MSI PRO Z690-A WIFI DDR4 + i7-12700K | Optimal BIOS settings

[WildDoktor]

Hi; I'm back! I got a small bonus at work and used it to buy the above as new (old stock) for a decent price. My reasoning is that my gpu is cpu-locked at 3x pcie speed; moving up to an 11th gen intel cpu or higher will unlock it's full 4x pcie speed capability. That won't really get me more FPS, but more stability and apparently an easier road to less stutters. I could have gotten an 11th gen cpu and kept this motherboard, but...the prices on the new mb and 12th gen cpu were too good to pass up. Moving up to the 12900k broke the budget, to, 12700k is where I landed.

The MB is here, and the cpu will arrive in a few days, so after having had 100% success with help from this community (@citay !) for my current setup about 5 months ago, I'd like to begin exploring optimal settings for the new setup!

I'm only replacing the current z590-A Pro and i9-10900KF; everything else stays the same:

RTX 4070 super, nzxt 850 gold power supply, mag liquid cooler (2 fans) v2, corsair 4000d case with 3x 120mm front intake, 1x 120mm rear exhaust, 2x 120mm radiator fans exhausting thru the radiator out the top, 32GB of G.SKILL Ripjaws V Series 16GB (2 x 8GB) 288-Pin PC RAM DDR4 3600 (PC4 28800) Desktop Memory Model F4-3600C18D-16GVK

• My current bios settings (on the 590) are: long and short duration wattage 240, lite load 4, [Turbo Ratio Offset] +1

Running MSFS 2024 full-tilt, my current cpu/gpu temps are averaging 65C.
I've seen that the 12700 will run hotter than this, but not by much, so we'll see.

I have already downloaded the latest BIOS and drivers, so I'll of course do that first thing and run at defaults for a few hours to get a feel for it and work out any bugs.

But after that: time to optimize! So: what suggestions do you have for me for this setup?

Thanks!

PS - I'm re-reading @citay's cpu power post atm.

 

[citay]

Yeah, basically you can go by my guide again. I would always advise to start off without any overclocking whatsoever, and instead use the CPU's full potential for undervolting, as in step 2) of my guide. Because overclocking and undervolting are two sides of the same coin. Whatever reserves the CPU has, you can use them to drive it further than factory, which requires more voltage, or make it more efficient than factory, lowering the voltage but keeping the full performance. Or even increase it slightly, should it be limited in any way (by power/temperature limits).

For most people, overclocking is a purely psychological thing, they will never notice 100 MHz more in gaming, but it's about getting "more bang for your buck". But to me, vastly improving the efficiency of the CPU by undervolting it, that is where the real reward lies. All of a sudden, while still performing like a full 12700K or better, you can achieve it at much lower power draw, heat, and potentially noise. When it comes to energy spent per workload, that CPU will blow any stock 12700K out of the water. Let alone an overclocked one, because with overclocking, you rapidly lose efficiency, once the voltage has to be driven up considerably for a minuscule frequency gain.

Now, the power limits, you can keep roughly at similar values than before, because they depend on what your cooling can deal with. Of course, you need to check at which temperatures you arrive at, and try to keep them below 90°C. Then the lowering of "CPU Lite Load", this has to be fully tested all over again, find the ideal mode for it. Testing one step above the first unstable mode, and if that is confirmed stable, add another step for stability headroom.

In this post you find some additional settings that I like to apply.

 

[WildDoktor]

Awesome; thanks @citay. I'm looking forward to tweaking the new setup. I've certainly changed my views on overclocking due to your excellent guidance; less power, better performance and stability, less heat and less noise over pure speed makes sense to me now, and I've seen it work in my own setup. So that's what I'll be going for with the new stuff! Probably won't get to work on it until Tuesday night.

 

[serj_7509]

Hi so i recently also upgraded my board to the msi z790 tomahawk and also have a i7 12700kf, do you happen to know on what undervolt settings you ended up sticking with? im pretty confused in all of this but i do want to reduce power draw since stock settings kinda suck

 

[citay]

The model of CPU might be the same, however, each CPU is individual, in regards to its undervolting ability. The CPUs that Intel produces are binned to achieve a certain spec (and then if the iGPU has a defect, they can sell it as an -F model), but that spec has a certain tolerance, so slightly higher-quality and slightly lower-quality silicon can still become the same CPU model. This reflects in how much you can undervolt each individual CPU later. But this also means that any values another user would tell you, in regards to undervolting, cannot be simply applied to your CPU. Either you'd leave something on the table, or you'd overdo it any have instability, because no two CPUs behave identically there.

