Question Need help to overclock i7-9700k

[katulen]

Hey guys.

I am currently attempting to overclock my i7-9700k.

Here is my current specs:
CPU: Intel Core i7 9700K / 3.6 GHz
CPU Cooler: NZXT Kraken X72
MOBO: Asus Rog Strix Z390-F GAMING
GPU: Asus Rog Strix RTX2070 O8G GAMING 8GB GDDR6
RAM: Corsair Vengeance RGB 16GB: 2x8GB DDR4 3000MHz DIMM
SSD: Samsung 970 EVO Plus MZ-V7S500BW 500GB - PCI Express 3.0 x4 (NVMe) - M.2 Card
HDD: Toshiba P300 3.5´´ 2TB - 7200rpm 64MB SATA-600
PSU: Corsair RM750x
CASE: NZXT H series H700i Window Black

Attempt 4.9GHz:
Multiplier: 49
AVX Offset: -3
XMP: XMP 1 Enabled
MCE: Disabled
SVID Support: Disabled
BLCK Aware Adaptive Voltage: Disabled
CPU Core/Cache Voltage: 1.350

• Cinebench r15 - Passed.
• Realbench - Passed.
• Prime95 - Passed.
• Highest temps was 85c.

Attempt 5.0GHz:
Multiplier: 50
AVX Offset: -3
XMP: XMP 1 Enabled
MCE: Disabled
SVID Support: Disabled
BLCK Aware Adaptive Voltage: Disabled
CPU Core/Cache Voltage: 1.360v
CPU Load-line Calibration: Level 6
CPU Current Capability: 170%
Long Duration Package Power Limit: 4095 (max)
Short Duration Package Power Limit: 4095 (max)
CPU Core/Cache Current Limit: 255.75 (max)

• Cinebench r15 - Passed.
• Realbench - Passed.
• Prime95 - BSOD after 2 minutes.

Failed.

2nd attempt 5.0GHz:
Multiplier: 50
AVX Offset: -3
XMP: XMP 1 Enabled
MCE: Disabled
SVID Support: Disabled
BLCK Aware Adaptive Voltage: Disabled
CPU Core/Cache Voltage: 1.360v
CPU Load-line Calibration: Level 5 (Changed)
CPU Current Capability: 170%
Long Duration Package Power Limit: 4095 (max)
Short Duration Package Power Limit: 4095 (max)
CPU Core/Cache Current Limit: 255.75 (max)
Changes: Only changed LLC from Level 6 to Level 5.

• Cinebench r15 - Passed.
• Realbench - Passed.
• Prime95 - Worker stopped after 3 minutes, no BSOD.
• XTU - Passed.

If you need some Min/Max/Avg screenshots of temps/voltages let me know, apparently cant paste them in this thread.

So a few questions here.
I've heard/read that if a worker stops in Prime95, it means your overclocking is unstable. Is it worth trying to let it finish and see what happens or pointless ?

Also i noticed when using XTU, that every 10 sec, 1 or more cores drop to 4.7GHz for a split second and go back to 5.0GHz. Is that due to AVX instructions coming into play or is it my cpu throttling? And how to fix it if needed.

Temps are fine, hitting max 82c on 1 core, rest high 70's.
VCORE is at 1.360v and droops to 1.332v and never go over 1.360v.

Update on my latest 5.0GHz attempt:

I just tryid following:
1.375v vcore fixed, LLC 4. Vdroop to 1.305v and BSOD in Prime 95.
1.375v vcore fixed, LLC 5. Vdroop to 1.350v and BSOD in Prime 95.
1.375v vcore fixed, LLC 7. Vboost to 1.421v and BSOD in Cinebench r15. (No vdroop - Probably overshooted the voltages with Level 7 i assume)

I tryid LLC 6 together with 1.360v without success.

