The Experience of Undervolting a 12th Gen CPU

As I began to experiment with undervolting my 12th gen CPU, I was excited to see how it would perform. The process started by testing the CPU's performance and temperature using various tools such as Time Spy Extreme, which measured in milliseconds. To my surprise, the 12th gen CPU performed remarkably well without any overclocking, rivaling the performance of older CPUs like the 9980xe on LN2.

One of the key aspects I wanted to explore was how the CPU would perform under different voltages. To do this, I used a tool called XTU, which allowed me to dynamically adjust the voltage and monitor the system's response. The goal was to find an optimal balance between performance and temperature, while also identifying any potential bottlenecks.

Initially, I started with a relatively conservative approach, keeping the voltage at stock settings. However, as I continued to experiment, I found that even small increments of voltage could have significant effects on the system's behavior. For example, increasing the voltage by just 100 megahertz seemed to have a noticeable impact on the system's performance and temperature.

To further test my hypothesis, I decided to push the CPU to its limits by adjusting the voltage and monitoring the system's response. In one instance, I bumped up all the clocks by 100 megahertz to see if that would reveal any potential issues. To my surprise, even at this increased frequency, the system remained stable.

One of the most interesting observations I made during this process was how the BIOS responded to different voltage settings. It seemed that the BIOS was able to recognize and adapt to the new voltage levels, adjusting its behavior accordingly. This led me to experiment with different profiles and voltages, ultimately allowing me to find an optimal setting for my system.

Throughout this experience, I found it fascinating how a small change in voltage could have such a significant impact on the system's performance and temperature. It also highlighted the importance of careful tuning and experimentation when trying to achieve optimal results.

In terms of practical applications, I believe that undervolting can be a useful technique for anyone looking to improve their system's performance and efficiency. Whether using air cooling or a small AIO, making these adjustments can lead to significant improvements in temperature and overall system stability.

I hope this experience has been informative and helpful to those who are interested in exploring the world of CPU tuning. If you're new to the topic, I encourage you to continue experimenting with different techniques and settings until you find what works best for your system. And if you're already familiar with undervolting, I'd love to hear about your own experiences and tips in the comments below.

In conclusion, my experience with undervolting a 12th gen CPU was both fascinating and eye-opening. By carefully experimenting with different voltages and settings, I was able to unlock significant performance gains and improvements in temperature control. While this technique is not without its risks, I firmly believe that it can be a powerful tool for anyone looking to take their system to the next level.

The Final Experiment: Crashing and Recovering

In one final experiment, I decided to push my 12th gen CPU to its limits by increasing the voltage even further. To do this, I bumped up all the clocks by another 100 megahertz, taking the total increase to 200 megahertz. This resulted in a significant increase in temperature, but I was eager to see how the system would respond.

To my surprise, when I restarted the program and checked the system's status, it indicated that there had been an application exit error due to platform instability. While this might seem alarming, I knew that this was simply a result of pushing the system too hard.

In the end, I decided to revert back to my previous settings and save the profile for future reference. This experience served as a reminder that undervolting can be both exhilarating and intimidating, and that it's essential to approach these experiments with caution and careful consideration.

The Importance of XTU

Throughout this experiment, I found myself relying heavily on the XTU tool to monitor and adjust my system's settings. The utility proved invaluable in helping me fine-tune my settings and identify any potential issues before they became major problems.

XTU allowed me to dynamically adjust the voltage and monitor the system's response in real-time, providing valuable insights into how different settings would affect the system's performance and temperature. By using XTU, I was able to avoid crashes and instability, ensuring that I could experiment safely and confidently.

In terms of practical applications, I highly recommend using tools like XTU when exploring undervolting techniques. These utilities provide a crucial safety net, allowing you to monitor your system's behavior and adjust settings accordingly.

