[SOLVED] Overclocked my i9 9900K to 5.1ghz with 1.325v, is that OK? Did I do it right?

[Albert Wesker]

Hello there. I am no expert, so kindly enlighten me.

So I overclocked my i9 9900k to 5.1ghz with 1.325v using a Kraken X62. 0 AVX target. It's stable.

Passed Cinebench R15 multiple times in different occasions. Passed Prime95. No issues during games.

Max temps when using Cinebench were 72-75, 85 on Prime95, and 50 when gaming (depends on the game, most of the time it stays between 35-40)

Voltage drops to 1.279v under load. LLC: level 7

My motherboard (Asus ROG Maximus XI Formula Z390) shows 1.314v in the bios. It does fluctuate between 1.314v and 1.323v on CPU-Z.

So is that all OK? Because when I looked for other i9ners who OCed their CPU to 5.1ghz, most of them had a much higher voltage like 1.38, 1.4... Do I need that???

Temps seem OK to me and speed frequency stays 5.1ghz at all times (it drops to 5.08, 5.09, but I guess that's normal).

Oh and RAM is OCed at 3600mhz C15 using 1.35v, 1.050v VCCIO and 1.050v CPU system agent voltage.

Thanks.

 

[Darkbreeze]

It's stable.

Passed Cinebench R15 multiple times in different occasions. Passed Prime95. No issues during games.

That, means very little. And it CERTAINLY doesn't mean that it's stable. Prime95 is really not a suitable stress test for stability, it's best usage is for thermal compliance testing.

Cinebench is not a reliable metric for stability testing. Realbench however, is.

As far as "Did I do it right", that answer is probably no. You can't just throw predetermined numbers at an overclock and assume that it's "right". Overclocking is not a set of numbers that works the same from one system to another. It's a PROCESS that must be followed to individually determine where the best configuration is at for any given system, because every motherboard, CPU and memory kit is going to act differently, and is going to like different settings, even if you compare two systems with identical parts.

If you can pass 15 minutes of Prime95 version 26.6 (And ONLY version 26.6) running the "Small FFT" option, and remain below 80°C, then it is technically thermally compliant.

If you can run 8 hours of Realbench stress test (Not the benchmark, click on the stress test option) with half your full memory capacity selected in the "memory to use" field, then the system is about as stable as it's ever going to be. There are further tests you can do with Prime95 to offer additional reassurance of stability, such as the Blend test with custom settings, and those are outlined as well.

Below is my quick and dirty methodology. At the bottom is a link to my full overclocking guide for beginners.

Quick and dirty overview of overclocking/stability validation procedure.

Set CPU multiplier and voltage at desired settings in BIOS. Do not use presets or automatic utilities. These will overcompensate on core and other voltages. It is much better to configure most core settings manually, and leave anything left over on auto until a later point in time if wish to come back and tweak settings such as cache (Uncore) frequency, System agent voltage, VCCIO (Internal memory controller) and memory speeds or timings (RAM) AFTER the CPU overclock is fully stable.

Save bios settings (As a new BIOS profile if your bios supports multiple profiles) and exit bios.

Boot into the Windows desktop environment. Download and install Prime95 version 26.6.

Download and install either HWinfo or CoreTemp.

Open HWinfo and run "Sensors only" or open CoreTemp.

Run Prime95 (ONLY version 26.6) and choose the "Small FFT test option". Run this for 15 minutes while monitoring your core/package temperatures to verify that you do not exceed the thermal specifications of your CPU.

(This should be considered to be 80°C for most generations of Intel processor and for current Ryzen CPUs. For older AMD FX and Phenom series, you should use a thermal monitor that has options for "Distance to TJmax" and you want to NOT see distance to TJmax drop below 10°C distance to TJmax. Anything that is MORE than 10°C distance to TJmax is within the allowed thermal envelope.)

If your CPU passes the thermal compliance test, move on to stability.

Download and install Realbench. Run Realbench and choose the Stress test option. Choose a value from the available memory (RAM) options that is equal to approximately half of your installed memory capacity. If you have 16GB, choose 8GB. If you have 8GB, choose 4GB, etc. Click start and allow the stability test to run for 8 hours. Do not plan to use the system for ANYTHING else while it is running. It will run realistic AVX and handbrake workloads and if it passes 8 hours of testing it is probably about as stable as you can reasonably expect.

If you wish to check stability further you can run 12-24 hours of Prime95 Blend mode or Small FFT.

You do not need to simultaneously run HWinfo or CoreTemp while running Realbench as you should have already performed the thermal compliance test PLUS Realbench will show current CPU temperatures while it is running.

If you run the additional stability test using Prime95 Blend/Small FFT modes for 12-24 hours, you will WANT to also run HWinfo alongside it. Monitor HWinfo periodically to verify that no cores/threads are showing less than 100% usage. If it is, then that worker has errored out and the test should be stopped.

