H100i V2 - Decent idle temps, awful load temps (90C)

[Knight_]

I've recently just upgraded my PC by grabbing an i9-9900K and the temps are a little worrying. At idle, it sits around 30-38C with some fluctuations into 40C seemingly at random. Not awful. Whilst gaming it's not terrible either with temps seeing 50-70C depending on the game and how CPU intensive it tends to be.

 

However, when running Cinebench R15 or playing RPCS3 (both very demanding, I know), the temps hit Tjunction Max at 90-99C. That's not right for a stock 9900K despite how demanding that is.

 

Here are my simple specs:

 

• i9-9900K @ 3.60Ghz (Stock)
  
• ROG Maximus Hero XI
  
• 16GB G.Skill Trident Z 3600Mhz CL16
  
• Corsair H100i V2
  
• MSI Gaming 6G GTX 980Ti
  
• Corsair 570X case
  
• EVGA Supernova 850P2
  

 

More details:

 

The H100i is about 2-3 years old, worked fine with my previous CPU and is setup in a push/pull configuration at the front of the case. Two of the three front fans are pushing air into the radiator and two fans connected to the radiator are pulling the air through the case. The third fan at the front is just pushing air in. I then have 2 top-mounted fans exhausting air along with an external fan above the I/O shield doing the same. This isn't my PC but it has the same case and H100i layout.

 

The 3 font fans are connected to CHA_FAN1 through FAN3. The 2 radiator fans are connected to the H100i pump directly and the 2 top-mounted fans are connected to CPU_FAN and CPU_OPT. The H100i pump itself is connected to AIO_PUMP. The external fan above the I/O shield is connected to H_AMP.

 

I used a peasized amount of CoolerMaster E1 IC Essential thermal compound when placing the heatsink over the CPU. The CPU itself fluctuates between 0.9V to 1.3V when idle. It averages 1.2V @ 4500-4800Mhz 35-37C (reading it right now from CoreTemp 1.14). Cable management is very tidy.

 

Some things I've tried:

 

• Setting all fans to 100% speed in the BIOS. No change
   
  
• Disabling XMP. No change to temps
   
  
• Disabling [AUTO] on ASUS' AI controlled overclocking / voltage control / clock control. Increase in temps to about 40C idle
   
  
• Disabled MCE in the ASUS bios. Temps increased by a small amount it seems
   
  
• Enabling Legacy USB Support and XHCI Support in the BIOS
   
  
• Ensuring the AIO_PUMP header is enabled and set to DC control / no Q-Control
   
  
• Reinstalled, downgraded, updated Corsair Link 4 as the H100i V2 doesn't show up, no drivers except for "Corsair Bus" exist in devman (This still didn't cause it to show up). Ran dism and hydros and siusb don't exist - I manually installed hydros and no change. I downloaded SIV from these forums and it couldn't recognise the H100i although it's clearly working
   
  
• I moved the 3-pin pump connector to "CPU_FAN" instead of "AIO_PUMP" and swapped the USB header the connector was plugged into
   
  
• [Not done yet] I've ordered some 12.5W/mK Kryonaut thermal compound and am going to reapply the paste tomorrow when it arrives
  

 

Can anyone else offer any advice or suggestions? I find it really odd that it's so terrible under extreme load but not awful whilst just gaming or idle. It seems to fluctuate in temperate and voltage but, from what I've read, it seems normal. I initially thought this was a thermal paste issue but now it's seeming more like an issue with the pump and Corsair Link not recognising it / perhaps adjusting it depending on the temperature and load. It's been like this since when I built it on Saturday and I've researched it and tested with it non-stop since - I'm so incredibly frustrated at this point.

Edited July 23, 2019 by Knight_

[c-attack]

The one thing missing from all of the above is your actual Vcore levels when running Cinebench or anything else. The core voltage is the single most dominant factor in end CPU temperature and the first thing you should always look at. Your Asus BIOS has quite a few little tweaks necessary to run any kind of synthetic benchmark. MCE is just one of them and it is likely impossible to do so on Auto voltage.