"WEBVTTKind: captionsLanguage: enso if you guys watch my video about the origin pc millennium uh i talked about the fact that 12th gen no matter what you install it in whether or not it's fully water cooled custom loop like this or an aio gets extremely hot well i'm using this as my desktop here now at work instead of using the small one so i figured i'd take this opportunity to kind of do some undervolting with 12th gen because i feel like that's something that a lot of people take advantage of with intel's latest generation and its voltage and heat and power draw so i want to see because this is the first full water cooled loop that i have my hands on for 12th gen how much will undervolting actually help in terms of trying to keep these temperatures under control bring your setup to the next level with cablemod's all-new custom coiled keyboard cables available in a variety of colors and connector types utilizing best-in-class connectors the keyboard cables give your setup a look and feel it deserves to see the complete lineup of custom keyboard cables available from cablemod click the link in the description below so this is the first msi board i've really kind of gone in and played around with the bios on 12th gen so i'm not as familiar with this bios layout so i'll be kind of figuring out what features it has available to it we've been using asus for a while now in our builds so i'm very familiar with that bios we did update the bios though because our build was all the way back to november which is not that far after 12th gen launch so i updated for maximum memory compatibility and all of that so i haven't actually even turned on xmp profile which i went ahead and did right there and let's just go ahead and do a boot restart with xmp to see if that will even start now one of the things with 12th gen is because they've got e core and p core we've noticed not just a massive amount of power draw um and heat generation from this chip i mean 250 plus watts under load in cinebench is a lot of heat for a block to manage it's not the loop there's more than enough thermal capacity in the two 360s to keep 250 watts cool it's just when you have 250 watts really like condensed into a a really kind of like small area which is that of the cpu die um it generates a lot of heat now as we move forward someone might say well j gpus like 3080 ti here is much smaller and much denser than the cpu yet it gets into like the low 50s under load wise the cpu gets so much harder well that's direct die contact we have the block touching the die directly whereas the cpu has to go through a heat spreader and then into the block as also if you compare the surface area of a gpu block versus out of a cpu block they're much more thermal capacity in a gpu block than a seat blue block so it might get to the point where we need bigger thicker cpu blocks to keep them cool just by having more thermal capacity but it's the transfer rate between the dye and the cpu block that is being interfered with with the inner integrated heat spreader the ihs so i just want to give a little background there as to why that is if there's one thing i understand very well it's cooling which is why now i have to get in here and do voltage mods to try and keep the cooling a little bit more under control so let's just take a look at the settings here real quick and this isn't a tutorial i'm just kind of showing you the methodology i go through so we're going to go to oc all the voltage stuff at least for msi is under oc we're in expert mode which gives us access to everything we're not touching any of the core frequency stuff we're not doing any of the like core enhancement stuff like you would see in asus this is just out of the box settings for msi so if i go all the way down here to voltage right here core voltage mode is just set to auto this one that gives us kind of a few options here we have adaptive mode which is basically what auto is going to be which means it will determine under the load conditions and the frequency at which the cpu is going to how much voltage does it think it needs to give to the cpu spoiler alert it will always err on the side of more voltage because that's more voltage means more stability the problem is that also means more heat and also means more um you know power draw out of the cpu which kind of exacerbates the problem starts to get too hot then it starts to drop the clocks to bring it down because the voltage is just pumped into the cpu to make sure it's as stable as possible it's kind of like in a car most cars err on the side of running rich rather than lean to account to calculate for all of the different altitudes at which cars operate some like if you're in denver your car runs very different than it would here in california when we're at nearly sea level versus a mile in the air which is what denver is so with that said i need to get a baseline so i'm just going to go ahead and boot the system here with stock settings with my xmp enabled and we need to get some baselines now we've already done that with our uh overview of the system it was reaching upper 80s and i think the hottest i saw it get while letting it just loop and loop the loop was like 91c on the cpu 91c is still what 14c below where we would start the throttle which is 105c i with the with two 360s and a big old loop like this and tons of fans and airflow and a 5000t that is still much warmer than i would ever expect it to get to and again that has everything to do with um the the voltage logic built into the motherboard so as always i'm using cinebench r23 for my temperature testing because i can let that loop this is the toughest scenario which a system of mine will ever like none of my systems will ever get hotter than doing cinebench r23 i've got harvard monitor open here to volt to keep an eye on different voltages and temperatures so vid is the voltage to each core directly you've got cpu core temperature which i don't care about i care about the core temps which are right here and as you can see they're in the 20s and then we've got cpu here under voltages which is going to be basically like the v core voltage it's just when they name it something different so cpu not vcore so as you can see right now under idle conditions it's about 1.272 and then our voltages to our cores are bouncing around between like 1.25 and 1.33 it appears um and then i'm just going to start this test and see where we sort of land the initial spiked temperatures will give me an indication of what things are looking like here so we are in the upper 60s low 70s on the e cores we are already at nearly 90 c on the p course 85 76 on core one and it's 1.315 volts to the course to do that our volt our core speeds right now are 4.9 all core p core and 3.7 all core ecore these temps are not bad but i fully expect that they could be better since we're spiking up to the high 80s that tells me over time with full saturation of this loop we would probably end up hitting 9192c on the core p cores package temp there's a lot of different areas where it will take temperature readings to