Question 9700K Overclock

Feb 12, 2019
10
0
10
Guys,

Please, be comprehensive with me. I used to do overclock a long time ago with a K6-2 500. I'm getting updated with the new options.

I have a Maximus XI Hero (Asus Rog) and a 9700k and I did an overclock to 4.8 GHz, I did it simple just changing (from default BIOS options):

  • Sync all cores to 48x multiplier
  • Load Line Calibration (LLC) to 3

I used the oficial Intel App and I ran the Burn In test for 30 minutes (https://ibb.co/mqCR3JT). I got the following results (see the image it's better):

  • clock speed stable and fixed to 4.8 GHz
  • hottest core (89 degree)
  • VID: 1.3345v
  • Mem frequency: 99.98

I have some questions:

  1. That temperature is normal? (I now that less than 85 is the best however the turn test is quite agressive, so how much can be considered ok to a burn in test)?
  2. Intel Boost stopped to work (I don't care much about, I just wanna now if it's normal)
  3. VID (every article said that less is better, it's running automatically, that 1.3345v is considered ok)?
  4. What other parameters should I change to try to get 4.9 GHz?

My cooling system is a Deepcool 120 EX (AIO) and I'm using the grizzly kryonaut thermal paste.
 

Eximo

Titan
Ambassador
  1. A little high, but I wouldn't worry about it. Chances are you won't see those temperatures under normal use.
  2. A fixed clock speed is fixed, doing what you told it to do.
  3. VID is the expected demand voltage programmed in at the factory if I recall. Core voltage is what you want to see, if you have left the voltage on automatic and used LLC it may be getting a bit higher than you want.

4. No real definitive answer, if you can manage to lower the voltage and increase the clock speed then you might be able to go a little farther. Otherwise you are going to need better cooling than a 120mm radiator.

4.8Ghz across 8 cores is pretty darn good.
 
Feb 12, 2019
10
0
10
Eximo,

Thank you so much for answering me. 4.8 Ghz is good however I like to race (race simulator - iRacing) using VR and I sadly discover that VR + Physics Calculation + Frame preparation consume a lot of CPU, my GPU is left over. The bottleneck is CPU.

Thank you for sharing your thoughts about temperature, I believe that too that the burn is a stability test - It can't be considered as a normal use. I'm glad that you share the same opinion. The Intel software set 105 degrees as limit to the Tj Max (my processor passed in all tests - including temp). The prime number test is the heaviest one and I don't think that just playing all the 8 cores will be running at 100% doing prime numbers calculation, for example. I will monitor the daily use temp and I guess that I will be happy.

I did a test now and I'm running with 4.9 (I just change the multiplier to 49x). The Intel Burn In Test ran for 30 minutes:
  • clock speed stable and fixed to 4.9 GHz
  • hottest core (94 degrees) - plus 5 degrees from 4.8 GHz
  • VID: 1.3793v (processor consumed a little more)
  • Mem frequency: 99.95
Just more three questions:

1. Why mem frequency changes if I didn't change any memory parameter? If I set to auto the sync all core it ran with default configuration for processor and the memory frequency keep the 100 Mhz... is it normal?

2. About the voltage, should I change it to manual or use the offset trying to find the best value? What's better option? (Now, I’m using the default - auto)

3. Are there any other BIOS options that should I look?

Sorry for writting so much, I found a lot of material on the internet however I didn't find people sharing their experiences in the detail. So I decided to write down my own experience.
 
Last edited:

Eximo

Titan
Ambassador
1. I am not sure what frequency you are measuring. The BCLK should be 100Mhz and 99.95 is close enough, but the memory frequency is going to be some multiplier of that. If you are running 2666Mhz memory then it would be a 13.3 multiplier. Cache is usually the CPUs base frequency.

2. Auto is almost never recommended when overclocking as motherboards typically apply too much voltage to ensure maximum compatibility. As you have not looked at core voltage I can't really say what you should do. You should be using more monitoring tools than Intel's provided ones.

Offset voltages are for the chips C-states mostly. It lets the CPU ramp down and use low voltage. They are just as dangerous as LLC though and can exceed the set core voltages when used. Negative offsets allow less power/temperature, but can drive the voltage too low for operation (crashes) With a fixed frequency it is best to set the voltages manually.

3. The most basics of overclocking should be looking at Vcore, VIN, LLC, Multiplier, maybe cache frequency. There are dozens more settings to look at when going for maximum performance. Plenty of guides out there you can look at. I like the ones on overclock.net, but there are many board/brand specific ones that might be more useful to you.

