My CPU overclocks badly

[VladusD]

Hello,

 

I recently bought a new PC. I have a Phanteks P400S chasis (with only 2 stock 120mm chasis fans ), i7-9700k, Asus Prime Z370-A (latest BIOS), GTX 1080, 16GB 3200Mhz RAM, the Noctua NH-U12S cooler and a 650W gold-rated PSU from EVGA. I've recently tried setting up an initial, modest overclock (4.8Ghz all-core with 1.25 vcore) and it failed. It even failed with 1.26 voltage. Temperatures were in the high 70s in cinebench and fine in games until it crashed.

 

Here are some of the BIOS settings that I set:

• Asus multi-core enhancement disabled
• SVID behaviour -  typical case
• XMP Profile
• AVX offset 0
• Sync all cores @ 4.8 Ghz
• BCLK Freq 100 Mhz
• BCLK  Freq: DRAM Frequency Ratio: auto
• Xtreme tweaking: enabled
• TPU: keep current settings
• CPU SVID support: disabled
• CPU Core/cache limit max: max (255.50)
• Ring down bin: auto
• Min/max cpu cache ratio: auto
• BCLK aware adaptive voltage: disabled
• CPU core/cache voltage: 1.25
• CPU load/line calibration: level 6 (out of 7)
• CPU current capability: 140% (max)
• CPU power duty control: extreme
• Intel speedStep: off
• turbo enabled
• long/short duration package power limit: max (4095)

Now, I am not experienced with overclocking at all, but I did my best to document myself by watching different YT videos from seemingly credible sources, read professional reviews, forum posts etc and I came up with this initial conservative in my view profile that I would have then slowly improved. The problem is even this fails, even though looking at other overclocks of the i7-9700k, this looks like it should work (many people recommend an initial 1.200 voltage @ 4.8Ghz for the i7-9700k). And yes, I know that it varies from CPU to CPU, but not buy this much, I guess....

I am thinking of a few possible explanations:

1. I am screwing something in the BIOS settings
2. Insufficient cooling, either for the CPU (the NH-U12S not enough for i7-9700k OC?) or for the VRMs (meaning I should buy after market chasis fans) or both. (the stock chasis fans in particular seem shitty and low-powered, i might be wrong tho)
3. A particularly unlucky case of the silicon lottery.

 

Sorry for the long post and thank you in advance for your help!

[Brooksie359]

Why would you have turbo enabled if you are setting a fixed overclock?

[xg32]

16 minutes ago, VladusD said:

.

bump the voltage up to 1.35 and see how far it goes, if it can't boot at 4.8 avx, then yes it is garbage, but i doubt it. i recommend 5.0, 1.35v avx offset 2, also try backing off on the LLC a bit, you are not even hitting 1.3v yet.

[VladusD]

4 minutes ago, xg32 said:

bump the voltage up to 1.35 and see how far it goes, if it can't boot at 4.8 avx, then yes it is garbage, but i doubt it.

Wow, I guess it would run, but it would be kind of ridiculous to run such a low frequency at such a high voltage, not to mention the generated heat.

 

6 minutes ago, Brooksie359 said:

Why would you have turbo enabled if you are setting a fixed overclock?

Could this have been the cause of my problems?

[xg32]

Just now, VladusD said:

Wow, I guess it would run, but it would be kind of ridiculous to run such a low frequency at such a high voltage, not to mention the generated heat.

 

Could this have been the cause?

1.35 isn't high voltage, and 4.8 avx isn't low frequency, are u checking your voltage in hwmonitor with LLC at 6? because it could be dropping to like 1.175v on loads.

[VladusD]

5 minutes ago, xg32 said:

bump the voltage up to 1.35 and see how far it goes, if it can't boot at 4.8 avx, then yes it is garbage, but i doubt it. i recommend 5.0, 1.35v avx offset 2, also try backing off on the LLC a bit, you are not even hitting 1.3v yet.

To be honest, I doubt my chasis & cpu cooling could take 5.0 Ghz and 1.35 v

[Jurrunio]

I'd day your expectations on your 9700k is unrealistically high. Not all 9th gen CPUs do 5GHz even at 1.4V, since I've just seen a 9900k crash at that settings in a MSI Z390 Godlike motherboard cooled by an Arctic Freezer 360

[VladusD]

5 minutes ago, xg32 said:

1.35 isn't high voltage, and 4.8 avx isn't low frequency, are u checking your voltage in hwmonitor with LLC at 6? because it could be dropping to like 1.175v on loads.