determine what it's going to do you've got the package temperature which is just the socket temp itself the entire array if you will individual e core p core temps as well as system agent like all voltages all these things come into play 1.336 in my opinion based on stock speeds and what i've already done on like my um my small build when i was under volting that is way more than i think we need i'm also not planning on doing any real overclocking with this i might push it to like a 5-0 all core maybe even a 5-1 all core because i found that i can undervolt it and still overclock it and get fairly decent results out of it so yeah 1.33 to 1.35 is where it tends to be going package attempts at 88 core temps the highest core is hitting 89. i'm not gonna change clocks first i'm gonna actually change voltage and i could do this with uh intel xtu in fact i'm gonna do that because intel xt or extreme tuning utility um will allow me to do this in the operating system where i can make quick changes and run the tests over and over and then when i sort of get that dialed in if i want i can go into the bios on a bios level and have entire dial or like enter in these figures and these numbers to where i don't have to worry about xtu running to make all this work another benefit to using xtu is the fact that you could leave all your bios settings stock that way you know it'll boot and then when xtu runs which would be set up to automatically start with the system it can then imply the under voltage and any clock adjustments that you've made memory speed adjustments that way you know you get a safe boot and then let xtu handle all of it xtu is very similar to how like ryzen master works um only it's a lot more like utilitarian whereas ryzen master is a very gooey kind of an interface xt is just like here's all the settings and all the dials good luck in fact another benefit to xtu is the fact that it's smart enough to know when it crashes for instance when you shut down the system you shut down xtu shuts down normally the log will basically say that it has a successful shutdown if you lock up under testing conditions and you have to force a restart when xtu starts back up and sees that it did not have a successful shutdown in the log it will not apply your latest overclocks it'll tell you what the defaults are and say do you want to load these defaults it looks like your system was not stable which makes it a lot easier process to figure out your overclocks and or your undervolts without doing it in the bios level and having to deal with a clear cmos every single time and then put all your memory settings and your fan speeds and your boot priorities and all that stuff back so xtu is actually a very powerful program uh when applied you know correctly it's got a built-in speed optimizer which i don't want to do because i want to do advanced tuning which is going to be under volting here so core voltage offset we can do a negative offset to our voltage or we can do core voltage here as just a fixed value now i want to do a fixed value because i don't want the offset is only going to apply to whatever its dynamic adjustment is i don't want that i just want to lock the voltage where i can see in static voltage situation with a very hard test where does it crash now i know that 1.33 is running by default i'm going to go ahead and just do 1.3 and then i've applied that i'm also going to max out our turbo power timer window i want everything as static as possible so already we've come down a couple of degrees it's just this reminds me of like the 4770k where you'll have certain cores that are just fine like we've got core 2 right here core 1 sitting at 74c and then we've got a core right next to it doing 85. like this is why d-litting became a thing because of the fact that there was such uneven thermal compound between the ihs and the dye or the ihs itself sometimes was kind of wavy or concaved or bowed which was causing certain cores to run hotter than others especially if you have a hot core right next to a cool one this is the frustrating thing here is to see like like i said core one is at 77 74 76 right next to a core doing 86 10c difference now 4770k we saw like 14c delta between the hottest and the coldest but those were usually like on two different sides of the die not right next to each other so our cpu right now is showing 1.28 volts and then if i look at vid it's still going 1.35 per core so you can see how just package core or cpu core voltage or vcore and vid are two very different numbers when it comes to voltages temperature wise though again we've hit 90 c on the package we've hit 9 dc on core 2 which is our hottest and core 7. so i'm going to drop the voltage even farther to 1.25 oh here we go let's try this again one does it run and not crash by dropping the voltage that low okay so vid is back up to 1.35 by setting it to static and we're at 96 c on the core so clearly that setting didn't help at all right there what we did find is by setting up the static and then having it like i had to go way farther south of where i wanted it to be to actually have it land where i wanted to be so let's go to 1.2 and let's see if 1.2 puts us where um we expected the voltage to land which is right now 1.25 it's exactly 0.1 volts higher than where we're setting it it's it's weird the way it does that but now we can at least sort of dial in based on that i kind of go in like 50 millivolt adjustments typically until i'm doing fine tuning like if it were to crash now then i would know i can land somewhere between 1.2 and 1.165 and then go from there so there's our vid 1.254 look at our package 78 80. our ecores are back down into the 60s it's really getting the wattage down is what's helping the temperatures and obviously the wattage is based on amps and volts right so it's working and the fact that it hasn't crashed is already a good sign so if that doesn't tell you already that we was doing 1.36 volts on vid and we're at 1.26 volts on vid and it's still stable like it didn't crash immediately that's already a good indicator so i'm going to go 1.15 interestingly enough our overclocking therma thermometer it was disabled by default which is interesting because this is a this is like a prime uh ever since like 10th gen it up this has been a very big deal we re-enabled it here i want to see what the difference is so it's one 1.40 on the vid 77 on the package 78 the core one which is like our amazing core is at 68 degrees right now we can get them all that cold i don't want to deliver one of these cpus i have like four of them i could have a sacrificial but folks have already been deleting these for for obvious reasons and it's just why why can't intel do this itself so like our max temp right now is 80 on the package our hottest core hit 80. and sure i'm not running this for like 30 minutes that'll be the last thing i do is okay i think i'm dialed in let's let it loop and see if it crashes over time if it crashes over time then i'd have to slowly add back like 10 millivolts if the fluid rises 1c the temperatures will rise 1c they