 

Any kind of cooler problem will reveal itself at idle as well in the coolant temperature (H100i Temp). Most people see a coolant temp of around +4-7C above room temp. Anything more than 30C probably bears investigation unless your room is 30C as well. Any Lake CPU is not going to sit still on the desktop and there is no “flat idle” like Sandy Bridge or even Haswell-E. You will see cycling Vcore and cpu core temps while at the desktop. The newer cpus use other methods to reduce energy consumption.

[Knight_]

The one thing missing from all of the above is your actual Vcore levels when running Cinebench or anything else. The core voltage is the single most dominant factor in end CPU temperature and the first thing you should always look at. Your Asus BIOS has quite a few little tweaks necessary to run any kind of synthetic benchmark. MCE is just one of them and it is likely impossible to do so on Auto voltage.

 

Any kind of cooler problem will reveal itself at idle as well in the coolant temperature (H100i Temp). Most people see a coolant temp of around +4-7C above room temp. Anything more than 30C probably bears investigation unless your room is 30C as well. Any Lake CPU is not going to sit still on the desktop and there is no “flat idle” like Sandy Bridge or even Haswell-E. You will see cycling Vcore and cpu core temps while at the desktop. The newer cpus use other methods to reduce energy consumption.

 

My CPU is idling at around 41-45 right now with just Chrome open. It's a hot day here in the UK but it's not 35-40C in my room, that's for sure. So perhaps the idle temps aren't as good as I expected then. Also, thanks, I think I understand the energy and vcore fluctuations now.

 

As for the vcore during games and Cinebench or RPCS3 (the latter isn't synthetic btw):

 

I'm getting 1.3V - 1.41V having just booted up the Sims 4 and going in-game. I'm almost worried to check Cinebench again given that it maintains 95-100C for about 3-5 seconds. Not something I'm super comfortable with right now. Nonetheless, seeing it hit 1.4V in games is pretty worrying and I believe gives us enough of an idea, right? That's well out of the ordinary, I'm assuming.

 

Could you give me a rundown on what 'tweaks' I'll need to do to get this under control? I've heard of people running 9900K's at 5Ghz on all cores at lower voltages than me right now.

[c-attack]

Ok, well this is likely mystery solved. You can run a 9900k at 5.0GHz for 8 cores at 1.25-1.275V, so 1.30-1.40 is a truckload too much. Welcome to the current state of affairs, at least with Asus boards. You need to “tune the board” right out of the box to make it run at default levels. I’ll try and make a BIOS setting list for you, but I need to be in front of my Z390 to do it. For immediate relief, set a fixed/manual voltage of 1.20. That should cover you to about 47x100 on all cores. This can be tuned to exactly what you want later.

[Knight_]

Ok, well this is likely mystery solved. You can run a 9900k at 5.0GHz for 8 cores at 1.25-1.275V, so 1.30-1.40 is a truckload too much. Welcome to the current state of affairs, at least with Asus boards. You need to “tune the board” right out of the box to make it run at default levels. I’ll try and make a BIOS setting list for you, but I need to be in front of my Z390 to do it. For immediate relief, set a fixed/manual voltage of 1.20. That should cover you to about 47x100 on all cores. This can be tuned to exactly what you want later.

 

I'll give this a shot right now. I appreciate your speedy replies, it provides some relief after having been so clueless and frustrated for the past few days.

 

I'll edit this comment and let you know how it goes.

 

EDIT: Disregard last edit if you saw it. I've set the voltage at 1.20 and, like you said, it's sat at 4700ish Mhz. Temps still occasionally shoot up to like 50/60C when first starting Windows. Cinebench still reaches 90-95C but at least it's not hitting 100C now. One thing I've noticed is the H100i fans getting loud *after* the test is done. It's almost like it's reacting to the temperatures but like 10-20 seconds after they've been reached. I feel like some more tweaking is needed in the BIOS but I feel at least more comfortable letting it just run naturally now.

Edited July 23, 2019 by Knight_

[c-attack]

The h100i fans will react to coolant temperature, not cpu temp. The fans won’t do a thing for the heat created at the pins under the cpu where the voltage is applied. That part of the cooling is all conductive. The coolant then picks up the heat at the block and transports the heat to the radiator where the fans help disperse it elsewhere. The fans do affect the rate of dispersion and thus “coolant temperature”. +1C coolant temp = +1C CPU temp. Same for reduction.