Keep in mind you should be monitoring all the voltages using a tool like Hardware Monitor. (Though you can never trust just one application, try a few and see which one best reflects reality, sometimes they'll pull in the wrong sensor) Skylake/Kabylake/Coffeelake are all pretty much the same architecture though, so I doubt there are many remaining errors.
 
How well you can do is determined by your luck in getting a good chip.
As of 12/07/2018
What percent can get an overclock at a somewhat sane 1.360v Vcore
And AVX offset = 2.

I7-9700K

5.1 27%
5.0 72%

The intel burn test does just that, it executes the most heat generating instructions.
Not really what you will do normally.

The best stress test is YOUR workload.
Set your ram to your best XMP profile first.

Monitor with CPU-Z and keep the vcore to that 1.36 number.
CPU-Z has a simple stress test that you can try.

Monitor your temperatures with HWmonitor
The throttle point will be about 100c.
Load temperatures up to 85c. should be ok.

Implement speedstep and adaptive voltage.
That will reduce the multiplier and vcore when the cpu has little to do.

Just run with whatever multiplier you feel good about.
A BSOD while gaming is not a disaster.
If you experience a problem in your normal operations back off your multiplier a notch.
 
Feb 12, 2019
10
0
10
Eximo,

Thank your for your time.

1. I'm using the 100Mhz, the default detected by BIOS

2. I installed CPU-Z and the results were:

CPU Core Voltage (idle): 1.314v @ 4.9
CPU Core Voltage (during burn-in): 1.181v @ 4.9

  • I have the AI Suite (from Asus) and it showed the same information for the Core Voltage than CPU-Z so I guess that the information is right.
  • I thought that under load (burn in) the voltage would be higher.

3. Why do you say that LLC is dangerous?


1. I am not sure what frequency you are measuring. The BCLK should be 100Mhz and 99.95 is close enough, but the memory frequency is going to be some multiplier of that. If you are running 2666Mhz memory then it would be a 13.3 multiplier. Cache is usually the CPUs base frequency.

2. Auto is almost never recommended when overclocking as motherboards typically apply too much voltage to ensure maximum compatibility. As you have not looked at core voltage I can't really say what you should do. You should be using more monitoring tools than Intel's provided ones.

Offset voltages are for the chips C-states mostly. It lets the CPU ramp down and use low voltage. They are just as dangerous as LLC though and can exceed the set core voltages when used. Negative offsets allow less power/temperature, but can drive the voltage too low for operation (crashes) With a fixed frequency it is best to set the voltages manually.

3. The most basics of overclocking should be looking at Vcore, VIN, LLC, Multiplier, maybe cache frequency. There are dozens more settings to look at when going for maximum performance. Plenty of guides out there you can look at. I like the ones on overclock.net, but there are many board/brand specific ones that might be more useful to you.

Keep in mind you should be monitoring all the voltages using a tool like Hardware Monitor. (Though you can never trust just one application, try a few and see which one best reflects reality, sometimes they'll pull in the wrong sensor) Skylake/Kabylake/Coffeelake are all pretty much the same architecture though, so I doubt there are many remaining errors.
 
Feb 12, 2019
10
0
10
My motherboard have the avx instruction core ratio negative offset.

Is it the same than Avx offset?

How well you can do is determined by your luck in getting a good chip.
As of 12/07/2018
What percent can get an overclock at a somewhat sane 1.360v Vcore
And AVX offset = 2.

I7-9700K

5.1 27%
5.0 72%

The intel burn test does just that, it executes the most heat generating instructions.
Not really what you will do normally.

The best stress test is YOUR workload.
Set your ram to your best XMP profile first.

Monitor with CPU-Z and keep the vcore to that 1.36 number.
CPU-Z has a simple stress test that you can try.

Monitor your temperatures with HWmonitor
The throttle point will be about 100c.
Load temperatures up to 85c. should be ok.

Implement speedstep and adaptive voltage.
That will reduce the multiplier and vcore when the cpu has little to do.

Just run with whatever multiplier you feel good about.
A BSOD while gaming is not a disaster.
If you experience a problem in your normal operations back off your multiplier a notch.
 
AVX instructions use high voltage and generate heat. Fortunately, they are not that common in suage for games and such.
The offset will reduce the multiplier when avx instructions are present.
I see no harm in that at 2.
Under load, the voltage might be reduced( I think it is called vdroop)
If you get sophisticated with overclocking you can adjust those parameters.
LLC applies more voltage under load for stability.
I suppose LLC can be dangerous if it results in excessive voltage at any time.
Under load your voltages are quite good and I think you could go higher if you want.
 
Feb 12, 2019
10
0
10
Thank you Geofet, let me ask you something else.

