Where is the IA CEP option in the bios?

ke157002db

New member
Joined
Feb 11, 2023
Messages
4
Just setup a new MSI MPG B760I Edge Wifi DDR4 + 13900KF. I'm trying to undervolt the CPU. If I select a lower CPU Lite Load mode (mode 9 -> mode 5), the CPU performance decreases according to benchmarks.

I've read that the IA CEP is often the culprit and have seen screenshots from other MSI boards on how to turn it off (Advanced CPU options). However, I don't see the option on my brand new board. Should I update the bios version?
 
@citay

hi,

"C1E Support" to Enabled (clocks down in idle)
Is safe to use on my pc(gaming as primary usage) without any performance drop ?

Then we have the Active State Power Management modes, Auto should default either to L0s only or maybe to L0s+L1.
The best would be to select L0sL1 (where available) or L1 manually, this means maximum energy saving.

Should Peg 0 and Peg 1 both be set ?
This setting affect also the gpu performance ?
 
Yes, it's safe, no, performance is not affected. The only thing that is affected by using a deeper power-saving state is the so-called exit latency: The time until the device fully wakes up. Once it's in fully awake state, it will perform at 100%. And even that exit latency is negligible, it is measured in nanoseconds or a few milliseconds at most. Anything else you are doing like reading/writing to SSD has higher latency, so it is completely masked by other stuff, you will never feel a difference. Set it up like on the screenshot.
 
hi,
What's difference about Intel Speed Shift Technologies and C1E support? .... Seem both downclocks the cpu in idle
 
Speed Shift, as the name also implies, activates a faster, CPU-controlled way of frequency switching according to the workload and other factors, see here. The CPU can react to changing conditions faster than the OS, so that's why enabling this option is beneficial.

C1E is an improvement over the C1 state, you can see it in the picture i included here.
 
PC Specs : 13600KF, MSI B660M MORTAR, DDR5 16GB*2, RADEON RX 6800 XT, 700W PSU

If Windows Power Plan is set to 'Power Saver' in C10, system freezing still occurs. Nothing has changed.

BIOS Version : 7D42vAB
 
Last edited:
If Windows Power Plan is set to 'Power Saver' in C10, system freezing still occurs.

Which BIOS version are you on now? I recommend the one with "Support 14th Gen CPU", the second newest one. The very newest version after that has some known bugs.

Also, instead of selecting "Power Saver", try leaving it on "Balanced" and then changing the individual settings yourself to find out which one triggers the problem. For example, is it PCIe Link State Power Management, is it USB power savings, is it Processor power management...
 
Additional information from my own new system, Z690 TOMAHAWK WIFI DDR4 + i5-13500. I updated the BIOS first using the Flash BIOS Button.

No option "IA CEP Support" available in the latest BIOS, but just as MSI's table shows, i can undervolt the CPU without losing any performance, so IA CEP doesn't intervene.
I am undervolting using CPU Lite Load, see here on how to do that.

View attachment 170340

The other settings on this page are also important. Set "Package C State Limit" to C10 instead of Auto. When setting it to C10, the CPU will enter C7 state in idle, instead of only C6. The C7 state lowers power consumption in idle by around 6W for me (37W -> 31W).

Results:
CPU Lite Load Mode 12 (default) - CB23 score: 20962 - total system power draw: 182W, CPU only: 135W (HWinfo64 CPU package power).
CPU Lite Load Mode 1 (crashes in Prime95 Small FFTs!) - CB23 score: 20935 - total system power draw: 141 W, CPU only: 105W.
CPU Lite Load Mode 4 (stable so far) - CB23 score: 20955 - total system power draw: 149 W, CPU only: 114W.

The CB scores are all within the margin of error, basically identical. They only drop by ~300 points when i also monitor the sensors with HWinfo64 in the background.
(wasn't sure if to open a specific thread for this, or if to ask here, since discussion is related to the CPU current)
How about the "CPU Current Limit(A)"? Shouldn't it be kept as per Intel's spec, listed under IccMax, @citay ?
Isn't that too high in your case/picture? I own a i5-13600K and Intel specs lists a limit of 200 Amps. This is the value which gets used when selecting the "Boxed Cooler" profile and leaving the explicit option on Auto. This profile seems to be the only one which respects Intel's limits in terms of PL1, PL2 and IccMax.
I'm wondering more in general what you know about this IccMax, whether it's safe to go above the Intel limits, or if this could harm the CPU in the long run.
Also, why are the MSI settings and profiles exceeding Intel recommendations on this: when using either Tower Air Cooler or Water Cooler, the Current Limit is set to 512A??!!
 