rise 1 to 1 ratio so it's one of those things where we have to let the loop get to full warmth to determine where the cpu's temperature is going to land at as well but right now this is looking very promising 78 on the package 78 on the cores 76 see asic quality is going to determine how much voltage it needs to maintain its clocks right if we have a very leaky cpu it'll tend to take a lot more voltage to maintain the the clock that's expected to run if we have a very tight cpu i i refer to as leaky voltage or tight in terms of of crosstalk and such and asic quality this cpu is already starting to look like it's one that's going to benefit very well from low voltage and not need to be you know stupid ridiculous so i'm gonna go 1.125 i've now reduced it 25 millivolts not 50 because i know i can't keep going 50. it's it's just not going to happen so at 1.125 the cpu voltage is landing at 1.188 and then our vid is 1.220 ish it every now and then spike though as you can see 1.339 and that's right when the test starts and this is going to have a lot to do with our droop settings as well which i haven't touched they're oem droop droop says how much will the voltage sag under load versus it applying more offset to keep a ma like maintain a level voltage because our vid is going higher than my target i'm not messing with volt with droop i'm leaving where it's at but that would kind of be like the next step is okay you sort of dialed in now let's play with droop and see if house level we can get the voltage to land under load because what will happen is this the test will start and then the voltage will spike real high and then come back down um the same thing can happen when we stop the test is it could shoot down which can also lead to a crash situation it'll shoot down and then come back up it's those spikes and those droops or drops that determine your stability at the start and stop of a test right now these temps 76 on the package our hottest core is hitting 78 and core 1 which is our amazing core is at 68 right now it was at now 66 67 i'm kind of surprised that i can still lower this voltage 1.1 see i would consider this a golden sample right now if i can maintain stock clocks by dropping the voltage now almost 225 millivolts that's a lot that's a huge percentage drop in terms of voltage which as you can see is having an immediate indicate like impact on our temperatures i've said this since the moment we first started touching 12th gen how much voltage are they adding that you just don't need motherboard manufacturers say the voltage is only going to what the cpu is commanding but i can take the same cpu and put it in different motherboards and have completely different voltages so clearly the motherboard also has a say in the voltages 73 on the package now i've been running this test over and over and over obviously not letting the loop cool down so by stopping adjusting and starting again it's a lot more towards equilibrium now our vid is finally under 1.2 it's like 1.9 1.1 or 1.9 1.19 1.189 this is absolutely beautiful and my fans are also not running as loud because the cpu is not as hot which could also have an indication of the fact that hey if i left the fan static which would be the next logical thing is you lock everything down so you control the variables as much as possible if i left my fan static and then tested temperatures that would give me an indication of what the effects of the voltages are doing but if the fans are slowing down because the cpu is colder that could also lead to a warmer temp over time because you have the fan curve adjusting so i'm not locking the fans though because it's water cooled it has a lot less of an effect than say an air cooled system would but that's something you'd want to do as well if you're truly trying to dial in voltages and tests and temperatures and remove as many variables as possible you would absolutely want to lock your fans off so i'm going to go 1.08 volts i'm only dropping at 20 millivolts now not even 25. so one of the things i'm going to want to do here too is once i find out where my voltage is stable i'm going to go back to default adaptive with an offset a negative offset the reason for this right now is this is working fine at idle and then hitting it with a load when you start loading programs and stuff where the voltages or the the clock speed can suddenly spike in a single core or dual core type of scenario where only one or two cores are being hit this voltage might be too low for that which could then lead to instability so that's why i'll just find where the negative offset puts it to this under load and then under adaptive that will allow it to adjust its voltages as it needs to as programs kind of come on and off offline and are hitting the core in a very different way right now i'm dialing in stability in a very specific test but this is the load test that i'm dialing in for and then i'll adjust from there so core voltage right now is at 1.168 vid is at 1.175 or so it's funny because it does it when i when i start the test and i watch it load the voltage does hit 1.33 and then come down so that tells me even though i have a set of static on here it's still doing adaptive voltage adjustments which is probably why it's not crashing when we start the test it probably would have crashed if it wasn't doing that adaptive thing so i'm not sure what setting in this motherboard is causing that to still spike like it is um it could be honestly the fact that the motherboard itself still sees an adaptive voltage setting whereas xtu is telling it no let's go static it's almost like the motherboard handles it for a second and then xtu takes over which is almost just i'm fine with that to be honest but temperatures though right now the core the package got 78 for a second that was during that load like that peak voltage we're at 7 dc on the package our hottest core is 71 but they're averaging 60s now and it hasn't crashed yet that's the crazy part it's like to me this should have crashed a long time ago like let's go 1.065 i'll drop it at 15. 