If you find there are errors on ANY of the stability tests including Realbench or Prime95, or any other stress testing utility, you need to make a change in the bios. This could be either dropping the multiplier to a lower factor or increasing the voltage while leaving the multiplier the same. If you change voltage or multiplier at ANY time, you need to start over again at the beginning and verify thermal compliance again.

A more in depth but general guide that is still intended for beginners or those who have had a small amount of experience overclocking can be found here:


*CPU overclocking guide for beginners

 

[Albert Wesker]

It's stable.

Passed Cinebench R15 multiple times in different occasions. Passed Prime95. No issues during games.

That, means very little. And it CERTAINLY doesn't mean that it's stable. Prime95 is really not a suitable stress test for stability, it's best usage is for thermal compliance testing.

Cinebench is not a reliable metric for stability testing. Realbench however, is.

As far as "Did I do it right", that answer is probably no. You can't just throw predetermined numbers at an overclock and assume that it's "right". Overclocking is not a set of numbers that works the same from one system to another. It's a PROCESS that must be followed to individually determine where the best configuration is at for any given system, because every motherboard, CPU and memory kit is going to act differently, and is going to like different settings, even if you compare two systems with identical parts.

If you can pass 15 minutes of Prime95 version 26.6 (And ONLY version 26.6) running the "Small FFT" option, and remain below 80°C, then it is technically thermally compliant.

If you can run 8 hours of Realbench stress test (Not the benchmark, click on the stress test option) with half your full memory capacity selected in the "memory to use" field, then the system is about as stable as it's ever going to be. There are further tests you can do with Prime95 to offer additional reassurance of stability, such as the Blend test with custom settings, and those are outlined as well.

Below is my quick and dirty methodology. At the bottom is a link to my full overclocking guide for beginners.

Quick and dirty overview of overclocking/stability validation procedure.

Set CPU multiplier and voltage at desired settings in BIOS. Do not use presets or automatic utilities. These will overcompensate on core and other voltages. It is much better to configure most core settings manually, and leave anything left over on auto until a later point in time if wish to come back and tweak settings such as cache (Uncore) frequency, System agent voltage, VCCIO (Internal memory controller) and memory speeds or timings (RAM) AFTER the CPU overclock is fully stable.

Save bios settings (As a new BIOS profile if your bios supports multiple profiles) and exit bios.

Boot into the Windows desktop environment. Download and install Prime95 version 26.6.

Download and install either HWinfo or CoreTemp.

Open HWinfo and run "Sensors only" or open CoreTemp.

Run Prime95 (ONLY version 26.6) and choose the "Small FFT test option". Run this for 15 minutes while monitoring your core/package temperatures to verify that you do not exceed the thermal specifications of your CPU.

(This should be considered to be 80°C for most generations of Intel processor and for current Ryzen CPUs. For older AMD FX and Phenom series, you should use a thermal monitor that has options for "Distance to TJmax" and you want to NOT see distance to TJmax drop below 10°C distance to TJmax. Anything that is MORE than 10°C distance to TJmax is within the allowed thermal envelope.)

If your CPU passes the thermal compliance test, move on to stability.

Download and install Realbench. Run Realbench and choose the Stress test option. Choose a value from the available memory (RAM) options that is equal to approximately half of your installed memory capacity. If you have 16GB, choose 8GB. If you have 8GB, choose 4GB, etc. Click start and allow the stability test to run for 8 hours. Do not plan to use the system for ANYTHING else while it is running. It will run realistic AVX and handbrake workloads and if it passes 8 hours of testing it is probably about as stable as you can reasonably expect.

If you wish to check stability further you can run 12-24 hours of Prime95 Blend mode or Small FFT.

You do not need to simultaneously run HWinfo or CoreTemp while running Realbench as you should have already performed the thermal compliance test PLUS Realbench will show current CPU temperatures while it is running.

If you run the additional stability test using Prime95 Blend/Small FFT modes for 12-24 hours, you will WANT to also run HWinfo alongside it. Monitor HWinfo periodically to verify that no cores/threads are showing less than 100% usage. If it is, then that worker has errored out and the test should be stopped.

If you find there are errors on ANY of the stability tests including Realbench or Prime95, or any other stress testing utility, you need to make a change in the bios. This could be either dropping the multiplier to a lower factor or increasing the voltage while leaving the multiplier the same. If you change voltage or multiplier at ANY time, you need to start over again at the beginning and verify thermal compliance again.

A more in depth but general guide that is still intended for beginners or those who have had a small amount of experience overclocking can be found here:


*CPU overclocking guide for beginners

Thanks for providing all of that.

So I should try Realbench now? I am happy with how my PC is operating at those settings to be honest.

 

[Darkbreeze]

Yes, you should. Because if it can pass 8 hours of Realbench, and you want to run ALL 8 hours. If it runs for 7 hours and 30 minutes, and errors out, it is not stable. If it runs for 7 hours and 59 minutes, it is not stable. Even if it runs for 8 hours, and errors out at 8 hours and five minutes, it is not FULLY stable, but the chances are very, very good that if it can run 8 hours of Realbench, you will never see an issue with corruption of data due to micro-errors, which is something you will never notice during normal use or gameplay until it has cumulatively corrupted your file system and data to the point where it begins showing problems and errors in the file system become evident.