 

It just occurred to me if you have run the overclocking/tuning program (Cpu, not fans) through AI Suite, you will want to reset that using AI Suite before messing around in the Bios. For AI Suite to work, it must be able to override the bios settings. Trying to change the bios without undoing AI Suite can lead to some nasty boot loops.

Edited July 23, 2019 by c-attack

[Knight_]

The h100i fans will react to coolant temperature, not cpu temp. The fans won’t do a thing for the heat created at the pins under the cpu where the voltage is applied. That part of the cooling is all conductive. The coolant then transports the heat to the radiator where the fans help disperse it elsewhere. The fans do affect the rate of dispersion and thus “coolant temperature”. +1C coolant temp = +1C CPU temp. Same for reduction.

 

It just occurred to me if you have run the overclocking/tuning program (Cpu, not fans) through AI Suite, you will want to reset that using AI Suite before messing around in the Bios. For AI Suite to work, it must be able to override the bios settings. Trying to change the bios without undoing AI Suite can lead to some nasty boot loops.

 

As far as I'm aware, I've not run any overclocking or tuning program through AI Suite. The closest thing I'm even aware of is the "AI Overclock Tuner" and "CPU Core Ratio". The latter I've not changed from default and the former I've changed from auto to manual at one point as I meant to say in the OP where I mention "disabling AI overclocking".

 

Is this what you mean? I have already changed the voltage and, fortunately, haven't run into any problems or boot loops.

[c-attack]

Voltage would not be catastrophic. AI Suite would use the old or the new and the worst case is a bsod for being too low. However, specific load line settings or the ones we need to deactivate to stop the voltage piling on could cause a irreconcilable conflict. You would know if you ran the tuner. It takes 5-10 minutes and tests a specific multiplier and voltage for 30-120s before moving on. Usually multiple restarts involved. Anyway, more precautionary. I did this to myself unintentionally several years back. Horrific next 60 minutes until I flashed the bios back to default.

[Knight_]

Voltage would not be catastrophic. AI Suite would use the old or the new and the worst case is a bsod for being too low. However, specific load line settings or the ones we need to deactivate to stop the voltage piling on could cause a irreconcilable conflict. You would know if you ran the tuner. It takes 5-10 minutes and tests a specific multiplier and voltage for 30-120s before moving on. Usually multiple restarts involved. Anyway, more precautionary. I did this to myself unintentionally several years back. Horrific next 60 minutes until I flashed the bios back to default.

 

Ah, I get you. That sounds awful.

 

Thanks for everything by the way. Although not completely fixing my issues, just that simple voltage change has given me a lot of relief.

 

In terms of a more complete solution, I'd really appreciate it if you could also write up that BIOS settings list. I've a feeling that a few more edits and I'll be good to go with temperatures hopefully perfect. No worries if not, I'd fully understand.

 

On another note, you mention the fans and the coolant from before. Could I not plug the radiator fans into the CPU_FAN and CPU_OPT headers to have them speed up when the CPU gets hot rather than the coolant? Perhaps there's another underlying issue but it did really feel like they were incredibly delayed when running Cinebench.

[c-attack]

You could move the fans to cpu/opt, but I won’t lower end cpu temp. When you initiate a large load and the Vcore jumps from idle ~0.70 to 1.2x volts, the cpu temps will jump 35-50C depending on CPU construction (aka silicon lottery). That value cannot be reduced except by lowering voltage. A heavy and extended load with a 9900 and 240mm radiator will increase coolant temp by +10C. Of that, slightly faster or slower fan speeds only affect performance by 1-2C for a CPU level wattage. So of the +45-55C increase, only 2C is adjustable. Not really what you are looking for. If all it took was faster fans, we would would all run 4000 rpm and 6GHz. Nope. Cpu voltage is the limiting factor.

 

I will get a checklist up for you, but I am still at work. It won’t be until tomorrow your time.

[c-attack]

BIOS -> F7 to Adv. BIOS if not already there

 

Extreme Tweaker Column

 

AI Overclock -> Manual or XMP 2 (if using XMP for RAM)

 

Asus Multi-core enhancement - Disabled (if you want all cores to run xx frequency, do that thought the multiplier)

 

SVID Behavior -> Best case scenario. This should only matter for Adaptive voltage, but you may wish to use that.