I never adjust "CPU Current Limit(A)" manually, and i never select "Boxed Cooler" manually. If anything, i set the power limits (Long/Short) in Watts. Yes, selecting Boxed Cooler sets it according to Intel specs. But Intel themselves are responsible that, ever since the 9900K, especially the higher CPU models have more and more ridiculously high power draw under full load. They also, if not encourage the board makers, at least completely tolerate them ignoring the limits in most cases, because then the Intel CPUs look much better in the benchmarks during the launch reviews (which is what ultimately sells those CPUs). I posted a bit more about that here. Also, the IccMax can be overzealous in the way it regulates, it's better to only use the Long/Short Duration Power Limits.

The thing about the MSI cooler selection screen goes back to what i wrote here: The three options are not enough. You have Boxed = theoretical Intel spec, which can limit you too much in some cases. You have Water = unlimited which nowadays makes even a 14th gen i5 run into thermal throttling and bouncing off of 100°C when you don't have a high-end AIO cooler. Then you have Tower cooler = something in between, but still way too high for most tower coolers, it would even make some water coolers struggle, depending on the CPU model!

Normally you should have at least a handful of options there: Boxed / low-end air cooler, mid-range air cooler, high-end air cooler / low-end AIO, mid-range AIO, high-end AIO / unlimited.

But the best thing to do is to set the power limits yourself, according to what your cooling can actually handle. I have described the entire procedure here (although in that thread, the user has a high-end AIO, so he can use high power limits). First you'd set power limits roughly according to what you think your cooling can get rid of, even with long periods of full load. Let's say you got a decent tower air cooler, but not a high-end one. That could deal with perhaps 130-150W. Then you'd set that in the BIOS, for example 150W, and check the resulting temperatures with Cinebench, which creates fully multithreaded conventional load (the highest load that does not come from a stress testing tool like Prime95). If they're mid-80°C, perfect. Above 90°C, you have to reduce the limits, below 80°C you can raise them if you want, of course it also depends on the noise you want to tolerate, that has to do with the fan curves. I recommend the highest point of the curve (full fan speeds) at 85° or 90°C. I would try to stay away from the 90°C range, that slowly enters thermal throttling territory, it is not good to rely on that, and you want to have some headroom for higher ambient temperatures.

Then, to reduce the CPU's power draw in all load states, you can work with CPU Lite Load. Again, i explain that in the thread, also follow the links in there. No need to change the power limits from what you determined though, because those limits only have to do with your cooling capabilities, they don't change once you determined what your cooling can handle. But when you make the CPU draw a bit less power, it will throttle a bit less under full load, and it can boost the clocks a bit higher.
 
Hello!
Thanks for the info and references, I was already following the CPU Lite Load methodology for undervolting/less heat:
* wanted to keep to Intel spec out of caution, so I enabled the "Boxed Cooler" profile which sets PL1 = PL2 = 181 W and IccMax = 200 A. I'm using a tower air cooler.
* CPU Lite Load = Mode 7 (Values for CPU AC/DC Loadline are 35/80 after applying this)
With these I'm reaching temperatures of mostly 80 C or a bit more in the summer in CB23 and CPU Package Power of 140W. I didn't bother to go lower with Lite Load to determine the CPU's stability limits, but maybe I'll do it when I have more time.

However, the Current limit is the one I'm most fearful about, and that's because I couldn't find that much information and reliable/official documentation about it, the risk of running higher than spec etc. I really wonder how safe it is to keep it over limits. There's an interesting discussion about this here and in the follow-up.
 
* wanted to keep to Intel spec out of caution

If you mean damaging/killing the CPU, well, it has self-protection mechanisms like thermal throttling, which prevent any damage from overtemperature. If you have a low-end board model, you might cause the VRM area around the CPU to heat up considerably when the CPU power draw is very high. And of course, you don't want your CPU cooling being overpowered because it can't get rid of the heat from the CPU fast enough anymore. So if you mean that by caution, then yes, it's good to set power limits. If you see that, otherwise, the temperatures get to 90°C and above, then absolutely, power limits are useful to keep it below that.

