10900K OC Guide for ASUS ROG Maximus Hero XII

[captain420 3,099]

Could someone please share some info on the best way to go about OC? What speed, voltages, settings, etc should I start off with? My goal is to reach a stable 5.2GHz OC. But I would like to take my time to make sure I'm doing everything correctly. I plan on starting with 5.0GHz on all cores and performing some tests to see if it's stable, and if it passes proceed to 5.1GHz and so forth. My questions are:

1. What CPU Voltage should I use for 5.0GHz, 5.1GHz, and 5.2GHz all core OC? 
2. What AVX offset to use?
3. Is Loadline Calibration needed?
4. Are there any other settings that I need to change?
5. What apps to use for testing stability? Prime95, Cinebench R20, R50, Intel Extreme Tuning, CPU-Z, etc.
6. For a stable OC, what are considered safe CPU temps and voltage range that I can expect?

I'm quit nervous when it comes to OC as I don't want to end up damaging anything because the components in my new build are quite expensive. I would love to start off with a safe and stable settings for an achievable 5.0GHz first and then take it from there. 

For example, if I'm running Prime95 to test my CPU overnight and the temps remain at around 90-95C. Does this mean I neeed to lower my settings to get those temps down? If yes, what temp range is considered good?

[Hyperfocal 186]

Nobody can give you voltages for a stable overclock because every CPU differs slightly. Here's how I approach overclocking the 10900K with an Asus motherboard.

1. Enter BIOS and enable MCE and XMP.
2. Boot to Windows and perform a test with something like Prime 95 while monitoring temperatures and voltages. You should see an all-core frequency of 4.9 GHz, likely at a voltage of 1.2X. Assuming you have thermal headroom (your load temperatures are under 85c):
3. Enter BIOS. Set CPU Core Ratio to Sync All cores. Set the All Core Ratio Limit to "49." Set AVX Instruction Core Ratio Negative Offset to a value of zero. Set Load Line Calibration to Level 4, which will limit the Vdroop when under load. Set CPU Core/Cache Voltage Override to something slightly higher than the voltage from your test in step two, probably around 1.3V.
4. Boot to Windows and perform the same test again to verify stability and temperature.
5. Enter BIOS and bump the All Core Ratio Limit to "50."
6. Boot to Windows and perform the same test again to verify stability and temperature.

Continue to bump the All core Ratio Limit up by 100MHz and test. You'll either run out of thermal headroom (temps will hit 85C) or the test will crash. If the test crashes you can try bumping up the voltage by ten millivolts (0.01V) and testing again or simply lower the core ratio back to its last "stable" value and call it a day. I would stop after either hitting 85C or reaching 1.35 V. Depending on the particular CPU, you may not reach 5.2 GHz. Some appear to reach 5.2 at a sane voltage while others can barely break 5.0 GHz. Once you've found a configuration that appears stable, you'll need to run a stress test for five hours or so to make sure it truly is.

Disclaimer: Do this at your own risk. This process works for me, but YMMV.

[Bob Scott 7,716]

11 hours ago, captain420 said:

Could someone please share some info on the best way to go about OC? What speed, voltages, settings, etc should I start off with? My goal is to reach a stable 5.2GHz OC. But I would like to take my time to make sure I'm doing everything correctly. I plan on starting with 5.0GHz on all cores and performing some tests to see if it's stable, and if it passes proceed to 5.1GHz and so forth. My questions are:

1. What CPU Voltage should I use for 5.0GHz, 5.1GHz, and 5.2GHz all core OC? 
2. What AVX offset to use?
3. Is Loadline Calibration needed?
4. Are there any other settings that I need to change?
5. What apps to use for testing stability? Prime95, Cinebench R20, R50, Intel Extreme Tuning, CPU-Z, etc.
6. For a stable OC, what are considered safe CPU temps and voltage range that I can expect?

I'm quit nervous when it comes to OC as I don't want to end up damaging anything because the components in my new build are quite expensive. I would love to start off with a safe and stable settings for an achievable 5.0GHz first and then take it from there. 

For example, if I'm running Prime95 to test my CPU overnight and the temps remain at around 90-95C. Does this mean I neeed to lower my settings to get those temps down? If yes, what temp range is considered good?

 

First, do NOT use the ASUS automatic overclock..  Lots of reports of crazy voltages when using that feature. 