 

AVX offset -> Only need to use this if you are going to frequently utilize programs with this instruction. However, many stressors use it, so set it to 0 for now for testing purposes.

 

CPU Core Ratio -> 47x100? What do you want to run. This can be set up as per core, but I don't think it is really more efficient and on most CPUs it will trigger the larger Vcore anyway, so you might as well take off the restriction. Still, user choice.

 

External Digi Power+ -> Sub-menu -> CPU Load Line Calibration (LLC) = 6. I am estimating based on a mid-level settings. I don't trust it on auto to not pile on voltage during R15. This should cap it. I use LLC 7 for 5.0. 8 is max and too much for anything but extreme overclocking.

 

Also in Digi Power sub menu -> VRM spread spectrum -> Disable

 

Internal CPU Power Management -> Sub-menu ->

1) Long duration Power = 4095

3) Short duration Power = 4095 (these two won't make things cooler, but may be necessary to run the stress tests without throttling)

5 & 6) IA AC/DC Load line = 0.01 for both. This is critical for adaptive voltage. Does nothing if fixed.

 

 

CPU Core/Cache Current Limit = 255.50 (same as other power limit. Necessary for testing. Should not come up for normal use).

 

CPU Core Cache Voltage = Adaptive or Manual. If adaptive, enter desired Vcore value on the "additional Turbo Mode voltage" line. You probably want adaptive for normal use. People talk about using fixed to find the OC point then changing, but with these settings the voltage should be regulated. You can skip the manual and set the adaptive voltage for testing. My view is if it fails adaptive but works on fixed, it's not really stable anyway. Too close to the mark.

 

 

Of course, the central element to all of this is what frequency do you want to run? That will determine the Vcore above. I pay attention to what others report, but I have not binned 100 9900Ks. I am making estimates based on what I see. Each CPU is relatively unique, so this likely requires some guess work to start.

[Knight_]

BIOS -> F7 to Adv. BIOS if not already there

 

Extreme Tweaker Column

 

AI Overclock -> Manual or XMP 2 (if using XMP for RAM)

 

Asus Multi-core enhancement - Disabled (if you want all cores to run xx frequency, do that thought the multiplier)

 

SVID Behavior -> Best case scenario. This should only matter for Adaptive voltage, but you may wish to use that.

 

AVX offset -> Only need to use this if you are going to frequently utilize programs with this instruction. However, many stressors use it, so set it to 0 for now for testing purposes.

 

CPU Core Ratio -> 47x100? What do you want to run. This can be set up as per core, but I don't think it is really more efficient and on most CPUs it will trigger the larger Vcore anyway, so you might as well take off the restriction. Still, user choice.

 

External Digi Power+ -> Sub-menu -> CPU Load Line Calibration (LLC) = 6. I am estimating based on a mid-level settings. I don't trust it on auto to not pile on voltage during R15. This should cap it. I use LLC 7 for 5.0. 8 is max and too much for anything but extreme overclocking.

 

Also in Digi Power sub menu -> VRM spread spectrum -> Disable

 

Internal CPU Power Management -> Sub-menu ->

1) Long duration Power = 4095

3) Short duration Power = 4095 (these two won't make things cooler, but may be necessary to run the stress tests without throttling)

5 & 6) IA AC/DC Load line = 0.01 for both. This is critical for adaptive voltage. Does nothing if fixed.

 

 

CPU Core/Cache Current Limit = 255.50 (same as other power limit. Necessary for testing. Should not come up for normal use).

 

CPU Core Cache Voltage = Adaptive or Manual. If adaptive, enter desired Vcore value on the "additional Turbo Mode voltage" line. You probably want adaptive for normal use. People talk about using fixed to find the OC point then changing, but with these settings the voltage should be regulated. You can skip the manual and set the adaptive voltage for testing. My view is if it fails adaptive but works on fixed, it's not really stable anyway. Too close to the mark.

 

 

Of course, the central element to all of this is what frequency do you want to run? That will determine the Vcore above. I pay attention to what others report, but I have not binned 100 9900Ks. I am making estimates based on what I see. Each CPU is relatively unique, so this likely requires some guess work to start.