What to do then? Well, that's why I wrote the Guide: How to set good power limits in the BIOS and reduce the CPU power draw. It gives you the way of finding the best settings for your individual CPU. Once you have found them, they will automatically be better than any values others tell you they ended up with. Those can only be ideal for their CPUs.

 

[serj_7509]

thanks for the guide, i ended up doing something around cpu lite load 10 (down from the defaulted 12) alongside doing a +2 ratio boost on both the E and P cores and it was stable during the extreme avx2 occt test for 30 minutes, and was able to achieve 23592 points in the multithreaded r23 cinebench benchmark. Is it a perfect undervolt? Probably not since i just hammered away with the cpu lite load values until it was stable but its getting 7% better performance alongside using less power so i stopped tweaking further

 

[citay]

Well, that's a good start, better than nothing. But two remarks there:

1) Undervolting and overclocking are two sides of the same medal. They each try to exploit the additional headroom that an individual CPU has from factory. But they each do it in the opposite direction. While, with undervolting, you would invest all that headroom into a lower voltage per frequency, resulting in lower power draw, lower temps, and higher efficiency at the same performance (or better performance, if it hits the power limits), with overclocking you start to invest the headroom primarily into higher performance, often at similar or higher voltage per frequency, usually resulting in similar or higher power draw, higher temps, as well as worse efficiency, if you overdo it.

Now, you can strike a good balance where, if you keep the OC modest, you will not suffer from the negatives much. But any more significant OC would directly conflict with the end goal of the guide, and make most CPUs behave worse than factory. You are lucky in that you probably have the last i7 that was somewhat civilized. The 13th and especially 14th gen i7 were pushed so hard from factory that even the tiniest OC would immediately result in extreme downsides. But even with a 12700K, I would already put all my eggs in the "undervolting" basket.

2) Why stop tweaking? You might have a "golden sample" CPU, whose silicon quality surpasses those of most other samples. I had some CPUs before that were stable at Mode 1, believe it or not. If you stop randomly at some mode, just because it was stable and already netted some improvements, then you haven't really optimized it fully yet. I would investigate further, see where you get the first instability. Stopping at Mode 10 is a shot in the dark, if you haven't also tested lower modes. And if you set power limits according to your cooling, and the CPU hits those limits (i.e. natively would want to draw more), then the more you can lower the mode while staying stable, the more the performance will increase, totally independent of any OC.

 

[hypedcritic]

Yeah, basically you can go by my guide again. I would always advise to start off without any overclocking whatsoever, and instead use the CPU's full potential for undervolting, as in step 2) of my guide. Because overclocking and undervolting are two sides of the same coin. Whatever reserves the CPU has, you can use them to drive it further than factory, which requires more voltage, or make it more efficient than factory, lowering the voltage but keeping the full performance. Or even increase it slightly, should it be limited in any way (by power/temperature limits).

For most people, overclocking is a purely psychological thing, they will never notice 100 MHz more in gaming, but it's about getting "more bang for your buck". But to me, vastly improving the efficiency of the CPU by undervolting it, that is where the real reward lies. All of a sudden, while still performing like a full 12700K or better, you can achieve it at much lower power draw, heat, and potentially noise. When it comes to energy spent per workload, that CPU will blow any stock 12700K out of the water. Let alone an overclocked one, because with overclocking, you rapidly lose efficiency, once the voltage has to be driven up considerably for a minuscule frequency gain.

Now, the power limits, you can keep roughly at similar values than before, because they depend on what your cooling can deal with. Of course, you need to check at which temperatures you arrive at, and try to keep them below 90°C. Then the lowering of "CPU Lite Load", this has to be fully tested all over again, find the ideal mode for it. Testing one step above the first unstable mode, and if that is confirmed stable, add another step for stability headroom.

In this post you find some additional settings that I like to apply.

I know this is probably an older post by now but I wanted to get some more information off from you even after following some of the BIOS tweaks that you had added yourself.