70s and 60s now on all the cores packages at 68 degrees 70 69. it is now if i stop the test there's one other there's one thing i want to do right now i don't want to keep dropping the voltages i honestly don't because i don't i think we're going to definitely get into low territory but let's just do random like loading of programs now so i'm going to i fully expect the voltage to like momentarily spike um yeah see vid is bouncing around like it should 1.2 1.3 so i should be able to start a program like let's say obs studio yeah that's fine let's let it let's just start recording and this is on gpu but look 1.35 oh yeah it's hitting cpu right now and i can tell because look we're having some course 1.404 1.3 1.4 so it is dynamically adjusting like it should okay let's try cpu test in 3dmark people would see the temps like in the most recent video we did about this and they'd say wow that loop really sucks and it's like no actually the loop is like doing its job because if you've ever tried to put even just like a basic aio on 12th gen and just leave it alone it's like it reminds me like the 9900k let me tell you what's nuts about this this 12th gen right now which is not overclocked this cpu tested in time spy extreme is measured in milliseconds it's as fast as like when we were doing the 9980xe on ln2 in the 7980xc on ln2 back when the 20 series first launched and we were doing like crazy overclocking with that it's nuts that this is that fast without overclocking it at all okay so there you go um i'll do one more thing before we get out of here which is i'm just gonna bump all the clocks up 100 megahertz to see if that's because if if it crashes under that condition at this voltage then i'll know like okay i was already getting towards the edge right there but i have a feeling it may not one active performance core i can make that 5.3 but i'm just bumping all this up one click so now we're looking for 5.0 all core and the nice thing is if this crashes and i restart the program it'll just tell me like hey do you want to go back to defaults and then i'll know i'll just put it back to static increase my turnover turbo timer and then just go back to the 1.17 is what the voltage was or maybe 1.16 ah see do you see how just 100 megahertz even idle was like nope so yeah you'll find sometimes too to get that extra 100 megahertz offset it might take a lot of voltage it's weird the way that works like you can just at stock voltage just add frequency add frequency add frequency and then you'll find this one number this one frequency where it's like nope no matter how much voltage you add no matter how much offset you try and do or cooling you throw at it well i shouldn't say cooling you throw it because ln2 obviously makes that number go up but it gets to the point hey look the bios actually recognized it um load default values and continue yeah so once you get it all dialed in where you're comfortable you can save the profile and xtu and then you can just have that lo that profile be what and it automatically loads once xtu loads with the system so there you go under volting still a necessity which is funny because this was such an amd thing for the longest time now it's the intel thing and the watts too dropped under 200 watts we went from over 244 to 250 watts under load to under 200 which is still a lot okay it's still a lot because amd's like 5950x still hits like 130 130 140. so anyway guys thanks for watching i hope this helps someone like oh my goodness my 12th gen's been super hot now i can maybe keep it cooler you can do this with any cooler any cooler in fact i recommend you do it if you're running air cooling or a small aio absolutely so all right guys thanks for watching what were your results like have you undervolted your 12th gen heck have you undervolted any of your cpus what was the result and how is it working out for you there's xtu loading let's see what it has to say i'm sure it's going to come back and be like bro you broke it yep see there it is right there the application exited unexpectedly this may be due to platform instability so anyway we can continue advanced tuning i agree because it's telling you you're going to break your stuff and as you can see i have to put all this back to like where i had it now which i'm going to go ahead and do again i'm just not going to take you guys along for the ride all right thanks for watching hope you enjoyed the video subscribe if you're new and share this video with someone that's complaining about their cpu temps it very well may help them we'll see in the next oneso if you guys watch my video about the origin pc millennium uh i talked about the fact that 12th gen no matter what you install it in whether or not it's fully water cooled custom loop like this or an aio gets extremely hot well i'm using this as my desktop here now at work instead of using the small one so i figured i'd take this opportunity to kind of do some undervolting with 12th gen because i feel like that's something that a lot of people take advantage of with intel's latest generation and its voltage and heat and power draw so i want to see because this is the first full water cooled loop that i have my hands on for 12th gen how much will undervolting actually help in terms of trying to keep these temperatures under control bring your setup to the next level with cablemod's all-new custom coiled keyboard cables available in a variety of colors and connector types utilizing best-in-class connectors the keyboard cables give your setup a look and feel it deserves to see the complete lineup of custom keyboard cables available from cablemod click the link in the description below so this is the first msi board i've really kind of gone in and played around with the bios on 12th gen so i'm not as familiar with this bios layout so i'll be kind of figuring out what features it has available to it we've been using asus for a while now in our builds so i'm very familiar with that bios we did update the bios though because our build was all the way back to november which is not that far after 12th gen launch so i updated for maximum memory compatibility and all of that so i haven't actually even turned on xmp profile which i went ahead and did right there and let's just go ahead and do a boot restart with xmp to see if that will even start now one of the things with 12th gen is because they've got e core and p core we've noticed not just a massive amount of power draw um and heat generation from this chip i mean 250 plus watts under load in cinebench is a lot of heat for a block to manage it's not the loop there's more than enough thermal capacity in the two 360s to keep 250 watts cool it's just when you have 250 watts really like condensed into a a really kind of like small area which is that of the cpu die um it generates a lot of heat now as we move forward someone might say well j gpus like 3080 ti here is much smaller and much denser than the cpu yet it gets into like the low 50s under load wise the cpu gets so much harder well that's direct die contact we have the block touching the die directly whereas the cpu has to go through a heat spreader and then into the block as also if you compare the surface area of a gpu block versus out of a cpu block they're much more thermal capacity in a gpu block than a seat blue block so it might get to the point where we need bigger thicker cpu blocks to keep them cool just by having more thermal capacity but it's the transfer rate between the dye and the cpu block that is being interfered with with the inner integrated heat spreader the ihs so i just want to give a little background there as to why that is if there's one thing i understand very well it's cooling which is why now i have to get in here and do voltage mods to try and keep the cooling a little bit more under control so let's just take a look at the settings here real quick and this isn't a tutorial i'm just kind of showing you the methodology i go through so we're going to go to oc all the voltage stuff at least for msi is under oc we're in expert mode which gives us access to everything