I barely know anything in depth about this stuff, so high LLC can cause voltage drops on loads? I thought it was the opposite? Meaning it adds voltage during high load to prevent a big decline in vcore

[VladusD]

5 minutes ago, Jurrunio said:

I'd day your expectations on your 9700k is unrealistically high. Not all 9th gen CPUs do 5GHz even at 1.4V, since I've just seen a 9900k crash at that settings in a MSI Z390 Godlike motherboard cooled by an Arctic Freezer 360

This could be true. Although you are speaking about the 9900k, which might be more difficult to reach higher frequencies due to hyper-threading (from the benchmarks & reviews I've seen). Ofc I could be wrong

[Enderman]

Ok cool, so just add more voltage or lower the frequency until stable.

That's how you overclock.

[xg32]

8 minutes ago, VladusD said:

I barely know anything in depth about this stuff, so high LLC can cause voltage drops on loads? I thought it was the opposite? Meaning adds voltage during high load to prevent a big decline in vcore

on asus and asrock boards it's high LLC=higher drop, there's even a graph for it on the newer boards, i wouldn't worry too much about heat just for benchmarks, and chances are your cpu will do 4.8 at 1.35v or lower (actual reading in hwmonitor after vdroop), and like @Jurrunio said, ur expectations were on the high side.

 

For reference

 

https://siliconlottery.com/collections/coffeelake-r/products/9700k50g

 

4.7 avx would be garbage tier, 4.8 would be average on 1.35v, if u find out ur cpu can indeed do 4.8 at 1.35, u can always go back down to 4.7 at a much lower voltage, since the 9700k and 9900k stock voltage is already in the parabolic part of the voltage curve.

 

I would not worry about temps in this process under 85C, just lower it back down later after you are done.

[Jurrunio]

7 minutes ago, VladusD said:

This could be true. Although you are speaking about the 9900k, which might be more difficult to reach higher frequencies due to hyper-threading (from the benchmarks I've seen). Ofc I could be wrong

I would try 4.5GHz at 1.3V from the start if I were you.

 

btw that 9900k refused to even post at 4.7GHz 1.35v back when it's sitting in an Asus Z370-E

[VladusD]

2 minutes ago, Enderman said:

Ok cool, so just add more voltage or lower the frequency until stable.

That's how you overclock.

I know and that's what I've been doing, just wanted a little feedback from people to make sure I am grounded in reality.

[Enderman]

Just now, VladusD said:

I know and that's what I've been doing, just wanted a little feedback from people to make sure I am grounded in reality.

1.2v sounds very low for 4.8GHz but you're supposed to work up from there, not expect that it would be stable there.

[VladusD]

5 minutes ago, Jurrunio said:

I would try 4.5GHz at 1.3V from the start if I were you.

 

btw that 9900k refused to even post at 4.7GHz 1.35v back when it's sitting in an Asus Z370-E

Actually I have it seemingly stable rn at 1.21v @ 4.6Ghz and I will try to lower the vcore further in a few days after I see that it is truly stable. Although I still plan to try higher freqs in the not so distant future

[VladusD]

2 minutes ago, Enderman said:

1.2v sounds very low for 4.8GHz but you're supposed to work up from there, not expect that it would be stable there.

Yeah, but mine crashed at 1.26v. Although yeah I got the point from this thread that even that might have been optimistic

[corrado33]

4 minutes ago, VladusD said:

I know and that's what I've been doing, just wanted a little feedback from people to make sure I am grounded in reality.

First off, intel 8th and 9th gen chips run hot. If you don't have a decent cooling system, say goodbye to your hopes at a high overclock.

 

Secondly, this is the silicon lottery. You MAY reach those speeds, you MAY not be able to overclock at all, it happens. 

[VladusD]

7 minutes ago, xg32 said:

on asus and asrock boards it's high LLC=higher drop, there's even a graph for it on the newer boards, i wouldn't worry too much about heat just for benchmarks, and chances are your cpu will do 4.8 at 1.35v or lower (actual reading in hwmonitor after vdroop), and like @Jurrunio said, ur expectations were on the high side.

 

For reference

 

https://siliconlottery.com/collections/coffeelake-r/products/9700k50g

 

4.7 avx would be garbage tier, 4.8 would be average on 1.35v, if u find out ur cpu can indeed do 4.8 at 1.35, u can always go back down to 4.7 at a much lower voltage, since the 9700k and 9900k stock voltage is already in the parabolic part of the voltage curve.