Im trying with LLC 6 now with a higher vcore voltage.
Settings:
Multiplier: 50
AVX Offset: -3
XMP: XMP 1 Enabled
MCE: Disabled
SVID Support: Disabled
BLCK Aware Adaptive Voltage: Disabled
CPU Core/Cache Voltage: 1.380v
CPU Load-line Calibration: Level 6
I removed all the other stuff i added earlier. (See above).

Did OCCT Linpack Non-AVX.
1.380v, vboost to 1.403v. No vdroop.
Max temp 94c.
Distance to TJMax 6c.
After 13min of OCCT i got BSOD.

I dont know what to try next tbh.
Is there ANYTHING I am missing or something i need to dial/tweak for 5.0ghz to happen? Or have i done everything correctly but my chip just won't hit 5.0GHz due to silicon lottery ?

 

[rigg42]

Can you change the switch freq on the VRM? That can help sometimes. Max that out. Disable the following power management settings: Intel Speed Shift Technology, CPU Enhanced Halt (C1E), C3 State Support, C6/C7 State Support, C8 State Support and C10 State Support. Leave LLC at 5. Set Uncore (cache ratio) at -3 to the clock multiplier. Disable XMP until you get stable. You're memory could be unstable. Eliminate it from the equation. Scrap the AVX offset unless you can't pass real bench.

Use HWinfo64 for monitoring. Check sensors only at startup. Its way more accurate than hwmonitor and will tell you if you are thermal throttling. Cinebench is a neat benchmark but is mostly useless for stability testing. Add intel burn test into your testing suite. Run this before you try prime or real bench. It's fast and lets you know in short order if you have a chance of being stable. Don't test with prime 95 with avx enabled. This is a completely unrealistic cpu load. I just use version 26.6 and let it run for an hour. Run real bench for 8 hours with half of your memory after you are stable in an hour of prime 95.

 

[katulen]

Can you change the switch freq on the VRM? That can help sometimes. Max that out. Disable the following power management settings: Intel Speed Shift Technology, CPU Enhanced Halt (C1E), C3 State Support, C6/C7 State Support, C8 State Support and C10 State Support. Leave LLC at 5. Set UNCORE at -3 to the clock multiplier. Disable XMP until you get stable. You're memory could be unstable. Eliminate it from the equation. Manually set system agent and vccio voltages at or slightly above stock. Asus bios can run these way too high on auto. Don't test with any version of prime 95 with avx. I just use 26.6. Scrap the AVX offset unless you can't pass real bench. Add intel burn test into your testing suite. Use HWinfo64 for monitoring. Its way more accurate than hwmonitor.

Hmm what would the option be called to switch freq on VRM - Any ideas?
I don't know about C1E, C6, C7, C8 or C10 State Support ?. I have seen C3 state support.
I do use HWiNFO tho.

Also i will look furthermore into your suggestions later today, and try my 5.0GHz profile again, to see if i can get it stable. And i think you're right about XMP, at least for testing for 5.0GHz. It seems stable tho, since XMP works with stock & my 4.9GHz profile. But ill try disable it on my 5.0GHz profile.

Thanks

 

[rigg42]

Hmm what would the option be called to switch freq on VRM - Any ideas?
I don't know about C1E, C6, C7, C8 or C10 State Support ?. I have seen C3 state support.
I do use HWiNFO tho.

Also i will look furthermore into your suggestions later today, and try my 5.0GHz profile again, to see if i can get it stable. And i think you're right about XMP, at least for testing for 5.0GHz. It seems stable tho, since XMP works with stock & my 4.9GHz profile. But ill try disable it on my 5.0GHz profile.

Thanks

I'm working with a maximus xi hero so you might not have all of the same options. I'm out of town until Sunday so I can't open my bios to check. VRM switching frequency should be in the same sub menu as LLC and power limits if I remember correctly. All of the c-state options should be in there somewhere. I'm not really sure though I've been working with intel mobos from all 4 of the big AIBs lately and I can't remember where to find all the options without the bios menus in front of me. Disabling MCE should essentially achieve the same result anyway.

