Really NEED Undervolt guide for 14700k+MSI z790 mobo

[AndyTheGreat]

Hi, community. Just bought 14700K for my MSI z790-P Wifi DDR5. It burns like hell. I get 99-101 in Cinebench 24 and 102 in AIDA, around 99 in Premiere Pro and After Effects. Did not even try other software since it's obvious my DeepCool LS720 (promised by Tom's Hardware to cope with 13900k's over 300 watts) cannot withstand the 14700k 278 without thermal throttling.
So I either have to get another AIO - and I have no idea which one because there are aio owners saying their particular aio beats the temps of 14900k getting not more than 87 in tetsts and there are ones saying even the best 420 mm aios cannot cope with their 13700k - or undervolt my CPU.
I followed the instructions from some post right here on the MSI forums (cannot find it) and lowered the long term and short term power to 253 watts and also set the Lite Load from the default Mode 9 to Mode 3 and the temps are now ten degrees less. Is it enough? Is there any other method or is this one correct even?
I'm completely dumb in undervolting - those are just numbers in BIOS for me, so I really need some sort of an MSI mobo guide to correct and safe undervolting - step by step with screensots or a video. Please, help.

 

[Arctucas]

Guys, nobody said anything about my bios settings I posted on the previous page. Are they ok or do I have to change anything

One thing I noticed was you do not have CPU Core Voltage setting or CPU Core Voltage Monitor? And I do not have CPU Core Voltage Offset Mode.

Must be differences in our motherboards.

 

[citay]

Guys, nobody said anything about my bios settings I posted on the previous page. Are they ok or do I have to change anything

In general they are ok, you are using negative offset undervolting instead of CPU Lite Load, it's a different approach but can lead to similar results in the end, so there's nothing to really criticize about it. Power limits roughly according to your cooling (not fully dialed in yet) with added 307A ICCMax, other settings seem reasonable too.

About your cooler, i wrote here before how i find the 300W claimed TDP dubious, then i suggested here to try with 220W power limits, which i still think is probably rather adequate for your situation. Also, i explained before, it's not about the last couple hundred points in CB. This kind of workload is not very common, and somewhere above 200W, at 250W at the very latest, the additional tiny performance gains are "junk performance" because the power draw still rises exponentially, while the performance gains stay miniscule, so it's not worth it to hunt for the last bit of points up there. If the voltage is somewhat optimized, if your cooling is protected, and if you are in the ballpark with the scores, then you can usually call it a day.

 

[AndyTheGreat]

In general they are ok, you are using negative offset undervolting instead of CPU Lite Load, it's a different approach but can lead to similar results in the end, so there's nothing to really criticize about it. Power limits roughly according to your cooling (not fully dialed in yet) with added 307A ICCMax, other settings seem reasonable too.

About your cooler, i wrote here before how i find the 300W claimed TDP dubious, then i suggested here to try with 220W power limits, which i still think is probably rather adequate for your situation. Also, i explained before, it's not about the last couple hundred points in CB. This kind of workload is not very common, and somewhere above 200W, at 250W at the very latest, the additional tiny performance gains are "junk performance" because the power draw still rises exponentially, while the performance gains stay miniscule, so it's not worth it to hunt for the last bit of points up there. If the voltage is somewhat optimized, if your cooling is protected, and if you are in the ballpark with the scores, then you can usually call it a day.

Thank you for clearing it out, so I must have set it all in the correct way. Or maybe if there's anything else we could optimize, just say a word. One thing I noticed though - despite the Cinebench being a totally synthetic workload, my videoediting workflow shows almost the same temps - I say 91 degrees today, though I didn't pay attention to the wattage and I'll check it next time. The main thing that matters to me now is temps below 85 and maximum possible performance I can get with it. I'm also thinking about getting a different AIO sometime in the future. But right now if I can achieve a lower temperature by just bios settings I'd be happy. By the way I installed the Fan control and it set the same fan speeds MSI Center had. I can hear it by their higher noise in comparison to my own settings you saw earlier in this discussion.