1.  Core voltage will depend on your cooling solution and how well you fare in the silicon lottery.  I started out at 1.35v Core Override and 5.1 GHz HT Off, but I have a very strong custom water loop.  With a less aggressive cooling solution, I might start a little lower and eek my way up.  Currently I'm at 5.2 GHz@1.36v HT Off.  WIth HT On, I'd start at 1.35v and 5.0 GHz. 

2.  AVX offset zero.  P3D does make use (but not extensive use) of the AVX instruction set.  I don't like the idea of my CPU oscillating the clock speed up/down every time it detects AVX instructions in the stack.

3.  Line Load Calibration at Level 4 is the ASUS-recommended setting for overclocking that board, and it's working fine for me

4.  You'll need MCE enabled to override the stock turbo logic and run all cores clocked up.  I set the min/max cache multipliers to 3 less than the core multiplier, so for 5.1 GHz (core mult 51) that'd be 48.  Depending on your memory, you may want/need to adjust Vccio and Vccsa...I use 1.18v Vccsa and 1.15v Vccio for a 4-DIMM config of fairly tight 3600 MHz CAS 15 RAM.  You don't need a lot there unless you're running some really fast 4GHz+ DIMMs, and in any event you don't want those voltages much above 1.25v for 24/7 operation.  The mobo will increase those voltages some under load--best to watch them with the monitor in HWInfo.

5.  Memtest86+ v8.4 for RAM testing--set it up on a bootable USB stick.  This is the first test to run after getting it to boot up and setting your XMP memory timings.  A bad memory clock will booger up everything that follows, so it's important to get that right first.  I typically will run 4 passes of Memtest86+ (~4.5 hours) before doing anything else.  After that I install the OS and all the relevant drivers, and make a disk image of the base OS so that if an overclocking excursion corrupts the drive (rare, but can happen), all I have to do is reinstall the image rather than do the entire OS install again.

RealBench 2.56 is a good (and free) stress test that presents a more realistic load profile than Prime95.  RealBench will give you a reasonable approximation of temps with P3D running.  If you do use Prime95, the newer versions present heavy AVX loads by default--you'll want to disable the AVX by setting the three AVX lines to 0 in the config (settings in local.txt)  If it passes P95 with AVX off, it should be OK for P3D's much more moderate loads.  If you intend to run other real world heavy loads like video rendering, then test with AVX, but expect a lower best overclock.  I typically run a P95 non AVX test for 2-4 hours, then a 4-8 hour run of RealBench. 

HWInfo 6.28 is a good all-around program for monitoring all the internals--speeds, voltages, etc. 

CoreTemp 1.51.1 for CPU core temp monitoring.

6.  I will accept core temps bumping into the mid 80s for a heavy stress test like P95+AVX, but I generally strive to keep cores at/below 80 in testing.  I would not want CPU core voltage above 1.40, which means setting something like 1.37v to account for LLC corrections, and that only with a very good water cooling system. 

 

There's no reason to push to the ragged edge with this chip.  After I get my OC stable, I generally downclock the CPU a couple steps for extra margin while I install/configure software.  I installed and test ran P3D v4.5 at a reduced speed of 5.0 GHz with the RealAir Duke and the result was just stunning.  If your cooling or your chip only gets you to 5.0 or even 4.9--on *10* cores--this thing is still a beast.

Have fun

[Bob Scott 7,716]

One more tip--once you get into the system with your memory still running at SPD (stock) settings, start by flashing your BIOS.  My board shipped with an old BIOS (version 0222), and that was causing a problem with a couple of the chipset component drivers.  0607 is current.  I didn't notice that at first (who would actually expect a new mobo to ship with a current BIOS??), and when I got around to flashing the new BIOS it wiped all my settings and stored profiles out.  Which brings me to another technique I use--keep copious notes on what you're doing and changing.  It can come in handy later!

[captain420 3,099]

7 minutes ago, w6kd said:

One more tip--once you get into the system with your memory still running at SPD (stock) settings, start by flashing your BIOS.  My board shipped with an old BIOS (version 0222), and that was causing a problem with a couple of the chipset component drivers.  0607 is current.  I didn't notice that at first (who would actually expect a new mobo to ship with a current BIOS??), and when I got around to flashing the new BIOS it wiped all my settings and stored profiles out.  Which brings me to another technique I use--keep copious notes on what you're doing and changing.  It can come in handy later!

