Need help reaching Package C8 on Intel N100

[tmkhiem]

Hi! I just bought a N100 laptop and performed a clean Windows 10 Installation on it. Since I will be using the machine for on the go working, after installing all drivers, I proceed to download ThrottleStop to optimize the battery usage. However, I'm at my wit's end because I have tried everything that I know (there is a list of them below) but no matter what I do, I cannot achieve package states deeper than C3.

Laptop's specification:

• Manufacturer: CHUWI
• Intel N100 Alder Lake
• LPDDR5 16GB 4800MHz soldered on
• 512GB AirDisk SSD
• BIOS: AMI
• LAN NIC: Internal Realtek USB FE
• WLAN NIC: Intel AX101

As with most noname manufacturers out there, I got almost every setting imaginable available for tuning. Almost nothing is hidden away. I could configure FIVR in the BIOS but whether it is effective is an another story, however I could see that disabling CFG lock in BIOS did reflect in ThrottleStop (no more "C States Locked").

Software specification:

• Windows 10 Home 22H2
• ThrottleStop 9.6
• Timer resolution in "Options" 15.62, a "relaxed" value I believe.
• powercfg /energy returns no error and some minor warnings.

What I have tried:

• Disabling the Internal Realtek LAN NIC. I also have an another laptop (i7-8550U) that only reaches PC8 with latest Realtek drivers. I tried updating the Internal NIC but it didn't work this time. Also

  powercfg /energy

  reported that it cannot do selective suspend and with it being a 100M only NIC I disabled the USB port altogether.
• Updating the Microsoft AHCI driver to "Intel 500 Series chipset AHCI controller". I have read somewhere that MS default AHCI drivers will prevent the system from going deeper than PC3 but still no luck. I know that Alder Lake would be equivalent to 600 series chipset but I can't find the AHCI drivers for 600 series anywhere.
• Followed this post on modifying relevant BIOS settings. I don't need to use the mentioned tool (AMISCE) because they're all available . But didn't work.
• Manually forcing DMI ASPM, SA PCIE ASPM and PCH PCIE ASPM to Auto, L1 and L0sL1. Both without success. I'm leaving it at L1.
• Manually entering C6/C7, C8, C9 and C10 Latency values that I copied from another system. My theory is that it's possible that a zero value may have somehow disabled the C state. It seems not to be the case, though.
• Enabled S0i2.0 and S0i3.0 -- no changes. From what I gather, they are supposed to be the "new sleeping" standard in place of S3 so I figured they might have something to do with low power.
• Turning on Speed Shift + EPP since it is disabled the first time I open ThrottleStop. I'm not sure if this matters with AlderLake.

What's weird:

• Apparently, there could be some BIOS settings that will render other settings ineffective (i.e. will be of no function without being grayed out). I tried limiting maximum Package C State to C2 and C1 but both ThrottleStop and HWINFO still shows some PC3 residency. However, some other tinkering around results in PC3 and PC2 not getting any altogether so I thought my current BIOS setting is as good as I can get.
• Right now both core C6 and C7 are at over 90%. I don't think this is possible because I think the sum of one core's C3%, C6% and C7% should not exceed 100% (Did I understand it right?). This is verified with MSR reading at 0x000003FD (CC6) of 00000020:2D553138 and reading at 0x000003FE (CC7) of 0000001F:5AE1F26F, while 0x000003F9 (PC6), 0x000003fa (PC7) and deeper states return all zeros.
• "C-State auto demotion" and "C-State undemotion" has only two possible values: Disabled and C1. Selecting C1 results in ThrottleStop showing "C1" button instead of "C9" but inside ThrottleStop's C States window still shows C9.
• In some combination of BIOS settings (I did not do a great job at keeping track of which settings has what value partly due to the vast amount of settings), requesting C7 will prevent Package C3 but allows Package C2.
• In some other combinations, like what I'm having currently, requesting C1 does not have any effect at all. The package still reaches C3.

Attached are some screenshots of ThrottleStop. Due to the vast amount of configurable settings in the BIOS I could not show them all here, and some of them have been changed, but any help would be appreciated!

EDIT: Added some logs.