 

I would not worry about temps in this process under 85C, just lower it back down later after you are done.

Alright. First of all, thank you for all your replies and I appreciate your feedback. I suspect that you are right regarding my high expectations and I guess I will continue working to find the min stable vcore for each freq. One thing tho, could you please tell me a little bit more about LLC? Should I always have it on level 5, or should I bump it to 6 when having ambitious OCs and high vcores?

[corrado33]

7 minutes ago, VladusD said:

Alright. First of all, thank you for all your replies and I appreciate your feedback. I suspect that you are right regarding my high expectations and I guess I will continue working to find the min stable vcore for each freq. One thing tho, could you please tell me a little bit more about LLC? Should I always have it on level 5, or should I bump it to 6 when having ambitious OCs and high vcores?

LLC (or Load Line Calibration) makes sure that your CPU core voltage stays at the voltage you set it at and not any lower. Depending on the motherboard, your settings may vary but essentially if you have the LLC set to high, the CPU voltage will definitely stay above your CPU voltage value but it will also overshoot (so it'll go a lot higher when the CPU is ramping up. Say your voltage is set at 1.35. If you have LLC at the highest, and you start a stress test, the voltage will likely shoot up to 1.39 or something before settling again at 1.35.

 

If you have LLC low, however, the voltage may "droop". So when you run a stress test instead of the CPU getting the 1.35V that you said it should have, it'll be getting less, maybe 1.33 or 1.32. This MAY cause an instability. Basically, monitor your VCore in HWinfo and see if it drops at all under load. If it doesn't, then lower the LLC value. 

 

EDIT: And by "high" and "low" LLC state, I mean check your mobo's manual. In my asus motherboard, 9 is the highest level, and 1 is the lowest. (9 will overshoot a lot, where as 1 may not provide enough voltage.) Apparently motherboards are different. 

[VladusD]

1 minute ago, corrado33 said:

LLC (or Load Line Calibration) makes sure that your CPU core voltage stays at the voltage you set it at and not any lower. Depending on the motherboard, your settings may vary but essentially if you have the LLC set to high, the CPU voltage will definitely stay above your CPU voltage value but it will also overshoot (so it'll go a lot higher when the CPU is ramping up. Say your voltage is set at 1.35. If you have LLC at the highest, and you start a stress test, the voltage will likely shoot up to 1.39 or something before settling again at 1.35.

 

If you have LLC low, however, the voltage may "droop". So when you run a stress test instead of the CPU getting the 1.35V that you said it should have, it'll be getting less, maybe 1.33 or 1.32. This MAY cause an instability. Basically, monitor your VCore in HWinfo and see if it drops at all under load. If it doesn't, then lower the LLC value. 

Yeah, I understand that. But poster xg32 in this thread said that on Asus mobos, high LLC can cause higher drops (I hope I interpreted his answer correctly, sorry if I didn't), which is seemingly opposite of what you said (and what I knew). Although in my experience, with a LLC of level 6 out of 7, I typically only have around -0.015 vdroop, which is ok, I guess? And so in HwInfo the min/max recorded vcore are typically around +- 0.015v from the vcore set in the BIOS,  in all scenarios.

[corrado33]

1 minute ago, VladusD said:

Yeah, I understand that. But poster xg32 in this thread said that on Asus mobos, high LLC can cause higher drops (I hope I interpreted his answer correctly, sorry if I didn't), which is seemingly opposite of what you said (and what I knew). Although in my experience, with a LLC of level 6 out of 7, I typically only have around -0.015 vdroop, which is ok, I guess? And so in HwInfo the min/max recorded vcore are typically around +- 0.015v from the vcore set in the BIOS,  in all scenarios.

That's why I said read your manual. It can be different for different motherboards. Mine literally says IN my bios what it does. 

[VladusD]

Alright, thanks to anyone who replied. Because of the nature of the topic, there is no exact solution for my question, but I think I received valuable information that will be helpful to me and probably others. To summarize, here are my takeaways:

• There probably isn't something "wrong" per se with my CPU. I might have been overly optimistic about the vcore-frequency combinations that I could do and I should continue trying to find the optimal settings for my particular chip.
• I will continue experimenting with various LLC levels and other settings to find the best setting for my PC/mobo.
• The 8th and 9th gen are hot chips, might want to buy better CPU/chasis cooling if I want to do more ambitious OCs.