we're not touching any of the core frequency stuff we're not doing any of the like core enhancement stuff like you would see in asus this is just out of the box settings for msi so if i go all the way down here to voltage right here core voltage mode is just set to auto this one that gives us kind of a few options here we have adaptive mode which is basically what auto is going to be which means it will determine under the load conditions and the frequency at which the cpu is going to how much voltage does it think it needs to give to the cpu spoiler alert it will always err on the side of more voltage because that's more voltage means more stability the problem is that also means more heat and also means more um you know power draw out of the cpu which kind of exacerbates the problem starts to get too hot then it starts to drop the clocks to bring it down because the voltage is just pumped into the cpu to make sure it's as stable as possible it's kind of like in a car most cars err on the side of running rich rather than lean to account to calculate for all of the different altitudes at which cars operate some like if you're in denver your car runs very different than it would here in california when we're at nearly sea level versus a mile in the air which is what denver is so with that said i need to get a baseline so i'm just going to go ahead and boot the system here with stock settings with my xmp enabled and we need to get some baselines now we've already done that with our uh overview of the system it was reaching upper 80s and i think the hottest i saw it get while letting it just loop and loop the loop was like 91c on the cpu 91c is still what 14c below where we would start the throttle which is 105c i with the with two 360s and a big old loop like this and tons of fans and airflow and a 5000t that is still much warmer than i would ever expect it to get to and again that has everything to do with um the the voltage logic built into the motherboard so as always i'm using cinebench r23 for my temperature testing because i can let that loop this is the toughest scenario which a system of mine will ever like none of my systems will ever get hotter than doing cinebench r23 i've got harvard monitor open here to volt to keep an eye on different voltages and temperatures so vid is the voltage to each core directly you've got cpu core temperature which i don't care about i care about the core temps which are right here and as you can see they're in the 20s and then we've got cpu here under voltages which is going to be basically like the v core voltage it's just when they name it something different so cpu not vcore so as you can see right now under idle conditions it's about 1.272 and then our voltages to our cores are bouncing around between like 1.25 and 1.33 it appears um and then i'm just going to start this test and see where we sort of land the initial spiked temperatures will give me an indication of what things are looking like here so we are in the upper 60s low 70s on the e cores we are already at nearly 90 c on the p course 85 76 on core one and it's 1.315 volts to the course to do that our volt our core speeds right now are 4.9 all core p core and 3.7 all core ecore these temps are not bad but i fully expect that they could be better since we're spiking up to the high 80s that tells me over time with full saturation of this loop we would probably end up hitting 9192c on the core p cores package temp there's a lot of different areas where it will take temperature readings to determine what it's going to do you've got the package temperature which is just the socket temp itself the entire array if you will individual e core p core temps as well as system agent like all voltages all these things come into play 1.336 in my opinion based on stock speeds and what i've already done on like my um my small build when i was under volting that is way more than i think we need i'm also not planning on doing any real overclocking with this i might push it to like a 5-0 all core maybe even a 5-1 all core because i found that i can undervolt it and still overclock it and get fairly decent results out of it so yeah 1.33 to 1.35 is where it tends to be going package attempts at 88 core temps the highest core is hitting 89. i'm not gonna change clocks first i'm gonna actually change voltage and i could do this with uh intel xtu in fact i'm gonna do that because intel xt or extreme tuning utility um will allow me to do this in the operating system where i can make quick changes and run the tests over and over and then when i sort of get that dialed in if i want i can go into the bios on a bios level and have entire dial or like enter in these figures and these numbers to where i don't have to worry about xtu running to make all this work another benefit to using xtu is the fact that you could leave all your bios settings stock that way you know it'll boot and then when xtu runs which would be set up to automatically start with the system it can then imply the under voltage and any clock adjustments that you've made memory speed adjustments that way you know you get a safe boot and then let xtu handle all of it xtu is very similar to how like ryzen master works um only it's a lot more like utilitarian whereas ryzen master is a very gooey kind of an interface xt is just like here's all the settings and all the dials good luck in fact another benefit to xtu is the fact that it's smart enough to know when it crashes for instance when you shut down the system you shut down xtu shuts down normally the log will basically say that it has a successful shutdown if you lock up under testing conditions and you have to force a restart when xtu starts back up and sees that it did not have a successful shutdown in the log it will not apply your latest overclocks it'll tell you what the defaults are and say do you want to load these defaults it looks like your system was not stable which makes it a lot easier process to figure out your overclocks and or your undervolts without doing it in the bios level and having to deal with a clear cmos every single time and then put all your memory settings and your fan speeds and your boot priorities and all that stuff back so xtu is actually a very powerful program uh when applied you know correctly it's got a built-in speed optimizer which i don't want to do because i want to do advanced tuning which is going to be under volting here so core voltage offset we can do a negative offset to our voltage or we can do core voltage here as just a fixed value now i want to do a fixed value because i don't want the offset is only going to apply to whatever its dynamic adjustment is i don't want that i just want to lock the voltage where i can see in static voltage situation with a very hard test where does it crash now i know that 1.33 is running by default i'm going to go ahead and just do 1.3 and then i've applied that i'm also going to max out our turbo power timer window i want everything as static as possible so already we've come down a couple of degrees it's just this reminds me of like the 