But don't go by what Intel choose. Those numbers lost all their meaning for normal end users. They have some relevance for OEMs who build PCs and want to have their cooling be able to deal with the official numbers, but not more. That's why they sometimes use special CPU models like the -T models (example i7-13700T: 106W Maximum Turbo Power, 35W Processor Base Power), and use a motherboard where the BIOS is strictly enforcing those limits. Then they can use a cheap cooler and never have to worry about their customers getting into problems. Of course these low power limits, especially the low PBP of 35W (Long Duration power limit) will completely castrate this CPU. On the other hand, if you were to use this CPU with considerably higher power limits or unlimited, then it would probably get quite close to a normal 13700(K). Also read the bottom of this post.

And for average Joe building his own PC, what good should the Intel numbers be, when they don't know what cooling you use? For a cheap cooler, the limits are too high. For a 360mm AIO for example, it could deal with more heat. And when you leave a modern i7/i9 CPU completely off the leash, it wants to draw 300W and more, that's all with default Intel voltages and frequencies (plus a little extra from the board, which can be reduced by lowering CPU Lite Load). And when you set up the BIOS in a modern board, it will tend to default to maxed out power limits, often it will be pre-selected, or people will select the value that allows a high power draw because it's worded in a certain way. You have a tower cooler, you select tower cooler. Surprise, now the power limits are close to 300W, good luck to the tower cooler...

So it's much better to do a Cinebench R23/R24 run, check the temperatures, then adjust the power limits accordingly you so it's never higher than mid-80°C. If you find that you only have a power draw of 140W in Cinebench, meaning you're not even hitting the Boxed cooler power limits, then the power limits are completely meaningless for your configuration. That's the same for my own PC. I have an i5-13500, which goes to maybe 120W power draw in Cinebench and 140W in Prime95. Temperatures are very low because of my Noctua NH-D15, just as intended. So it doesn't matter if the power limits are set to 181W, 288W or 4096W, i just leave it on Water cooler.

However, the Current limit is the one I'm most fearful about, and that's because I couldn't find that much information and reliable/official documentation about it, the risk of running higher than spec etc. I really wonder how safe it is to keep it over limits. There's an interesting discussion about this here and in the follow-up.

That's exactly what i was referring to here. That's why relying on the Current limit is not the best method for keeping the cooling on top of things. First of all, it can throttle too aggressively if set too low. Also, we want to go by the resulting power draw in Watt, the product of voltage (V) * current (A). When we only control the maximum current in Amperes, we don't know what the voltage is. Every single CPU requests a slightly different voltage from the board, according to its quality. So we don't go by Amps, we go by Watts, the end result of it all, this is much more precise than trying to limit the current, for the purpose of a "cooling capability optimization".

But come on, you have 140W power draw, there is nothing to be fearful about. People here on the forum are running their 14700K/14900K at over 300W power draw without batting an eyelid, i have to explain to them that the CPU bouncing off 100°C and a thermal throttling from hell is not the ideal setup for daily use. Yes, ten years ago, 140W would've been extraordinarily high, but by now we are used to much more extreme numbers from Intel (not that that's good).

Anyway, you are fine even with maxed out power limits, your CPU is not even sniffing at the power limits. Try to find out how much additional voltage you can shave off by tuning CPU Lite Load and testing the stability, this is the best thing you can do. And then when it's stable at a relatively low power draw, and you see that your cooling is not overstressed, then there is nothing to worry about.
 
Last edited:
Anyway, you are fine even with maxed out power limits, your CPU is not even sniffing at the power limits. Try to find out how much additional voltage you can shave off by tuning CPU Lite Load and testing the stability, this is the best thing you can do. And then when it's stable at a relatively low power draw, and you see that your cooling is not overstressed, then there is nothing to worry about.
Do you know if enabling Undervolt Protection (UVP) ("CPU Under Voltage Protection" in MSI BIOS) works with AC Load Line undervolting method? Or does it only prevent undervolt when using Offset or other things? I was wondering since I wanted to find out when does Intel consider you go under spec, and I understand that's what UVP is for, to not allow that to happen.
 
I wouldn't worry about Intel's specs too much, they are usually very conservative. I mean, it's understandable, they can't allow wildly large ranges for anything, they have to set a narrow band where the CPU is guaranteed to work as intended. But when it comes to undervolting, you are tuning the CPU to perform better than spec (lower power draw and temps, higher efficency), so the specs don't matter much anymore. You are finding your own CPU's limits by stability-testing. No generalized spec can be of any use for it.