 

[AndyTheGreat]

One thing I noticed was you do not have CPU Core Voltage setting or CPU Core Voltage Monitor? And I do not have CPU Core Voltage Offset Mode.

View attachment 190029

Must be differences in our motherboards.

I think the core voltage disappears once you choose a negative offset by -CPU, but I definitely don't have the monitor. Try choosing Offset instead of offset+adaptive

 

[citay]

One thing I noticed though - despite the Cinebench being a totally synthetic workload, my videoediting workflow shows almost the same temps - I say 91 degrees today, though I didn't pay attention to the wattage and I'll check it next time.

No, i didn't say it's totally synthetic, i said it's not a very common workload in daily use for most people. Cinebench is realistic in that it is a fully multithreaded workload including AVX instructions, but without using any dirty tricks to drive up the power consumption artificially. So it's representative of a multithreading-optimized rendering/encoding/etc. workload. If those are something you do regularly, then Cinebench is relevant, because the other workloads will cause a similar load. I would only hunt for the last bit of performance if time is money though (meaning, if shaving off a few minutes from the job is valuable because you earn your money with that stuff). Otherwise, just make sure to stay below the 90s in temps, for example try 220W power limits, and you're ok. With 220W you already get most of the performance out of the CPU (the more, the lower you can undervolt while staying stable).

 

[AndyTheGreat]

@citay 220 watts, got it, thank you

 

[Arctucas]

I think the core voltage disappears once you choose a negative offset by -CPU, but I definitely don't have the monitor. Try choosing Offset instead of offset+adaptive

You are correct, I never have used just Offset, I use Adaptive plus Offset.

As you can see, when I choose offset and - by CPU, my VCore goes to 1.351V.

Also, CPU lite Load (which I did not screenshot) changes from my manual setting of 3 back to Auto.


So, by my manually setting VCore, I am able to shave off .03V. I could probably go lower with more tweaking of the LLC, but I am good with my settings as they are now.

 

[FlyingScot]

You are correct, I never have used just Offset, I use Adaptive plus Offset.

If you're going to be manually messing with voltage, Adaptive plus Offset is definitely the better choice over a plain Offset mode, and I would add IMO Fixed voltage, too. I think you are better of allowing some of the smarts built into the Intel+MSI BIOS to help you out as the CPU goes up and down the frequency curve from idle to full load. In other words, a 125mV offset is a huge percentage deduct at idle (e.g. 0.700 volts) vs. at full load Vcore (e.g. 1.35 volts).

 

[FlyingScot]

Don't know if today's Super I/O chips offer that possibility already, but you are right, where there's a will, there's a way. So there's no will, evidently.

So, this is a perfect illustration of the limitations of tying all the system fans to CPU temp. I just implemented a FPS cap on a AAA game (to smooth out the frame times) and all my component temps (chipset, memory, even VRAM) shot up. Why? The HWInfo logfile shows that avg CPU temps dropped dramatically and all the system fan rpms along with it. Some days I think I’m pushing on a string...

 

[Arctucas]

If you're going to be manually messing with voltage, Adaptive plus Offset is definitely the better choice over a plain Offset mode, and I would add IMO Fixed voltage, too. I think you are better of allowing some of the smarts built into the Intel+MSI BIOS to help you out as the CPU goes up and down the frequency curve from idle to full load. In other words, a 125mV offset is a huge percentage deduct at idle (e.g. 0.700 volts) vs. at full load Vcore (e.g. 1.35 volts).

Fixed voltage is a second choice, Adapative + Offset seems to be the best for my needs.

The negative .120V appears to be (for my system) compensation for the amount of overvoltage the BIOS is applying by default.

I have no issues running any games or benchmarks with VCore at 1.32V (set at 1.33 in BIOS), down from the 1.35V the BIOS wants to apply.