4770k where you'll have certain cores that are just fine like we've got core 2 right here core 1 sitting at 74c and then we've got a core right next to it doing 85. like this is why d-litting became a thing because of the fact that there was such uneven thermal compound between the ihs and the dye or the ihs itself sometimes was kind of wavy or concaved or bowed which was causing certain cores to run hotter than others especially if you have a hot core right next to a cool one this is the frustrating thing here is to see like like i said core one is at 77 74 76 right next to a core doing 86 10c difference now 4770k we saw like 14c delta between the hottest and the coldest but those were usually like on two different sides of the die not right next to each other so our cpu right now is showing 1.28 volts and then if i look at vid it's still going 1.35 per core so you can see how just package core or cpu core voltage or vcore and vid are two very different numbers when it comes to voltages temperature wise though again we've hit 90 c on the package we've hit 9 dc on core 2 which is our hottest and core 7. so i'm going to drop the voltage even farther to 1.25 oh here we go let's try this again one does it run and not crash by dropping the voltage that low okay so vid is back up to 1.35 by setting it to static and we're at 96 c on the core so clearly that setting didn't help at all right there what we did find is by setting up the static and then having it like i had to go way farther south of where i wanted it to be to actually have it land where i wanted to be so let's go to 1.2 and let's see if 1.2 puts us where um we expected the voltage to land which is right now 1.25 it's exactly 0.1 volts higher than where we're setting it it's it's weird the way it does that but now we can at least sort of dial in based on that i kind of go in like 50 millivolt adjustments typically until i'm doing fine tuning like if it were to crash now then i would know i can land somewhere between 1.2 and 1.165 and then go from there so there's our vid 1.254 look at our package 78 80. our ecores are back down into the 60s it's really getting the wattage down is what's helping the temperatures and obviously the wattage is based on amps and volts right so it's working and the fact that it hasn't crashed is already a good sign so if that doesn't tell you already that we was doing 1.36 volts on vid and we're at 1.26 volts on vid and it's still stable like it didn't crash immediately that's already a good indicator so i'm going to go 1.15 interestingly enough our overclocking therma thermometer it was disabled by default which is interesting because this is a this is like a prime uh ever since like 10th gen it up this has been a very big deal we re-enabled it here i want to see what the difference is so it's one 1.40 on the vid 77 on the package 78 the core one which is like our amazing core is at 68 degrees right now we can get them all that cold i don't want to deliver one of these cpus i have like four of them i could have a sacrificial but folks have already been deleting these for for obvious reasons and it's just why why can't intel do this itself so like our max temp right now is 80 on the package our hottest core hit 80. and sure i'm not running this for like 30 minutes that'll be the last thing i do is okay i think i'm dialed in let's let it loop and see if it crashes over time if it crashes over time then i'd have to slowly add back like 10 millivolts if the fluid rises 1c the temperatures will rise 1c they rise 1 to 1 ratio so it's one of those things where we have to let the loop get to full warmth to determine where the cpu's temperature is going to land at as well but right now this is looking very promising 78 on the package 78 on the cores 76 see asic quality is going to determine how much voltage it needs to maintain its clocks right if we have a very leaky cpu it'll tend to take a lot more voltage to maintain the the clock that's expected to run if we have a very tight cpu i i refer to as leaky voltage or tight in terms of of crosstalk and such and asic quality this cpu is already starting to look like it's one that's going to benefit very well from low voltage and not need to be you know stupid ridiculous so i'm gonna go 1.125 i've now reduced it 25 millivolts not 50 because i know i can't keep going 50. it's it's just not going to happen so at 1.125 the cpu voltage is landing at 1.188 and then our vid is 1.220 ish it every now and then spike though as you can see 1.339 and that's right when the test starts and this is going to have a lot to do with our droop settings as well which i haven't touched they're oem droop droop says how much will the voltage sag under load versus it applying more offset to keep a ma like maintain a level voltage because our vid is going higher than my target i'm not messing with volt with droop i'm leaving where it's at but that would kind of be like the next step is okay you sort of dialed in now let's play with droop and see if house level we can get the voltage to land under load because what will happen is this the test will start and then the voltage will spike real high and then come back down um the same thing can happen when we stop the test is it could shoot down which can also lead to a crash situation it'll shoot down and then come back up it's those spikes and those droops or drops that determine your stability at the start and stop of a test right now these temps 76 on the package our hottest core is hitting 78 and core 1 which is our amazing core is at 68 right now it was at now 66 67 i'm kind of surprised that i can still lower this voltage 1.1 see i would consider this a golden sample right now if i can maintain stock clocks by dropping the voltage now almost 225 millivolts that's a lot that's a huge percentage drop in terms of voltage which as you can see is having an immediate indicate like impact on our temperatures i've said this since the moment we first started touching 12th gen how much voltage are they adding that you just don't need motherboard manufacturers say the voltage is only going to what the cpu is commanding but i can take the same cpu and put it in different motherboards and have completely different voltages so clearly the motherboard also has a say in the voltages 73 on the package now i've been running this test over and over and over obviously not letting the loop cool down so by stopping adjusting and starting again it's a lot more towards equilibrium now our vid is finally under 1.2 it's like 1.9 1.1 or 1.9 1.19 1.189 this is absolutely beautiful and my fans are also not running as loud because the cpu is not as hot which could also have an indication of the fact that hey if i left the fan static which would be the next logical thing is you lock everything down so you control the variables as much as possible if i left my fan static and then tested temperatures that would give me an indication of what the effects of the voltages are doing but if the fans are slowing down because the cpu is colder that could also lead to a warmer temp over time because