 

[unclewebb]

both core C6 and C7 are at over 90%. I don't think this is possible

I agree. On most Intel CPUs, the sum of the individual core C states as reported in ThrottleStop does not exceed 100%. The values in the C6% column of the ThrottleStop C state monitoring table are all higher than the C7% column. Whatever is being reported in the C6% column might be multiple different C states including C7%. I have never owned or had access to an Intel N100 CPU for testing purposes. How well ThrottleStop works on the N100 is unknown. There might be bugs like the C state bug you found.

On my 10th Gen desktop CPU, I have never made it beyond package C3. It might be a driver that blocks the deeper package C states. I gave up trying to figure out why. On a desktop CPU, a tiny amount of idle power is not worth chasing after. Hopefully you can find a way to unlock the package C state mystery. Lots of laptop owners could use some extra battery run time.

 

[mkdr]

I just got a Chuwi Herobox 2023 with Intel N100 (same configuration as your Chuwi laptop it seems, same ram, same Intel AX101) and set it up the last hours. The included fan is really annoying. But I had high hopes that I could just tweak the device, so I could detach the fan and make it run fanless. Which I actually just achieved. I limited PL1 and PL2 to 8W, and now the device runs fanless (I just opened it and detached the fan cable) between 70°C to 82°C.

The only thing so far though: I am not able to make the package enter any C state higher than C3 it seems. I found this post, and I get same results in ThrottleStop for the N100. Around 2.5W idle package but TS shows for package highest C state is C3 and for cores C7.

Wondering if the N100 cant go lower for package than C3? Or if this is some configuration issue. I tried to look into the bios, but all looks normal to me, couldnt find any settings so far where C states were disabled.

My Dell XPS 15 9570 with 8th gen 8750h CPU, it enters C8 for package (C10 only but only when I remove the power cord and it runs on battery) and can go low as 0.6W, with modern standby it enters C10 (I notice there is kinda no difference in power draw anymore it seems between C8 and C10). I am sure the N100 could do the same, the 2W for idle difference is actually a lot. Wonder if it is some stupid Windows or driver issue. I tried both power profiles with no luck, balanced and energy saving, no difference.

Update: After some more testing, when I enable modern standby, aka "Low Power S0 Idle Capability" in Chuwi bios to enabled, I get C10 state shown in TS, but only, when I put the box to sleep / modern standby. Under normal running condition, I still dont get lower than C3 for package and around 2.4W.

 

[tmkhiem]

Does your system also show the CC6 and CC7 values over 90% simultaneously? I think there is something wrong (possibly a bug unrelated to ThrottleStop but maybe msr/microcode?) here.

 

[mkdr]

@tmkhiem yes indeed, it shows the exact same values as yours. so that might be a bug actually showing both C6 and C7 kinda the same values for core, or there is some weird definitin how C7 is counted that it also counts C6 with it. but the state for package must be correct, because that is kinda the same value I also get for my 8th gen 8750h when it wont enter C8 and just C3 for package, it would be idle between 2-3w and when it can go down to C8 or C10, package goes to around 0.5w. so thats kinda 2w of waste for idle which is a lot.

I went through bios again and tried a few things nothing worked. I also updated the Realtek GBit Adapter driver to latest from Realtek page for the one which includes power saving, also didnt help. wonder if some of the countless drivers maybe preventing it, like bluetooth, usb, or some of the Intel drivers. I tried to disable the AX101 which didnt help, also removed all USB cables I had on the box. no luck so far. might also be a bios issue of the Chuwi devices which all share the same bios.

I have tried to look if I see anywhere a screenshot posted by anyone on the web from a N100 in TS, proofing that it can indeed enter C8 and go down to 0.5w for package, not found one so far. might also be some combination of the iGPU maybe from the N100.

for your laptop, did you update the Intel GPU drivers, did you check in the Intel GPU tool, the old one, not the new purple tool, under display, if you have a display power save option? I remember I had this issue on my Dell XPS before, that if you dont enable it, it also wont enter pkg C8.

 

[tmkhiem]

I did not remember if I had updated the GPU drivers. The current Intel UHD Graphics driver on my system is at version 31.0.101.5081, and in Intel Graphics Command Center (not sure if its the tool that you were referring to, but it's not purple) it looks like this:

EDIT: After updating the drivers to 31.0.101.5522 from Intel, nothing seems to change.