Re-read my previous post as I made some edits. Hope this helps and good luck.

 

[katulen]

Use HWinfo64 for monitoring. Check sensors only at startup

What do you mean by , check sensors only at startup? :O I mean, shouldn't i monitor my temps while stress testing ?

If i disable the power management options you've listed above, i am not able to run with Adaptive Voltages am i? Because that is my goal. I don't see the point, longterm, to run with static voltages to be honest.

UPDATE on my 5.0GHz attempt:

Settings:
Multiplier: 50
AVX Offset: Auto
XMP: Auto (No XMP)
MCE: Auto
SVID Support: Auto
Turbo Mode CPU Core/Cache Voltage: 1.300v (Adaptive Voltages)
CPU Load-Line Calibration: Level 6

Cinebench r15 - Passed
Cinebench r20 - Passed
Prime 95 v26.6 - Passed
Realbench - Passed

So this is quite interesting. Because i havent been able to pass Prime95 with 5.0GHz before, with manual voltages, but i do with adaptive voltages. However, when i tested 5.0GHz before, i did have XMP ENABLED, and now i dont. So that could be a thing i assume.

Here is some info & screenshots.

Prime 95 v26.6:
It ran flawlessly for 30 minutes, and hit a max temp of 86c. Clocks were stable and so were my voltages under load. Max VCORE was 1.403v, however it was 99% of the time on 1.385v, which you can see in the following VCORE graph. Also my VID was stable at 1.385v.

Realbench:
At first, it pretty much "crashed" or failed, because Furemark(or the GPU part of the test). So i assumed this was due to my GPU Overclock. I then put my GPU to stock, ran the test again, and it ran flawless for 30 minutes. However my temps did go higher, up to 97c on one core, but this was only for 1 second, and it dropped to mid 80c again. This happened a few times throughout the test.

So what do you guys think, how come these settings seem somewhat stable so far ? Because of adaptive voltages or the XMP I profile failing ?

 

[rigg42]

If you were running a newer version of P95 previously that is most likely the difference. I'd start from scratch with the voltage. I'd set a fixed voltage. Adaptive voltages will send more voltage than you actually need to be stable. If you are not getting a slight voltage drop at LLC 6 drop it down to 5. Use intel burn test initially before running prime and real bench. You can probably enable XMP again. I've had real bench crash after a couple of hours of being seemingly stable so you'll need to do an extended test at some point.

 

[rigg42]

What do you mean by , check sensors only at startup? :O I mean, shouldn't i monitor my temps while stress testing ?

This just opens the monitoring window and does away with all the extraneous windows. All relevant info is still there. Yes you should definitely monitor temps. The average is more important than peak if doing extended testing. Temporary spikes are fine. I'd aim for Mid 80's or less average after an extended test.

 

[katulen]

Ive been running P95 v26.6 at all times.

I am actually doing a 1hr P95 as we speak.
With 5.0ghz, 1.350 fixed voltages, LLC 6, svid disabled & xmp disabled.
It ran p95 for 30 mins no problem. Highest temp was 83 and that was a peak temp. I didnt get any vdroops But getting vboost to 1.368 stable.

About Intel Burn Test .. if i run it on Maximum test, it “fries” my system with peaks on 99c .. soo dont know about ITB?

About enabling XMP. I Think thats where the failure is, since i didnt found unstability yet on current settings. Now im trying this settings and see if its stable for 2 hrs. Then ill try enable xmp with same settings to see if i crash.

 

[rigg42]

Ive been running P95 v26.6 at all times.

I am actually doing a 1hr P95 as we speak.
With 5.0ghz, 1.350 fixed voltages, LLC 6, svid disabled & xmp disabled.
It ran p95 for 30 mins no problem. Highest temp was 83 and that was a peak temp. I didnt get any vdroops But getting vboost to 1.368 stable.