 

[FlyingScot]

Fixed voltage is a second choice, Adapative + Offset seems to be the best for my needs.

The negative .120V appears to be (for my system) compensation for the amount of overvoltage the BIOS is applying by default.

I have no issues running any games or benchmarks with VCore at 1.32V (set at 1.33 in BIOS), down from the 1.35V the BIOS wants to apply.

I’m not surprised at your results - you are still in the ballpark of conservative.
Although, I don‘t know if I would ever have thought to combined a manual voltage with an offset on top. That’s a novel approach. Am I missing something, or could you achieve the same result by adding to the negative offset (if you had a good feeling for the auto voltage) and eliminate the manual voltage altogether?

 

[AndyTheGreat]

Nah... Something is wrong with my AIO, see no other variants... And this is just video editing, not even video rendering

 

[FlyingScot]

Voltage looks like the main culprit, but I think we should put that to the side for a moment. What we need to do is make you feel better about your AIO or reach a point where we can confirm your suspicions.
Here’s what I would do: I would start with the basics. Lets take your AIO first. How are your idle temps, are they stable over 30mins or do they steadily creep up? They look like they are in the mid 30’s, so that looks normal to me. What is your room temp in C?
Next step: Do you have a AAA game that you can play? Put that 4080 to work and use HWINFO in the background to monitor temps, and other things. I see that you now have the Average column exposed. Good. Reset all the counters before you start playing the game. Play for 30mins to make sure your water temp has reached a steady state. Exit the game and take a snapshot of all the HWINFO sensors like you’re doing. Let us take a look at those results and perhaps I can say one way or the other whether the AIO cooling performance is as expected. Your case fans are spinning at relatively high RPM. Are they all 140s? Also, how is your AIO installed? Top mounted or front mounted? If it is front mounted, do you have at least a couple of inches of radiator above the CPU? If you have air in the system, it should end up in the highest point. You want the CPU pump to be lower than that. Also, did you check to make sure one pipe is hotter than the other yet? If not, you need to tick that off the list.

 

[AndyTheGreat]

The idle temps are just what they should be (considering my ambient temp) - 36-39 degrees. Normally it settles somewhere around 36. I don't have a thermometer in my place but it feels like 30-32 now with 35-36 outside. I use a fan all the time and am still sweating - this summer turned to be rather hot. Nonetheless I've never seen the temp go below 34 even in winter and those periods they turn off the central heating for maintenance when it's still cold.
Don't have an AAA game but gonna download Cyberpunk tomorrow. Well I finished Horizon Forbidden West 2 months ago and I don't remember temps being higher than standard 75-80 for the cpu and 65-68 for the gpu. Today I started Still wakes the Deep, but it's not so hardware demanding I think... Gonna check it with Cyberpunk tomorrow (still haven't played it).
The fans are all 140s - I knew you would ask so I added them to my signature below. They are spinning at high rps because the Fan control decided so after the calibration and I didn't have a spare moment to adjust them, besides I always wear headphones when at the PC, don't even have any external monitors. So it's not a problem.
The AIO is installed in the upper position. My case is the Meshify 2 by Fractal design, the original version, non-compact, non-C, there's tons of space inside for the air. I would even describe it like should you decide to instal the AIO afront, tubes down (like one should, according to Gamers nexus), the pump won't reach the processor
Haven't ever checked the pipes temperature before, will do.

 

[FlyingScot]

Okay, good info. I'll wait for your next update.
In an effort to give us some perspective, I ran Cinebench R23 on my cold, freshly booted system. I averaged 2 runs at 22C Ambient. This is what I got. Note that I just have a 6 core / 12 thread processor and I'm using one of the best air coolers on the market, probably better than most 240 AIOs. Even so, I hit 74C. Again, that's with a room temp of 22C. Now think about how much you are dealing with. Mine only produces 144W peak (142 avg). I tried to come up with a calculation for cooler efficiency and, if I compare our two results, I would say that it looks as I would expect. Maybe CiTay can add his 2cents. Maybe there's a better calculation out there on the "Internets".