I checked through all of them and did notice some things that I wouldn't really understand on my own and was thinking if you had any advice for them, is MSI Enchanced Turbo beneficial or should I just stick to keeping it disabled? I was also wondering what other things you had changed if not mentioned in other guides, like do I bother with doing any Turbo-related changes onto the cores themselves? (I am using PRO Z690-A WIFI with an i7-12700KF)

I did a run on CinebenchR23 (20189) and got some decent temps around 80c with no thermal throttling, with some OK wattage usage when I was on Lite Load 7. I'm not really sure what to go off from this. Thanks for the help if you responded!

 

[citay]

is MSI Enchanced Turbo beneficial or should I just stick to keeping it disabled?

Leave it disabled. "Enhanced Turbo" is a function which actually equates to overclocking. It's mostly a way for the boards of one brand to perform better than the boards of competing brands in reviews. Because if all other components are comparable, the boards would all have roughly the same benchmarks results, as the board itself has very little say on the overall performance. It's mostly down to the CPU, RAM and GPU. But obviously there can be a huge impact on performance once the BIOS settings are changed from the accepted norm or the Intel standard (also in regards to the power limits).

Now, "Enhanced Turbo" would be going way above a simple power limit increase. What this does is, it allows the highest turbo multiplicator for single-core load to be active no matter how many cores are fully loaded. Usually, the more cores are loaded, the less high the CPU is allowed to clock, in order to keep power consumption and heat production in check. But not so with this function. So when you have a rather high-end CPU, this can wreak havoc. It's better to set it to disabled. Other brands also have this "dirty trick" up their sleeve of course, on ASUS it's called "MultiCore Enhancement".

I was also wondering what other things you had changed if not mentioned in other guides, like do I bother with doing any Turbo-related changes onto the cores themselves? (I am using PRO Z690-A WIFI with an i7-12700KF)

I don't bother changing the multipliers or anything like that. I only aim to lower the power draw per load and increase the efficiency, by tuning via "CPU Lite Load".

I did a run on CinebenchR23 (20189) and got some decent temps around 80c with no thermal throttling, with some OK wattage usage when I was on Lite Load 7. I'm not really sure what to go off from this. Thanks for the help if you responded!

Well, did you try even lower CPU Lite Load modes? If not, that's the thing to try. Of course, keep a close eye on stability. Unless you have a rare CPU that's still stable at Mode 1 (not unheard of, but I wouldn't bank on it), the stability will slip in some mode. Then you have to thoroughly check the mode above that, and if that's stable, add another step for stability headroom.

To arrive at Mode 7, it should mean that Mode 5 showed the first instability, then you confirmed that Mode 6 was stable and thus arrived at Mode 7 for the final setting. If you haven't done this testing fully, then it's possible that the first instability would only happen in Mode 3 for example, so you could use Mode 5. It simply has to be tested.

 

[hypedcritic]

Leave it disabled. "Enhanced Turbo" is a function which actually equates to overclocking. It's mostly a way for the boards of one brand to perform better than the boards of competing brands in reviews. Because if all other components are comparable, the boards would all have roughly the same benchmarks results, as the board itself has very little say on the overall performance. It's mostly down to the CPU, RAM and GPU. But obviously there can be a huge impact on performance once the BIOS settings are changed from the accepted norm or the Intel standard (also in regards to the power limits).

Now, "Enhanced Turbo" would be going way above a simple power limit increase. What this does is, it allows the highest turbo multiplicator for single-core load to be active no matter how many cores are fully loaded. Usually, the more cores are loaded, the less high the CPU is allowed to clock, in order to keep power consumption and heat production in check. But not so with this function. So when you have a rather high-end CPU, this can wreak havoc. It's better to set it to disabled. Other brands also have this "dirty trick" up their sleeve of course, on ASUS it's called "MultiCore Enhancement".



I don't bother changing the multipliers or anything like that. I only aim to lower the power draw per load and increase the efficiency, by tuning via "CPU Lite Load".



Well, did you try even lower CPU Lite Load modes? If not, that's the thing to try. Of course, keep a close eye on stability. Unless you have a rare CPU that's still stable at Mode 1 (not unheard of, but I wouldn't bank on it), the stability will slip in some mode. Then you have to thoroughly check the mode above that, and if that's stable, add another step for stability headroom.

To arrive at Mode 7, it should mean that Mode 5 showed the first instability, then you confirmed that Mode 6 was stable and thus arrived at Mode 7 for the final setting. If you haven't done this testing fully, then it's possible that the first instability would only happen in Mode 3 for example, so you could use Mode 5. It simply has to be tested.