you have the fan curve adjusting so i'm not locking the fans though because it's water cooled it has a lot less of an effect than say an air cooled system would but that's something you'd want to do as well if you're truly trying to dial in voltages and tests and temperatures and remove as many variables as possible you would absolutely want to lock your fans off so i'm going to go 1.08 volts i'm only dropping at 20 millivolts now not even 25. so one of the things i'm going to want to do here too is once i find out where my voltage is stable i'm going to go back to default adaptive with an offset a negative offset the reason for this right now is this is working fine at idle and then hitting it with a load when you start loading programs and stuff where the voltages or the the clock speed can suddenly spike in a single core or dual core type of scenario where only one or two cores are being hit this voltage might be too low for that which could then lead to instability so that's why i'll just find where the negative offset puts it to this under load and then under adaptive that will allow it to adjust its voltages as it needs to as programs kind of come on and off offline and are hitting the core in a very different way right now i'm dialing in stability in a very specific test but this is the load test that i'm dialing in for and then i'll adjust from there so core voltage right now is at 1.168 vid is at 1.175 or so it's funny because it does it when i when i start the test and i watch it load the voltage does hit 1.33 and then come down so that tells me even though i have a set of static on here it's still doing adaptive voltage adjustments which is probably why it's not crashing when we start the test it probably would have crashed if it wasn't doing that adaptive thing so i'm not sure what setting in this motherboard is causing that to still spike like it is um it could be honestly the fact that the motherboard itself still sees an adaptive voltage setting whereas xtu is telling it no let's go static it's almost like the motherboard handles it for a second and then xtu takes over which is almost just i'm fine with that to be honest but temperatures though right now the core the package got 78 for a second that was during that load like that peak voltage we're at 7 dc on the package our hottest core is 71 but they're averaging 60s now and it hasn't crashed yet that's the crazy part it's like to me this should have crashed a long time ago like let's go 1.065 i'll drop it at 15. 70s and 60s now on all the cores packages at 68 degrees 70 69. it is now if i stop the test there's one other there's one thing i want to do right now i don't want to keep dropping the voltages i honestly don't because i don't i think we're going to definitely get into low territory but let's just do random like loading of programs now so i'm going to i fully expect the voltage to like momentarily spike um yeah see vid is bouncing around like it should 1.2 1.3 so i should be able to start a program like let's say obs studio yeah that's fine let's let it let's just start recording and this is on gpu but look 1.35 oh yeah it's hitting cpu right now and i can tell because look we're having some course 1.404 1.3 1.4 so it is dynamically adjusting like it should okay let's try cpu test in 3dmark people would see the temps like in the most recent video we did about this and they'd say wow that loop really sucks and it's like no actually the loop is like doing its job because if you've ever tried to put even just like a basic aio on 12th gen and just leave it alone it's like it reminds me like the 9900k let me tell you what's nuts about this this 12th gen right now which is not overclocked this cpu tested in time spy extreme is measured in milliseconds it's as fast as like when we were doing the 9980xe on ln2 in the 7980xc on ln2 back when the 20 series first launched and we were doing like crazy overclocking with that it's nuts that this is that fast without overclocking it at all okay so there you go um i'll do one more thing before we get out of here which is i'm just gonna bump all the clocks up 100 megahertz to see if that's because if if it crashes under that condition at this voltage then i'll know like okay i was already getting towards the edge right there but i have a feeling it may not one active performance core i can make that 5.3 but i'm just bumping all this up one click so now we're looking for 5.0 all core and the nice thing is if this crashes and i restart the program it'll just tell me like hey do you want to go back to defaults and then i'll know i'll just put it back to static increase my turnover turbo timer and then just go back to the 1.17 is what the voltage was or maybe 1.16 ah see do you see how just 100 megahertz even idle was like nope so yeah you'll find sometimes too to get that extra 100 megahertz offset it might take a lot of voltage it's weird the way that works like you can just at stock voltage just add frequency add frequency add frequency and then you'll find this one number this one frequency where it's like nope no matter how much voltage you add no matter how much offset you try and do or cooling you throw at it well i shouldn't say cooling you throw it because ln2 obviously makes that number go up but it gets to the point hey look the bios actually recognized it um load default values and continue yeah so once you get it all dialed in where you're comfortable you can save the profile and xtu and then you can just have that lo that profile be what and it automatically loads once xtu loads with the system so there you go under volting still a necessity which is funny because this was such an amd thing for the longest time now it's the intel thing and the watts too dropped under 200 watts we went from over 244 to 250 watts under load to under 200 which is still a lot okay it's still a lot because amd's like 5950x still hits like 130 130 140. so anyway guys thanks for watching i hope this helps someone like oh my goodness my 12th gen's been super hot now i can maybe keep it cooler you can do this with any cooler any cooler in fact i recommend you do it if you're running air cooling or a small aio absolutely so all right guys thanks for watching what were your results like have you undervolted your 12th gen heck have you undervolted any of your cpus what was the result and how is it working out for you there's xtu loading let's see what it has to say i'm sure it's going to come back and be like bro you broke it yep see there it is right there the application exited unexpectedly this may be due to platform instability so anyway we can continue advanced tuning i agree because it's telling you you're going to break your stuff and as you can see i have to put all this back to like where i had it now which i'm going to go ahead and do again i'm just not going to take you guys along for the ride all right thanks for watching hope you enjoyed the video subscribe if you're new and share this video with someone that's complaining about their cpu temps it very well may help them we'll see in the next one\n"