About Intel Burn Test .. if i run it on Maximum test, it “fries” my system with peaks on 99c .. soo dont know about ITB?

About enabling XMP. I Think thats where the failure is, since i didnt found unstability yet on current settings. Now im trying this settings and see if its stable for 2 hrs. Then ill try enable xmp with same settings to see if i crash.

V boost is bad. This can damage your CPU when transitioning from load to idle by sending voltage spikes to the CPU. You are better off with a slight droop and increasing vcore to compensate. Standard is all that is required in IBT. I use it as a quick "proof of concept" if you will before doing extended testing.

 

[katulen]

I see. So perhaps change from LLC 6 to LLC 5 to see if that gives me vdroops?

Ahh okay. Ill try ibt standard when p95 is done.

 

[hardyhack]

Where do I see the vboost?

Kind regards from one that has OCd a 9700k to 5Ghz.

 

[katulen]

Where do I see the vboost?

If you use HWiNFO, you can see it, otherwise more simply use CPU-Z and see whenever your voltage spikes over than what you set it to.
I like to use HWiNFO and right+click VCORE and Display Graph. Then i can see the most used voltage and where it vboost/vdroops too.

Standard is all that is required in IBT. I use it as a quick "proof of concept" if you will before doing extended testing.

So i tryid running IBT Standard. However it "fails" after 2-3 minutes, kicking my CPU temp up to around 85c, and comes up with a warning saying some stuff about my cooler or clock speed.
However this settings did pass 2 hours of Prime95.

UPDATE:

Settings:
5.0GHz
CPU Voltage: 1.355
AVX: Auto
MCE: Auto
LLC: Level 5
VDroops: 1.332v-1.355v

Passed 30 minutes of P95 with max temp of 78c, but as soon as i try IBT, it fails after 10-20 seconds with "warning". My temps spikes really high & really fast, hitting 85c before it fails.
It makes me wonder if i got a corrupted version of IBT or something is wrong somewhere in my settings/system.

I read this article about failing in IBT, and turned out it was his DRAM that caused issues. Something with timings. https://forums.tomshardware.com/threads/intel-burn-test-failing-while-stock.2801845/
My ram is on default/auto and this is the timings set by auto:
CAS# to Latency (CL) 15.0 clocks
RAS# to CAS# delay (tRCD) 15 clocks
RAS# Precharge (tRP) 15 clocks
Cycle time (tRAS) 36 clocks
Row Refresh Cycle Time (tRFC) 374 clocks
Command Rate (CR) 1T

Do you see any issues there?

 

[rigg42]

Increase the voltage slightly. You're not stable in IBT. That window pops up if instability is detected. You'd likely not pass realbench either. Are you running small FFT in Prime? You won't get useful temps if you're not. Those temps you're hitting in IBT are par for the course with my 9900k and 8700k at this voltage. These chips run hot.

 

[katulen]

Increase the voltage slightly. You're not stable in IBT. You'd likely not pass realbench either.

I did 30 minutes of OCCT Linpack and 30 minutes of Realbench, no issues. For some weird reason i feel like its my ram. Did my timings look correct tho?

Im running 1344 - 1344 fft. Read somewhere that was the settings i should use.

Well 100c is tjmax for i7 9700k right?

And if im on 1.355v now, should i go 1.360v or 1.365v ?

 

[rigg42]

Try 1.360v first. See if it stabilizes IBT. If not just keep bumping it up 5 mv. I've always just used XMP profiles without issue. I've never really dove into RAM timings extensively. I'm not the guy to ask on that front.

 

[katulen]

Try 1.360v first. See if it stabilizes IBT. If not just keep bumping it up 5 mv. I've always just used XMP profiles without issue. I've never really dove into RAM timings extensively. I'm not the guy to ask on that front.

Already, well it for sure seems like XMP is causing issues with my CPU Overclock stability. However, maybe default is causing issues as well.
But i mean if i can run OCCT Linpack with 90% ram usage, i guess its alrite.