 

Massively appreciate it, mate. Just a few questions;

 

Biggest one first; I've noticed that my H100i is getting 125 points in the BIOS. That's well below what others are getting and is on-par with a 120mm radiator or air cooler. I also do plan on using RPCS3 which really likes AVX and, given my poor cooling score, wouldn't it be better to have the AVX offset as 1-2? I imagine that once I've applied this new thermal paste that the score will go up but 125pts is pretty concerning right now.

 

Also, I noticed yesterday that my CPU System Agent voltage was around 1.4V. I've heard that's insane and causes a lot of heat and anything above 1.3V is unnecessary. According to OCCCT, my VCCIO is 1.4V too... Given my RAM, I really don't think I should have 1.4V on either, right? DRAM voltage is obviously at 1.35.

 

For CPU core ratio: I wanna hit 5Ghz at some point for sure. The BIOS reckons 1.33v for 5Ghz but CPU-Z right now is reporting 4800Mhz / 48x100 at 1.19v. It is fluctuating massively though, sometimes dropping to 3600Mhz every few seconds. I've heard this could be due to C-States and SpeedStep and can easily be made more consistent. Also, wouldn't I be better off on "Sync all Cores" and just setting them as 48-50?

 

As you can see, I've done some research of my own looking into overclocking guides for my exact motherboard, CPU and sometimes even the same AIO. I think I'm going to be more conservative for now just so I can get a good look at my temps and stability. Like I mentioned, I'm probably going to use an AVX offset of 1, Min/Max cache ratio of 43 and an LLC of 6 whilst below 5Ghz. I'm definitely not going to touch anything until you've gotten back to me though. Research is good an' all but some circumstance specific advice and information is safer.

 

Cheers again.

Edited July 24, 2019 by Knight_

[c-attack]

Ugh.. Always copy your long posts before clicking submit. There is often a white screen and a lost post waiting for you on the other side.

 

Take 2.

 

 

1) Any kind of "cooler score" from the BIOS is garbage. It has no means to access or measure your coolant temperature differential. Presumably it was designed for air coolers and is a simple CPU temp rating, but then it can't really measure your software usage from there either. Do I get a good score if I sit there and do nothing? Do I get a bad score if I run Prime 95 a lot? Who knows. No matter what, this is a worthless metric intended for people who have no understanding of how their computer work and need some value as a reassurance.

 

I don't use and am not familiar with RPCS3. However, you can use whatever AVX offset you want when finished, but if you are trying to test 5.0@1.32v (or whatever), nearly every stress test program uses AXV and will downclock you to the AVX level, thus you are testing 4.8@1.32 with AVX, not 5.0. While theoretically possible, it is unlikely you would be 5.0/1.xx vols stable, and then shaky at 4.8 AVX. A 200MHz offset should cover things.

 

2) VCCIO and VCCSA - I have no idea why Asus is still piling on the IO and System Agent voltages. Unlike most other values, more voltage is not "hotter but more stable". It can be be hotter and less stable. Try 1.175v for both. That is what I use for 3600C16 (4x8GB). I use 1.20 for both at 4000C17. Most likely you are in that range. This platform is not so picky that you will have problems shifted one way or the other and anything in that area is likely fine. Like others settings, this one is individual CPU specific. Warning signs are a frozen screen, black screen, or loss of input.

 

3) Sync All vs Per Core - Sync all cores sets the turbo multiplier for all 8 cores to the specified level and it will run this when the conditions permit (sufficient load, under power/temp limits). This is the same as what MCE does, except you get to set the voltage. Overclocking with auto voltage (MCE) is what gets you into the worry zone immediately. All core lets you set the voltage, as long as you don't use auto. Per Core synchronization is similar to the Intel default specifications. When 1 core is loaded, you can have 50x100. When 2 cores are loaded, 48x100. 3=47, 4=46... 8 cores = 3.6GHz. The per core setting will let you set each core count to whatever you want (6 cores for 5.0, 8 for 4.8), but the problem is you are always going to have 8 cores active The CPU constantly shifts the load between cores and this mechanic does not keep in step with those parameters. You will never ever see 5.0, except maybe opening a folder on the desktop. You will run 4.8 under load all the time. All this does is add uncertainly to your results.