I had jumped down from Mode 12 (Auto) to 7 because I thought it would be a decent jump down. I took a skip again and made it to Mode 5 but I discovered that I would crash before I even went past a 1 minute in Cinebench23. I knocked it down to Mode 6 and it didn't crash but the report from the bench was lower with lower temps and wattage. (19966, 77C, 157W) Should I stick with using Mode 7 or try to go lower than 5, or would that cause more instability? I do also use Auto PLs but they're limit is 253 as addressed with a tower cooler recommendation.

Also, what is your way to go when it comes to undervolting GPUs? Especially running them at 1440p. I'm running a 4070 Super and get a bit skeptical when I see high 70s as the temp. Not really sure if it's something to be so uneasy about but I feel that way. I have yet to do anything with my GPU as I would get very inconsistent instability when I would try and do my own thing with it.

Thank you for your response, I'm glad I'm able to talk with someone who is knowledgeable about this instead of just watching a bunch of videos and learning without asking questions, giving my own experiments, etc.

 

[citay]

I had jumped down from Mode 12 (Auto) to 7 because I thought it would be a decent jump down.

It is a decent jump, but you have no idea where you land. It's ok to jump to Mode 7, but then you have to check further. Is it fully stable here? Then try lowering the mode further. Or if it's already showing hints of instability in stress tests, obviously you know that you will land somewhere higher, but probably still below Mode 12 (since Mode 12 tends to be quite conservative).

I took a skip again and made it to Mode 5 but I discovered that I would crash before I even went past a 1 minute in Cinebench23.

Yeah, if CB crashes, it's not good, it's not even a proper stress test, just a (demanding, admittedly) benchmark.

I knocked it down to Mode 6 and it didn't crash but the report from the bench was lower with lower temps and wattage. (19966, 77C, 157W)

Did you set "IA CEP Support" to [Disabled], as explained in my guide? But normally, a couple hundred points difference from run to run doesn't mean that much, it might still be within the margin of error when converted to percentages from such high numbers. Of course the power draw is reduced with each step, as the undervolt increases, therefore the max temperatures are also lower. But nothing means anything without stability.

You know Mode 5 is unstable, so the lowest mode you should consider using is Mode 7. Even that has to be stability-tested thoroughly, if it hasn't happened yet. For example with OCCT.

I do also use Auto PLs but they're limit is 253 as addressed with a tower cooler recommendation.

I guess you didn't perform step 1) from my guide. The defaults are not fitting for 99% of the situation. An air cooler that can comfortably deal with 253W worth of heat, there's not even a handful of those available (the very top air cooler models from Noctua and be quiet! for example). Most normal tower coolers people have, staying below 90°C at 180-200W CPU Package Power would be a nice result.

Now, since you mentioned 157W at Mode 6, I'm guessing you get around 165W or so max with Mode 7, so the power limits don't even come into play. With the reduced CPU Lite Load, your CPU cooler doesn't seem to be in danger of being overpowered by the heat output. That's the scenario where you can skip step 1) of the guide, when your CPU cannot cause problems for your cooler under full load.

So, stay at Mode 7, since 5 is unstable and you need stability headroom, you can't use Mode 6. But test 7 more thoroughly.

Also, what is your way to go when it comes to undervolting GPUs? Especially running them at 1440p. I'm running a 4070 Super and get a bit skeptical when I see high 70s as the temp. Not really sure if it's something to be so uneasy about but I feel that way. I have yet to do anything with my GPU as I would get very inconsistent instability when I would try and do my own thing with it.

80°C is not much for a GPU. But it does pay dividends to undervolt it too. You'd use the Curve Editor in MSI Afterburner to undervolt the GPU (often times they want you to also slightly overclock it at the same time, but even a pure undervolt is very good). See here for example. You can most likely find specific guides for the 4070 Super on Youtube.

 

[hypedcritic]

It is a decent jump, but you have no idea where you land. It's ok to jump to Mode 7, but then you have to check further. Is it fully stable here? Then try lowering the mode further. Or if it's already showing hints of instability in stress tests, obviously you know that you will land somewhere higher, but probably still below Mode 12 (since Mode 12 tends to be quite conservative).



Yeah, if CB crashes, it's not good, it's not even a proper stress test, just a (demanding, admittedly) benchmark.