I will try 1.360v.

 

[rigg42]

I did 30 minutes of OCCT Linpack and 30 minutes of Realbench, no issues. For some weird reason i feel like its my ram. Did my timings look correct tho?

Im running 1344 - 1344 fft. Read somewhere that was the settings i should use.

Well 100c is tjmax for i7 9700k right?

And if im on 1.355v now, should i go 1.360v or 1.365v ?

I've had realbench crash after several hours. You'll need to do an extended test at some point if you want to be truly confident in your overclock. I don't really use Prime as a stability test. I use it to find out how much voltage I can cool. An hour of small FFT is useful for this task. The CPU will throttle at 100c. I think tjmax is actually 115c. Keep in mind these are max temps during a stress test. I doubt your cpu would even hit 80c while gaming at this voltage.

 

[katulen]

I've had realbench crash after several hours. You'll need to do an extended test at some point if you want to be truly confident in your overclock. I don't really use Prime as a stability test. I use it to find out how much voltage I can cool. An hour of small FFT is useful for this task. The CPU will throttle at 100c. I think tjmax is actually 115c. Keep in mind these are max temps during a stress test. I doubt your cpu would even hit 80c while gaming at this voltage.

Well there we go.

I had to go all the way to 1.385v and it passed IBT.
Max temp: 88c
Vdroops to 1.350v, no vboosts.

So what do you recommend me to do now?

 

[rigg42]

Is your XMP enabled?

 

[katulen]

Is your XMP enabled?

No disabled.

I have a few days ago tryid with 1.380v + XMP running OCCT & P95 and got bsod 15 mins in. So concluded my XMP wasnt stable, since i now can run P95 & OCCT as low as 1.350v

Want me to try enable XMP again and see? If so, run which test ?

 

[rigg42]

Might be worth trying at this point. I've used corsair vengeance with several Intel and AMD systems in the last 6 months. I've never once had issues with the XMP profiles. Maybe load the XMP and downclock to 2666.

 

[rigg42]

You can try to increase VCCIO voltage if the memory is still being flukey. I usually run at 1.1v. I believe its safe up to 1.2v or more. Overclocking on Intel is a pain in the ass. Ryzen is so much easier.

 

[katulen]

Might be worth trying at this point. I've used corsair vengeance with several Intel and AMD systems in the last 6 months. I've never once had issues with the XMP profiles. Maybe load the XMP and downclock to 2666.

Okay i passed IBT with XMP I enabled. I didnt downclock them. They are at 3000mhz.
But i have 2 options with XMP. I got XMP I and XMP II.

And it says something in the buttom of bios, where the "description text is". It says:
[XMP I] Load the DIMM's default XMP memory timings with BCLK frequency and other parameters, optimized by Asus.
[XMP II] Load the DIMM's complete default XMP profile.

You can try to increase VCCIO voltage if the memory is still being flukey. I usually run at 1.1v. I believe its safe up to 1.2v or more. Overclocking on Intel is a pain in the ass. Ryzen is so much easier.

When i enable XMP I, it also changes some voltages stuff.
DRAM voltages to 1.350v

And it also changes VCCIO & VCCSA. Don't know their defaults, but they are now, with XMP I enabled:
VCCIO: 1.312v - 1.344v
VCCSA: 1.216v.

So suggestion for next move?

 

[rigg42]

Stick with XMP-1. Those voltages seem high. Asus tends to run these too high on auto. I'd put them below 1.2 and try again.

 

[katulen]

Stick with XMP-1. Those voltages seem high. Asus tends to run these too high on auto. I'd put them below 1.2 and try again.

Okay, so:
XMP 1
Manually change VCCIO to 1.2v
Manually change VCCSA to 1.2v <-- or is the "auto" at 1.216v good enough?
And what about DRAM voltages?

And just run IBT again or?