 

3.6 is the standard frequency for the 9900K and probably whatever you were doing was light enough not to invoke the the turbo boost. It is also possible to unnecessarily create this condition by fiddling with the c-states/turbo/Speedstep settings in the BIOS. You want Speedshift On, Speedstep Auto, C-states Auto (to start). That should disable C-states whenever there is any load. It will partially relax them when at the desktop. If you need the lowest possible desktop power levels, you need to manually enable C-states and enable most of them. That may have some unintended effects on gaming and full near full load operations. If for whatever reason you need 100% uptime from the processor, use the Windows power plan to set it to Maximum Performance (from Balanced). This should be a pre-existing setting in the power plan. This will keep the cores up at the max turbo frequency when any load is present. It will appear to be loaded at the desktop, although Speedshift will cut voltage below the monitoring levels if no work is being done. You can observe this through core temps. Frequency will be up. Vcore is up. But my core temps are still 38C? That is what Speedshift does. There is no downside and you should leave it enabled. Speedstep (the old tech) will be automatically disabled when speed shift is running. That is why you leave it on auto. Again, it is normal to see cycling Vcore and frequencies on the desktop with Lake processors. However, the actual power consumption is more stable and a sign things are not as turbulent as they seem.

 

I have not had a chance to really test the new Asus Z390 "AI" predictions. I would really need to run it with multiple CPUs to verify. However, what it is telling for my 8700K I have had for 18 months is accurate. Once again, the means and method it uses to arrive at these values are unknown. I know my CPU to be AVX stable at 5.0/1.30v. It is telling me it is for 1.2985v non-avx and 1.3025 AVX. Am I am that good? Is it? Or is it just spitting my own settings back to me? I don't know. I do think you can use the predictions as starting points. Make sure you make a system restore point or other C:drive image capture before really getting into your overclocking. While OS corruption is more common with memory tweaking, it is possible on any BSOD. You likely will have one at some point. In the better safe than sorry category.

 

**I had to go back and look at this cooler score thing. For the record, my dual 360mm custom loop comes in at 189 pts. That seems like an improvement over your 125, until you go to the user settings and see the range. 125 is the minimum. 2000 is the maximum. Gee, I guess I needed to go quadruple 560mm rads. Anyway, garbage metric. All it can do it assess CPU temperature in relation to TJ Max. I suspect my score is low, because since I updated the BIOS 36 hours ago all I have done is run RAM and CPU stress tests making sure my 5.0 cache OC was stable. Average temps are high. We'll see what it looks like next week after 40 hours of MS Word and Excel. Don't get caught up in this. It does not matter if your cooler is 120mm or 10m, the CPU voltage is the limiting factor. I cannot run 1.60v Vcore on a 120mm cooler or 30ft wall panel of fins. It will be too hot at the pins. Whatever the "AI" is doing, it is assessing the the amount of time you spend within X% of TJ Max. I suspect lowering your clocks and vcore would boost it.

Edited July 24, 2019 by c-attack

[Knight_]

Ugh.. Always copy your long posts before clicking submit. There is often a white screen and a lost post waiting for you on the other side.

 

Take 2.

 

 

1) Any kind of "cooler score" from the BIOS is garbage. It has no means to access or measure your coolant temperature differential. Presumably it was designed for air coolers and is a simple CPU temp rating, but then it can't really measure your software usage from there either. Do I get a good score if I sit there and do nothing? Do I get a bad score if I run Prime 95 a lot? Who knows. No matter what, this is a worthless metric intended for people who have no understanding of how their computer work and need some value as a reassurance.

 

I don't use and am not familiar with RPCS3. However, you can use whatever AVX offset you want when finished, but if you are trying to test 5.0@1.32v (or whatever), nearly every stress test program uses AXV and will downclock you to the AVX level, thus you are testing 4.8@1.32 with AVX, not 5.0. While theoretically possible, it is unlikely you would be 5.0/1.xx vols stable, and then shaky at 4.8 AVX. A 200MHz offset should cover things.

 

2) VCCIO and VCCSA - I have no idea why Asus is still piling on the IO and System Agent voltages. Unlike most other values, more voltage is not "hotter but more stable". It can be be hotter and less stable. Try 1.175v for both. That is what I use for 3600C16 (4x8GB). I use 1.20 for both at 4000C17. Most likely you are in that range. This platform is not so picky that you will have problems shifted one way or the other and anything in that area is likely fine. Like others settings, this one is individual CPU specific. Warning signs are a frozen screen, black screen, or loss of input.