Did you set "IA CEP Support" to [Disabled], as explained in my guide? But normally, a couple hundred points difference from run to run doesn't mean that much, it might still be within the margin of error when converted to percentages from such high numbers. Of course the power draw is reduced with each step, as the undervolt increases, therefore the max temperatures are also lower. But nothing means anything without stability.

You know Mode 5 is unstable, so the lowest mode you should consider using is Mode 7. Even that has to be stability-tested thoroughly, if it hasn't happened yet. For example with OCCT.



I guess you didn't perform step 1) from my guide. The defaults are not fitting for 99% of the situation. An air cooler that can comfortably deal with 253W worth of heat, there's not even a handful of those available (the very top air cooler models from Noctua and be quiet! for example). Most normal tower coolers people have, staying below 90°C at 180-200W CPU Package Power would be a nice result.

Now, since you mentioned 157W at Mode 6, I'm guessing you get around 165W or so max with Mode 7, so the power limits don't even come into play. With the reduced CPU Lite Load, your CPU cooler doesn't seem to be in danger of being overpowered by the heat output. That's the scenario where you can skip step 1) of the guide, when your CPU cannot cause problems for your cooler under full load.

So, stay at Mode 7, since 5 is unstable and you need stability headroom, you can't use Mode 6. But test 7 more thoroughly.




80°C is not much for a GPU. But it does pay dividends to undervolt it too. You'd use the Curve Editor in MSI Afterburner to undervolt the GPU (often times they want you to also slightly overclock it at the same time, but even a pure undervolt is very good). See here for example. You can most likely find specific guides for the 4070 Super on Youtube.

I did disable IA CEP Support. I only didn't do step 1 because I didn't do many tests so I just ended up sitting with the Auto, I have just now lowered it to 200W for both of them and I can lower it if needed down to 165W~ if the wattage doesn't need to be higher.

I followed up on trying to use OCCT and it kept crashing, not sure if i was using it right but I chose CPU Stability test only on Normal and it crashed before it could even start after 1s. So I'm assuming Lite Load 7 has some instability as well? Running Stability Tests seems to crash it but the Benchmarks are OK?

 

[citay]

Well, try Mode 9 for example. If you can suddenly run the CPU test in OCCT just fine, then Mode 7 was too optimistic and not stable in a real stress-test.

 

[hypedcritic]

I tried Mode 9 and it lasted for 9s before freezing. I have changed IA CEP Support to Auto seeing if it was the problem, and it didn't crash when I had started the stability test and it didn't crash during it for the 10m stability test. Should I lower down the Lite Load now since it worked fine with 9 when IA CEP Support was set to Auto?

 

[citay]

"IA CEP Support" will prevent effective undervolting, by essentially lowering the clocks/performance accordingly. With Mode 9, it might still be just about ok, but you have to check the Cinebench score. If it has dropped too much, then it's no good. Looks like your CPU might need a higher voltage than you thought. It can happen. After all, a Mode like 12 was chosen to make all CPUs of varying voltage requirements work and be stable. So there are some CPUs out there that you can barely undervolt at all before they start to show an instability. Yours could be one of them.

It's not the end of the world, but it means you might have to stay at a higher mode. You could set IA CEP to Disabled and then test Mode 10.

 

[hypedcritic]

Ok. I've disabled IA CEP and moved onto Mode 10. It has not crashed with that but my CPU Package Power is above the PL1 and PL2 limit? Not by much, by like 2-4W, I am not sure what to do about that, I guess I could try and lower them more but I have them both at 200W.


Does having the C-State thing at C10 cause any interference as well? When I read your guide you mentioned it could have some instability but not common

 

[citay]

I guess I could try and lower them more but I have them both at 200W.

Depends on your temperatures. If you're still quite comfortably under 90°C during the end of a Cinebench Multi run, you could even raise the power limits a bit. But if your cooler is already struggling a lot, you may need to lower them instead. The power limits are the insurance policy for your cooling, that it's never overpowered and you never run into thermal throttling. You always want the power limits to kick in before thermal throttling has a chance to kick in.

Does having the C-State thing at C10 cause any interference as well? When I read your guide you mentioned it could have some instability but not common

Nah, that's only for idle. It can put more of the CPU to deeper sleep when you set it to the deepest C-state (higher = deeper = lower power draw in idle).

 

[hypedcritic]

Alright. Thanks for the help, man. I'll tune the wattage if need be in the future