Cinebench R23
CPU = 6 Core, 12 threads	Points = 10,020 (monitored)
Ambient Temp	22ᵒC
Max R23 Temp (Cold System)	74ᵒC
Avg R23 Temp	68ᵒC (Run1) - 72ᵒC (Run2)
Watts R23 Peak	144W (avg = 142W)
Volts (with Vdroop)	1.268 - 70V (avg = 1.250V)
My rudimentary CPR "Cooler Performance Rating" (higher number = more efficiency)	CPR Calculation = Peak Watts divided by (Peak ᵒC minus Ambient ᵒC)
My system:	Noctua NH-D15 Air Cooler
Watts per ᵒC	144W / (74C - 22C) = 2.77 CPR
Your system	AIO
Watts per ᵒC	230W / (96C - 22C*) = 3.11 CPR * = using my room temp as worst case scenario 230W / (96C - 30C) = 3.49 CPR

You will see that your cooler is allowing more watts per C. While these numbers are by no means scientific, they do suggest that we have got to try and address the high voltages that your CPU is running at. With Vdroop, I'm running 1.270V max. Your CPU appears to be running over 1.4V with Cinebench. That's one hot Tamales!

 

[AndyTheGreat]

Ok, about 40 minutes in Cyberpunk. I must also mention, that I set the AVX2 from -3 to -1, the core voltage offset - from -0.125 to -0.09 and set the Lite load to Mode 5. Without it I couldn't set the Lite load without BSODs.

 

[citay]

Negative offset for VCore and CPU Lite Load basically do something similar, they lower the voltage-per-frequency curve by different methods. While you can use both, i usually recommend just using CPU Lite Load for simplicity.

I don't have a thermometer in my place but it feels like 30-32 now with 35-36 outside. I use a fan all the time and am still sweating - this summer turned to be rather hot.

As soon as i read that, it all started to make sense. I always recommend to set the power limits so that the CPU ideally stays in the mid-80s range under full load with Cinebench, "because that leaves enough headroom for higher ambient temperatures in the summer". Well, looking at this, you already have extremely high ambient temperatures! We've had heatwaves here in Germany before, but i have yet to see my thermometers rising to something like 32°C inside. I always manage to keep it in the high 20s inside at the most, by keeping the sun out and the windows closed during the day, and opening the hatches at night when it's cooler.

So you and your PC are basically living in the worst-case scenario that we usually prepare for by tuning the power limits to stay in the mid-80s, a figure which was obviously meant for an average room temperature of 20-24°C or so. Going by this, your cooler has a higher cooling performance potential, but since it's working with unusually hot ambient air, you don't have it available at the moment. If these temperatures are only a short-time event, you can actually allow higher power limits again, it will even itself out with the temperatures once your room temperature drops. For a few days in the year, if it really gets that hot in the room (which is honestly hinting towards installing an AC), you and your PC may have to suffer a little. If you want to hunt for performance, these 220W power limits will start to look a bit tame once the room temperature is 10° lower.

 

[FlyingScot]

I was going to say the same thing, but CiTay beat me to it. Your ambient temps are the source of many of your concerns. And I think it would tilt me in the direction of believing that your AIO is doing the best job it can.

Your ambient temps are like +10C higher than any reviewer's environment that I have ever seen. In my mind, the normal room temp for reviewing cases, CPUs, CPU coolers, etc. is usually 21- 22C. So, your room temp is where 10 degrees (+/-) is coming from. We can't overlook that reality; that’s a big deal. Instead of seeing 80’s, you are seeing 90’s. 80’s doesn’t sound so bad, but 90’s is always instinctively alarming, perhaps because we have been trained to think that way. The new Intel and AMD folks would probably argue that 90's is fine. I beg to differ. Even if it is fine for the CPU, the CPU shares its space with all the other components in your system, including the CPU socket, VRM, etc. Frankly, I’m not surprised that you are also personally struggling with the high summer heat. That system of yours is also contributing to your room temps! Incidentally, it’s a good thing you sold those stock Fractal case fans. You’d probably be looking at a very unstable system on top of everything else.