Also, where did you find UVP? I don't seem to have it:

MSI_SnapShot_40.png


I don't think this has anything to do with Intel specs though, this is just what MSI thinks are safe ranges, this is all controlled by the BIOS, it doesn't read out anything from the CPU. An UVP also doesn't make sense for the CPU, there is no risk of damage from undervoltage, just risk of instability, and you can find that out yourself from doing stability tests. UVP makes sense in PSUs, because the PSU voltages leaving specs on the bottom range means that there is an abnormal situation going on, something is drawing way too much current (like a short for example), so it's a good protection circuit to have that cuts power before something blows. But for the CPU, i don't see a need.
 
Hi! The setting is among the Advanced CPU Configuration ones on my board, here's an older screenshot. It's disabled by default, when using Auto.
Right underneath the CPU Lite Load and Over Temp Protection.
CPU-UVP.png

And here is Intel's documentation about it: https://www.intel.com/content/www/us/en/support/articles/000094219/processors.html
Hmm, by that page, it seems like it only prevents undervolt from OS, not from BIOS, so guess I also got my answer sort of.
Yes, I see that undervolting should be fine as long as you have stability, but is this stability a binary thing: either ON or OFF? I would be so happy if it was like that. I'm just a bit worried about other side effects of this undervolting and wonder why Intel don't want you to do it. I guess they also have some security reasons not to, as far as I read there, but I gues those are mostly theoretical, not found that easily in the wild.
There was an older thread here about potential input latency introduced by undervolting, but I doubt it could have any effect on it, maybe the testing method wasn't the best. I wouldn't want to inroduce additional input delay by doing it, but I think it's not that plausible, what do you say? I mean, how could USB polling of input devices be delayed and add up? Hard for me to figure how, if the performance is the same by game benchmark or Cinebench, still, extra delay could be introduced, but I'm no low level OS device polling expert.
 
I don't have it there either,

MSI_SnapShot_02.png



Hmm, by that page, it seems like it only prevents undervolt from OS, not from BIOS, so guess I also got my answer sort of.

Yeah, they talk about runtime undervolting, so with a program from the OS. Not necessary. This seems mostly for OEMs who want to lock down their BIOS from any shenanigans. They will use a custom BIOS with a lot of options missing, and set restrictively.

Yes, I see that undervolting should be fine as long as you have stability, but is this stability a binary thing: either ON or OFF?

Well yes, in the way that you will have errors with one setting when you stability-test properly, but not anymore if you raise the voltage again. And if you want to be really sure, and that's what i always recommend is, when you lower CPU Lite Load for example, and find the first mode that results in errors, then you don't raise it by one mode again, you raise it by two modes. This way, you have "stability headroom". You are not at the edge of stability, you have a buffer of additional voltage.

There are no other side effects worth mentioning, in my experience. That older post was rather esoteric, additional factors probably not controlled, and yes, it's implausible. The CPU shouldn't function any differently, unless there is some undetected instability, but that should be ruled out by doing proper stability testing and having additional stability headroom in the way i described. Of course, if that person was operating his CPU at the very edge of stability, then it could be possible that some algorithms in the CPU detected data errors and re-requested some data, which could introduce additional latency. But those sort of things you try to weed out by not going to the very edge. Also, whenever you notice something odd, it's easy just to go up another mode or two and see the problem disappears.
 
Well yes, in the way that you will have errors with one setting when you stability-test properly, but not anymore if you raise the voltage again. And if you want to be really sure, and that's what i always recommend is, when you lower CPU Lite Load for example, and find the first mode that results in errors, then you don't raise it by one mode again, you raise it by two modes. This way, you have "stability headroom". You are not at the edge of stability, you have a buffer of additional voltage.
Of course, if that person was operating his CPU at the very edge of stability, then it could be possible that some algorithms in the CPU detected data errors and re-requested some data, which could introduce additional latency.
Yes, this is the way I also understand it, thank you for confirming it.
These errors, would they be the WHEA errors that also have a section in HWInfo, for example? Which can be checked in the Event Log under "Applications and Services Logs"/Microsoft/Windows/Kernel-WHEA/Errors? Guess that's the reason why HWInfo offers an easy way to track them.
Are there any other places where these errors would show up or be logged, aside for the other, more obvious types, like freezing, application crashes (which could also be caused by other factors, not necessarily instability) or BSODs?
 
These could be low-level checks inside the CPU that the OS doesn't even register, you could only notice them through indirect measurements. But as i said, the guy who measured stuff in that other thread, that didn't seem like the most well-controlled test setup to me, so for now this is all mostly theory.

Errors that the OS notices would show up in the event logs. The stability tests will be more useful, they detect slight miscalculations that don't lead to any crashes/errors for the OS.
 
Back
Top