For AVX Instruction Core Ratio Negative Offset my BIOS says the following text in the information bullet:

"Substract this value from your core ratio to get the ratio at wich AVG applications run. Setting a negative offset may sometimes require a higher core voltage to stabilize."

I've read the same information that you shared, when the AVX is reduced the the heat could be reduced too (open space for higher clock), however my motherboard said that negative offset can require more voltage to stabilize. It is that right?


AVX instructions use high voltage and generate heat. Fortunately, they are not that common in suage for games and such.
The offset will reduce the multiplier when avx instructions are present.
I see no harm in that at 2.
Under load, the voltage might be reduced( I think it is called vdroop)
If you get sophisticated with overclocking you can adjust those parameters.
LLC applies more voltage under load for stability.
I suppose LLC can be dangerous if it results in excessive voltage at any time.
Under load your voltages are quite good and I think you could go higher if you want.
 
Sorry, I am no expert on that level of overclocking.

The statements are a bit confusing in that increasing a negative offset might really be increasing it.
You might look to overclocking forums for your brand of motherboard.
Perhaps it is best to not fool with that particular setting.
 
Feb 12, 2019
10
0
10
Just to share more information. I tried to run the 4.9 GHz with LLC setted to 2. The Windows showed the blue screen of death. I noticed that running the burn in with LLC setted to 2 the processor tried to run with lower voltage from 1.181v to 1.146v. I think that was the freeze cause.
 

Eximo

Titan
Ambassador
LLC and offest voltages apply voltage on top of the existing parameters. LLC (Load Line Calibration) is to prevent vdroop (voltage drop) from creating instability. However, the timing isn't perfect. So if the CPU comes off load that higher voltage is still available for brief periods, or during the duration of the higher load. Offsets for things like AVX will apply extra voltage when AVX instructions are running, same issue, when the AVX instructions are no longer in use there is a brief period where the voltage may exceed the base settings.

On your latest attempt that says you need more LLC or to increase the base core voltage. As long as temperature allows.

Setting a negative offset is to control temperature/power usage. Typically you would also set the CPUs core clocks to go lower when running AVX as well. Since it is a higher than normal load on the CPU, but that circuitry is more purpose built, the idea there is that the sacrifice in clock speed is worth it for the gain in performance on AVX tasks.
 
Feb 12, 2019
10
0
10
I've tested other configurations. I was not happy with auto vcore, I think it was not optimized. I achieved the following results:
  • Sync all cores to 49x multiplier
  • Load Line Calibration (LLC) to 5
  • Vcore (manual set): 1.310
  • AVX 0
  • Hottest core: 95 (using IBT stress test)

Second option:
  • Sync all cores to 50x multiplier
  • Load Line Calibration (LLC) to 5
  • Vcore (manual set): 1.310
  • AVX -2
  • Hottest core: 93 (using IBT stress test)
In fact I don't know if higher clock is better with avx -2. Running my race simulator the average clock as 4.82 (because under avx instruction the clock is reduced to 4.800)... so my stable 4.9 with avx 0 is at average, better.
 
Feb 12, 2019
10
0
10
While gaming the temp achieve 80 degrees at maximum, for the below configuration:

I stabilize with avx -1.

  • Sync all cores to 50x multiplier
  • Load Line Calibration (LLC) to 5
  • Vcore (manual set): 1.310
  • AVX -1
  • Hottest core: 95 (using IBT stress test - it's really stressfull. I ran prime95 blend test during two hours and the max temp was 82 degrees )
 
Feb 12, 2019
10
0
10
Eximo,

Just other thing... first, thank you for sharing.

I don't know if I'm right however in my IMHO the prime small fft and large fft are dangerous and they are far away from our processor reality needs. For me, it's a program created by crazy guys to crazy other people that love to find the limit regardless the real application of things.

I'm saying that because I ran the small fft and I got round error and the processor literally burned out (100 degrees in less than one minute for @5.0Ghz OC AVX -1)... so I didn't get happy with this. I set back to stock configuration and I ran it again: the result was the same 100 degrees (except by no round erros during the three minutes that I have courage to let it running). I have a really good water cooler (120x deep cooler), grizzly thermal paste, MB rog maximus XI, case with good wind flow, a processor without HT (9700k) and a 650w font to supply energy.
So the stock configuration should ran smooth however prime95 goes beyond any person needs. I thing that it can be good for some scientifc processing stability test because it really has crazy math operations that literally burns the processor.

For me, this app should be banned from threads talk about stability, it's way beyond our user needs: that's to the most is gaming or personal use. It's a dangerous application because most of us have some cheap component and it take they from way of their limits.