As CiTay mentioned, the quick fix is to keep lowering PL1/PL2 levels until you can do your video editing with more reasonable temps. In terms of more complex solutions, I doubt there’s much you can do about indoor temps in your part of the world. But the next two problems you can do something about. You could find the absolutely best AIO cooler on the market, [see poilshsirhawk's recent tips below] or you could take a leaf out of Arctucas's book and do a custom water loop. Maybe that’s something you look at down the road.

"Intel's newest CPU heatspreader (IHS) is typically banana shaped (concave), where you've got a valley in the middle of it. Meanwhile the Arctic Liquid Freezer II coldplate is as flat as it gets. This situation results in a usually horrible geometry mismatch and temperatures way higher than you'd expect. A aftermarket contact frame helps but doesn't totally resolve the issue." "In terms of coolers on the market, the Valkyrie AIO is probably the only one with coldplate the most suitable for current gen Intels."
"Applying liquid metal (DC2 Pro, Conductonaut) should help tremendously with mismatch."

So that just leaves us with case fan optimization and CPU voltage to play with in the short term. When gaming, that Graphics card is going to heat up everything above it. It's why I was discussing it in this thread. I think it's relevant. So, perhaps you can work on optimizing your case cooling.

For case cooling, JayZ has some novel approaches in these two videos:

For future reference:

Voltage wise, we know it's on the high side. So we have to come up with a plan for lowering your voltages. It would be nice to see it in the 1.2’s or even 1.3s rather than the 1.4s. But it's going to take a lot of testing to find out how low you can go, which takes us right back to the beginning of your adventure. Incidentally, a hot system requires more voltage to remain stable, and unfortunately you get this feedback loop that happens. More voltage (when things are hot) results in more heat, which results in more voltage…This feedback loop continues until something steps in and puts a stop to it, which is where your power, amps, frequency limiters come into the picture. Again, your ambient room temp is on the extreme side. And it’s a major reason (I think) why you are struggling with your system. On top of that, you have chosen the 14700K, which is pushed to its thermal limits at the factory. No doubt, Intel expects room temp to be lower, too. I’m sure their factories run cool…
It's not a hopeless situation, though. And summer won't last forever.

 

[AndyTheGreat]

Damn you, Intel. Ok, tomorrow I'm setting up the 220 watts for the PLs, hope it helps. Don't like seeing 90s, worst years in my life they were )) I hope the weather's really the one to blame because I've even started to save the money for Lian Li Trinity Performance or Phanteks 360, those with the 7th gen Acetek pumps... Btw this Valkyrie you mentioned, @FlyingScot , is it any good? I mean I keep hearing about it all the time lately but this brand is not familiar to me at all, I suppose it's China-made, like my Deepcool AIO.

 

[FlyingScot]

Damn you, Intel. Ok, tomorrow I'm setting up the 220 watts for the PLs, hope it helps. Don't like seeing 90s, worst years in my life they were )) I hope the weather's really the one to blame because I've even started to save the money for Lian Li Trinity Performance or Phanteks 360, those with the 7th gen Acetek pumps... Btw this Valkyrie you mentioned, @FlyingScot , is it any good? I mean I keep hearing about it all the time lately but this brand is not familiar to me at all, I suppose it's China-made, like my Deepcool AIO.

I hear ya! But at least we're all on the same page now. You were correct in thinking that you were going to have to tackle this problem from both ends, cooling and voltage. In regards to water cooling, I'm definitely not the guy to ask. I don't think I'll ever need anything more sophisticated than my Arctic Liquid Freezer 280. Polishsirhawk was the one who recently offered up that tip to another 13th/14th gen user with cooling issues.