At that point, all is lost. None of the data can be "fixed" or recovered. If there are errors, you want to know about them NOW so you can either reduce the multi or increase the voltage, and then test again. Full instructions on testing can be found in the guide.

 

[Albert Wesker]

Yes, you should. Because if it can pass 8 hours of Realbench, and you want to run ALL 8 hours. If it runs for 7 hours and 30 minutes, and errors out, it is not stable. If it runs for 7 hours and 59 minutes, it is not stable. Even if it runs for 8 hours, and errors out at 8 hours and five minutes, it is not FULLY stable, but the chances are very, very good that if it can run 8 hours of Realbench, you will never see an issue with corruption of data due to micro-errors, which is something you will never notice during normal use or gameplay until it has cumulatively corrupted your file system and data to the point where it begins showing problems and errors in the file system become evident.

At that point, all is lost. None of the data can be "fixed" or recovered. If there are errors, you want to know about them NOW so you can either reduce the multi or increase the voltage, and then test again. Full instructions on testing can be found in the guide.

Damn, this is interesting. I am excited. So I cannot use my PC during Realbench?

 

[Darkbreeze]

No, that is why I recommend running it overnight or while you are away for the day. It is VERY intensive with both CPU and memory resources. In fact, if you can pass with half your memory allocated, it's a pretty good idea to run it again with the full installed capacity of memory as well. OR, run it WITH the full amount the first time, and if there are errors, try it again with half the memory. So if you have 16GB installed, then 8GB in the "memory to use" field.

But if you try with the full amount from the start, and it passes, then there is no need to run it again, as it is fine. If it doesn't, and you CAN pass with half the memory selected, then the CPU is likely stable, fine, but there may be configuration issues with the memory that need to be addressed.

I also have a guide on that.

*Resolving memory problems and setting up XMP/DOCP/AMP profiles

 

[Albert Wesker]

No, that is why I recommend running it overnight or while you are away for the day. It is VERY intensive with both CPU and memory resources. In fact, if you can pass with half your memory allocated, it's a pretty good idea to run it again with the full installed capacity of memory as well. OR, run it WITH the full amount the first time, and if there are errors, try it again with half the memory. So if you have 16GB installed, then 8GB in the "memory to use" field.

But if you try with the full amount from the start, and it passes, then there is no need to run it again, as it is fine. If it doesn't, and you CAN pass with half the memory selected, then the CPU is likely stable, fine, but there may be configuration issues with the memory that need to be addressed.

I also have a guide on that.

*Resolving memory problems and setting up XMP/DOCP/AMP profiles

This is great man. Thanks so much for the help. Will report back once I use Realbench.

 

[Albert Wesker]

No, that is why I recommend running it overnight or while you are away for the day. It is VERY intensive with both CPU and memory resources. In fact, if you can pass with half your memory allocated, it's a pretty good idea to run it again with the full installed capacity of memory as well. OR, run it WITH the full amount the first time, and if there are errors, try it again with half the memory. So if you have 16GB installed, then 8GB in the "memory to use" field.

But if you try with the full amount from the start, and it passes, then there is no need to run it again, as it is fine. If it doesn't, and you CAN pass with half the memory selected, then the CPU is likely stable, fine, but there may be configuration issues with the memory that need to be addressed.

I also have a guide on that.

*Resolving memory problems and setting up XMP/DOCP/AMP profiles

BTW, I just rechecked the voltage under load, it's actually 1.279v.

 

[Darkbreeze]

Yes, your actual core voltage will change under a load. That's knows as vdroop. Changing the LLC (Load line calibration) setting to a higher, or lower, depending on the board and which direction numerically they've chosen to be an INCREASE in LLC, will help to alleviate vdroop, but you have to be careful because while increasing LLC will help with stability issues without having to increase the core voltage (Vcore, and several other names depending on the platform and generation) it will also FOR SURE increase thermal response as well.

A CPU with a 1.35v core voltage and and LLC of 3 (Assuming 1 is lowest and 10 is highest, although some boards don't have numerical schemes and instead have something like standard, high, extreme, etc.) that sits at about 80°C under a full Prime95 Small FFT load might easily jump to 85 or 90 degrees by changing the LLC to 4 or 5, or any stepped increase, so you have to be careful with it and be sure to immediately retest the thermal compliance EVERY time you make a change to LLC.

So dropping to 1.279v under a load probably isn't terrible unexpected if you have a minimal LLC set. I'd check your LLC settings in the BIOS to see what they are currently set at. There are lots of tutorials covering LLC settings for just about every platform out there.

Other settings that might have a big impact on stability and thermals are the System agent and VCCIO voltages.