.365v... The max was 1.336v under adaptive for vcore.

So how can I have it scale down voltages when not at full load but have it hit the same voltages as fixed when at full load?

Is this just a matter of using a higher LLC or is it likely something else?

Thanks

You need to look at the actual voltages going to the CPU in both your scenarios. VR VOUT (as shown in HWiNFO64) is the best thing to look at. It is likely that when you switched to adaptive, you're not getting quite as much voltage to the CPU, regardless of what you specified in the BIOS. I had a similar issue. The voltage the CPU gets is proportional to what you set in the BIOS, but not necessarily exactly that. Find VR VOUT in HWiNFO64 and see what it's at when under full load in both your circumstances. Changes are it's lower when you're setting adaptive (which is why it isn't stable) and you need to increase the offset until VR VOUT reaches what you had when using Fixed voltage.

Falkentyne · Jan 5, 2019

Zer0CoolX said:
Actually still need advise.

As per above 1.365v LLC 3 In fixed mode proved to be stable.

I changed from fixed to adaptive and the test failed after about 10 mins. I noticed vcore readings on adaptive never went even close to 1.365v... The max was 1.336v under adaptive for vcore.

So how can I have it scale down voltages when not at full load but have it hit the same voltages as fixed when at full load?

Is this just a matter of using a higher LLC or is it likely something else?

Thanks

As Jfriend said, VR VOUT is what you should be looking at, not vcore. Because vdroop functions differently at different loads. Some vcore sensors will keep the same vcore at medium amps load as very high amps load (like the ITE 8792E sensor) when a medium high LLC is set, but if you check VR VOUT, you will see that vdroop is actually about 20-30mv *more* at the high amps load than the medium amps load. And there you go. If you have an Asus Maximus XI board, then their vcore sensor is already recalibrated to match VR VOUT.

(Proof:
https://www.overclock.net/forum/27686004-post2664.html )

And with adaptive voltage, several factors extra come into play with make things hard to configure:
1) the IA AC loadline setting (the default 2.10 mOhms or 1.60 mOhms value causes the "starting" VID to rise above the preset VID, based on some factor I am not sure of.
2) The IA DC loadline setting causing the VID to droop by the mOhms value at load (but the cpu vcore seems to use the IA AC loadline value, and is only very slightly affected by the DC value, but the DC value affects power measurements (power draw) massively.
3) The Loadline calibration (LLC) level, which affects cpu vcore directly. (if LLC is kept at disabled or "standard/default", then the cpu vdroop will follow the default VID, based on the default IA AC and IA DC loadline preset values, because the default Loadline calibration level is either 2.10 mOhms or 1.60 mOhms (depending on what the OEM used for it--increasing loadline calibration reduces the mOhms--e.g. maximum LLC is =no vdroop = 0 mOhm loadline).
4) the default CPU VID, which is based on the CPU cache ratio (not core ratio), which seems to stop scaling upwards at 4.7 ghz ring ratio.

When you use static (manual) voltages, everything above is thrown out the window, except the CPU voltage you set in bios and the loadline calibration setting (which are again in mOhms, but we only see levels or words...the OEM's never tell us how many mOhms is each LLC level, but maximum is usually 0 mOhms (flat) loadline--Ultra Extreme on gigabyte, LLC8 on Asus, LLC1 on Asrock, Mode 1 on MSI, etc).

Zer0CoolX · Jan 6, 2019

Thanks guys. My HWInfo does not have an entry for VR VOUT.

I got stable by using LLC 2 and adaptive + override @ +0.010. Ran prime95 4 hours without issue at 50x cpu, 47x ring, 1.365v CPU, enhanced turbo off. Max temps leveled off to ~78-82 depending on core using Be Quiet Dark Rock Pro 4. I am feeling pretty good about it. I will go through some games to make sure in the coming days but think I have got it now.

I appreciate everyone's input and help.

white owl · Jan 6, 2019

Zer0CoolX said:
Exchanged CPU, seems much better (making it at least average overall).

Testing 50x CPU, 47x ring, 1.365v, LLC 3, Fixed voltage, enhanced turbo off.

Prime95 as recommended in first 2 responses with avx and fma3 disabled, going on 35 mins. If I get to 2 hours I will consider it stable and then try with adaptive voltage and stress test again. Max temps are 80-82c.

Noticed right away a big difference. Old CPU couldn't hold 4.9Ghz with more voltage (tried up to 1.4v I think) and higher LLC regardless of ring ratio.

In testing old CPU Prime95 and Intel XTU would crash almost instantly when failing. New one ran my last attempt with lower voltage for 30mins before crashing.

Pretty confident 5.0 stable is now possible and if I get it I am not even going to push it harder, I'll be content.

Thanks all for the help, I can finally join the 5Ghz club!

Running manual then stressing again with adaptive defeats the purpose of using manual mode to stress with.
You stress in manual and use adaptive daily for power saving, not wise to do both.

Cant Get i7-9700k Stable at 5Ghz or Above

Zer0CoolX

Zer0CoolX

Zer0CoolX

jfriend00

Falkentyne

Zer0CoolX

white owl

Top Contributors this Month

Recommended Communities