Yup, I made sure that was the 1st thing I did. I'm on the latest BIOS 0607 right now. I will follow the above tips and go from there! I appreciate the help guys! 🙂 Will keep you updated on my OC. I heard a majority of the 10900k will easily OC to 5.0GHz. I hope that's true. Because my goal is to hit 5.2GHz stable.

Right now I'm running stock settings, the only setting that I've changed is the XMP1. But pretty much everything else is stock. Should I change my XMP1 setting back to Auto before I proceed with the OC?

My SP is 63 and Cooler Pts is 141-160. Not sure if that means anything.

I'm running a NZXT Kraken Z73 AIO.

[captain420 3,099]

Are these settings okay for stress test in Prime95?

With these default settings. My average CPU temp is around 65C and voltage around 0.906v - 0.932v. CPU Core speed around 3798.00 - 3900.94MHz. Multiplier around x38-x39.

I haven't changed anything in my BIOS yet. So it's still default with the exception of XMP1 for my RAM. 

image uploader

Edited July 22, 2020 by captain420

[Bob Scott 7,716]

Yikes...I don't like what i'm seeing with your CPU case temp at default settings.  Your core voltage is default at 0.84v, and the Turbo Boost logic has downclocked all the cores to 3.8GHz under an all-core load, and yet you still have 69 deg C CPU case temp.  That's too high under those conditions.

You need to check the pump/fan curves and possibly the mount of the water block to the CPU before doing anything else.  My guess is the fans or pump are not spooling up enough and/or too late.  On the H100i on my desktop PC, I have the fan curve set to start coming up with coolant temp rising above 30 deg C and it hits max at 60 deg.  The fluid in the loop will act as a heat reservoir if the pump and fans are late to the game.

 

[captain420 3,099]

[G-RFRY 1,708]

https://www.youtube.com/watch?v=dHVzRV1VwvA

5.3 to far 5.2 all cores 24/7 very unlikely, it may not CTD but motherboards now days use the VRM to protect the hardware (throttle)

Edited July 22, 2020 by G-RFRY

[Hyperfocal 186]

I would begin ramping up the fans no later than 40C, and preferably around 35C.

[G-RFRY 1,708]

Don`t get fooled by the myth that CTD`S are due heat unstable voltage is also a common cause.  

[Bob Scott 7,716]

I'd recommend using the coolant temp rather than the CPU temp to control the fan/pump speeds.  That way it won't constantly and rapidly cycle up/down as the CPU load and temp changes--the coolant temp changes relatively gradually, where the CPU temp changes quickly with the load.  That way you won't have that really annoying surging fan noise...and it's better to react to heat accumulation in the system rather than an instantaneous measure of heat input to the system.

Your fan curve is far too passive, IMHO.  On my 7700K system (no longer used for simming), I start my fans and pump ramping-up at 30 deg C coolant temp, and hit 100% at 60 deg C.  With a 10-core CPU kicking out 250W or more, you need that cooler to come in much earlier and more aggressively.  Maxxing it at 50 deg and above would not be out of the question.

 

[captain420 3,099]

7 hours ago, w6kd said:

Your fan curve is far too passive, IMHO.

What do you mean by this?

7 hours ago, w6kd said:

I'd recommend using the coolant temp rather than the CPU temp to control the fan/pump speeds.

How do I do this?

[Bob Scott 7,716]

1 hour ago, captain420 said:

What do you mean by this?

How do I do this?

I mean that your fans and pump are set to ramp up later than they need to.  You posted the curves above...your radiator fans are set to run at 50% until the CPU gets to 55 deg, then it ramps up to 100% at 60 deg.  Not sure what that 140mm rear fan is about.

I can't help you with a specific how-to for the Kraken config...I've never used one.  From what I can see in your posted screenshot, though, there is very likely an option to configure the fans and pumps as a function of coolant rather than CPU temp, and you should be able to drag the nodes on those curves around to reset the curve so that the fans and pump come in earlier than 55 deg C.

 

 

[captain420 3,099]

This is the first system I had with liquid cooling. So I'm still new to this. I will play around with it some more and do some fine tuning. Thanks for your input. I appreciate it.