 

3) Sync All vs Per Core - Sync all cores sets the turbo multiplier for all 8 cores to the specified level and it will run this when the conditions permit (sufficient load, under power/temp limits). This is the same as what MCE does, except you get to set the voltage. Overclocking with auto voltage (MCE) is what gets you into the worry zone immediately. All core lets you set the voltage, as long as you don't use auto. Per Core synchronization is similar to the Intel default specifications. When 1 core is loaded, you can have 50x100. When 2 cores are loaded, 48x100. 3=47, 4=46... 8 cores = 3.6GHz. The per core setting will let you set each core count to whatever you want (6 cores for 5.0, 8 for 4.8), but the problem is you are always going to have 8 cores active The CPU constantly shifts the load between cores and this mechanic does not keep in step with those parameters. You will never ever see 5.0, except maybe opening a folder on the desktop. You will run 4.8 under load all the time. All this does is add uncertainly to your results.

 

3.6 is the standard frequency for the 9900K and probably whatever you were doing was light enough not to invoke the the turbo boost. It is also possible to unnecessarily create this condition by fiddling with the c-states/turbo/Speedstep settings in the BIOS. You want Speedshift On, Speedstep Auto, C-states Auto (to start). That should disable C-states whenever there is any load. It will partially relax them when at the desktop. If you need the lowest possible desktop power levels, you need to manually enable C-states and enable most of them. That may have some unintended effects on gaming and full near full load operations. If for whatever reason you need 100% uptime from the processor, use the Windows power plan to set it to Maximum Performance (from Balanced). This should be a pre-existing setting in the power plan. This will keep the cores up at the max turbo frequency when any load is present. It will appear to be loaded at the desktop, although Speedshift will cut voltage below the monitoring levels if no work is being done. You can observe this through core temps. Frequency will be up. Vcore is up. But my core temps are still 38C? That is what Speedshift does. There is no downside and you should leave it enabled. Speedstep (the old tech) will be automatically disabled when speed shift is running. That is why you leave it on auto. Again, it is normal to see cycling Vcore and frequencies on the desktop with Lake processors. However, the actual power consumption is more stable and a sign things are not as turbulent as they seem.

 

I have not had a chance to really test the new Asus Z390 "AI" predictions. I would really need to run it with multiple CPUs to verify. However, what it is telling for my 8700K I have had for 18 months is accurate. Once again, the means and method it uses to arrive at these values are unknown. I know my CPU to be AVX stable at 5.0/1.30v. It is telling me it is for 1.2985v non-avx and 1.3025 AVX. Am I am that good? Is it? Or is it just spitting my own settings back to me? I don't know. I do think you can use the predictions as starting points. Make sure you make a system restore point or other C:drive image capture before really getting into your overclocking. While OS corruption is more common with memory tweaking, it is possible on any BSOD. You likely will have one at some point. In the better safe than sorry category.

 

1) I've gotta say, I've heard good things about the predictions and namely the cooler ratings. It's an arbitrary rating system but at least consistent. It actually does use data from your temps, load and power consumption. I think the general rule is if your CPU is under high load, has a lot of power flowing through it and it's relatively stable but low temperature wise, you get a good 'score'. You can train it and manually set how optimistic it is about its predictions, etc. I don't think it's completely right but I think it's good to keep in mind.

 

2) I thought the VCCIO would be too high. I'm wary of dropping it straight down to 1.175V, but I'll try 1.2V now. It seems like most people have it between 1.15V - 1.35V (the latter for 4200Mhz+) so I don't want to accidentally undervolt it if my RAM happens to be on the more demanding side (despite its specifications). Could you be more specific about the warning signs? I wouldn't want no input or a black screen before I can enter the BIOS, that sounds irrecoverable.

 

3) Understood. For now, I think I'm going to leave them all at 4.7GHz using "Sync All Cores" and just have those conditions dictate when that speed is reached or not. Seems the safest and most stable whilst I'm still getting the grips.

 

4) SpeedStep and C-States are on "Auto" by default. SpeedShift isn't an option so I'm assuming it's just always on. I'll leave them as they are.

 

5) Yeah, people say it's a good place to start. Not always perfect, but it does a pretty good job. I watched a few videos on it and it's said to use the SVID table built into the processor and then mixes that with your cooler rating. You can artificially change its predictions if you manually set your cooler rating if you are anticipating an upgrade or think its values are wrong.

 

I'll make a system restore now, change the AVX offset, reduce the VCCIO / VCCSA voltage to 1.2V and set all cores to a synchronised "47" and let you know how it goes. Thanks again.

[c-attack]

2) I thought the VCCIO would be too high. I'm wary of dropping it straight down to 1.175V, but I'll try 1.2V now. It seems like most people have it between 1.15V - 1.35V (the latter for 4200Mhz+) so I don't want to accidentally undervolt it if my RAM happens to be on the more demanding side (despite its specifications). Could you be more specific about the warning signs? I wouldn't want no input or a black screen before I can enter the BIOS, that sounds irrecoverable.

 

2) The freeze would be on the desktop. Just hitting the reset button is enough to send things to the BIOS. It is not a boot loop or hard lockout. I'd have to go back and find the original post, but Asus through their guides put the target IO/SA at 1.15-1.25 for most people. I am sure those on the bleeding edge of stability have to find their own marks.

 

4) Possible SpeedShift is in a different location on 9000-series processors vs 8000. I will need to dig around.

[Knight_]

I'm back. I followed your original list along with the suggestions on the VCCIO/SA voltages, ran Cinebench R15 and then ran OCCCT for 5 minutes for a quick stress test.

 

Everything's fine so far. Cinebench just brushed 90C at its highest but was mostly in the 80's. It also didn't instantly shoot up to 100C like it did originally back when I made this post. I'm fine with that, it's something I'd expect and I believe is about normal. Overall score wasn't amazing at 1935 but I guess that's because I'm not hitting 5Ghz yet.

 

OCCCT was absolutely fine. It was 68-73C and stable around there. It didn't increase. The clock speed was hitting 4500Mhz but I'm assuming that's because it's using AVX instructions? I set the negative AVX offset to 2, so that would make complete sense. CPU Voltage dropped to about 1.6V / 1.5V here too. Again, I'm assuming completely normal.

 

Now I feel more comfortable, I'm probably going to begin playing around with it to get higher clocks. I think my issue with temperatures is largely solved and, where it could be better, I think my new thermal paste will improve it.

 

Just one more question:

 

CPU-Z reports 1.181V vcore. This being 0.02V lower than the value I set is called "vdroop", right? Slight follow-up; If it's having no issue hitting 4700Mhz at this voltage, the silicon lottery is looking to be in my favour, right? Haha.

 

I know I've said it loads, but thanks again. When you start off knowing absolutely nothing about overclocking and your £400 purchase starts burning in front of you, it's really frustrating and upsetting. I'm glad to have it working and to now have the knowledge on how to deal with something like this in the future.

 

Not sure if this forum has it but I'd say the post can be labelled "solved" now.

[c-attack]

Yes, you will see the Vcore just under the set value at some points and likely just a bit over when running Cinebench or another max load test. At LLC 7 I see about a 0.012 to 0.015v add on, so your value makes sense for LLC6.

 

Silicon lottery? You won’t know until you test, but the results so far are certainly good, not bad. The funny thing about approaching the limit is how drastic the changes are in temperature or efficiency in that last tick or two. You think you are ahead of the curve and then the limit jumps right in your face. You go no further. On the other hand, there are also CPUs that are piggy on voltage at lower levels, so you think it will be a low limit. Then they just keep going. Yeah, it needs 1.40v, but it is 5.3 GHz stable. Of course, we always want both — high limit and voltage efficient.

 

You may be interested in this. I didn’t remember it also had R15 data or I would thrown it out right away. Keep in mind this is the top 10% of his binned CPUs, but look how much variation there is in the R15 numbers — 20C at 5.2/5.3 and this is among top binned samples.

 

https://www.tomshardware.com/reviews/intel-9900kf-vs-9900k-overclock-comparison,6038.html

 

 

Solved. Lol. They don’t close threads here unless people are nasty. You’ll get an email 5 years from now that someone has responded to your thread asking if their “Salton Sea 22290K should be under 90C like this person or if they should bought the Corsair H480 Titanium Pro cooler.”