I am wondering if the fact that I'm refusing to use Windows 11 that Alder Lake officially supports plays a role in this. I have always thought that since N100 only contains E cores and no P cores, it might not matter whether Intel's Windows 11-exclusive ThreadDirector is present or not.

 

[mkdr]

@tmkhiem I meant these display saving settings:

It seems these settings are missing for you. Thought they might be present for a Laptop maybe.

Like I said. I still have not seen a 100% proof shown by anyone, the N100 can go lower than 2.5w idle and reach C8 package power state under Windows shown in TS. I have not seen anyknow shown a proof for it with a picture. I have no idea, if the N100 can do, or if Intel changed something for it, or if it maybe is a error in TS not properly shown. My 8750h cpu on my Dell XPS 15 9570 though can do it, reach c8 or even c10 and go low as 0.5w package power. So something is fishy here. 2.5w is a huge waste of power and heat for idle of the package.

I couldnt unlock undervolting so far too, even with CFG lock off in bios I can set in the Chuwi bios. Did you try to get undervolting working? There seems to be a special undervolt lock flag you need to flip, but I cant find it in the Chuwi bios.

Speedshift was shown as disabled too for mine in TS, not sure why. In bios Speedshift is enabled. I enabled it in TS, didnt make any difference though.

 

[mkdr]

Here is an example btw of my Dell XPS 15 9570:

So I am 99% sure the N100 can do the same, yet something is blocking the C8 state. Or it is a bug in TS? It shouldnt show also like you said C6 and C7 for the cores at the same time. Maybe the E cores kinda work different than older cores or P cores?

Also I noticed for my N100 the button in TS showing C1 instead of C10 for the N100 @unclewebb ? On the above screenshot it shows C9 for @tmkhiem

Edit:

I noticed that in the Chuwi bios, the option ACPI D3cold ist disabled, I tried to enable it but it still had no impact. Yet as I understand D3cold has something to do with low energy states for USB, GPU, PCIe and so on. Also noticed some options are foced to D3hot in bios too, for GPU, and also SATA/hdd it seems. Maybe the included SSD in both your laptop and my Chuwi cant support low power state and causing package power to not go deeper?

 

[unclewebb]

Also I noticed for my N100 the button in TS showing C1 instead of C10 for the N100

The ThrottleStop C State button will show whatever the package C state limit is set to.
When the package limit is set to C1, my 10850K does not use any of the package C states. The package C states will all report 0.0. It appears from the screenshot above that the N100 does not work the same way. A setting of C1 on the N100 might be equivalent to an Unlimited setting. The Intel documentation and how a CPU actually works has always been a a little fuzzy when it comes to the package C states.

None of the 12th Gen and newer screenshots that I have seen show the E cores using core C7. When the P cores show core C7, the E cores seem to all be limited to core C6. I am not sure why.

Or it is a bug in TS?

Without access to a laptop with a N100 for testing purposes, I cannot guarantee that all ThrottleStop features will work 100% correctly on these CPUs.

Have you tried using HWiNFO? It can report C state activity.

 

[mkdr]

Here is another screenshot with HwInfo @unclewebb It seems to show the same weird data for the N100, with showing C6 and C7 at the same/close similiar vlaues for the cores.

What are these settings with Demotion / Undemotion, whats that all about? I tried to change them in bios, on/off/c1 for core demotion/undemotion but it seems to have no effect, or I couldnt see any:

There are a bunch of setting all over the bios which might give some hints but I couldnt find out some so far and I dont want to just test them without knowing what they do.

Noticed Storage is forced to D3hot, mostly means storage never will enter sleep mode?

Not sure what DeepSx power policies mean, it is off.

Also some options with the name C10 in them off, like C10 dynamic threshold adjustment and

Then noticed this was off ACPI D3cold but activating it seems to do nothing:

And for the C states:

I tried to change all of them with no luck. Setting package c state limit from auto to "cpu default", or setting it to c10, doesnt seem to change anything. Set c-state auto demotion and undemotion from c1 to disabled. Set package c-state demotion undemotion from enabled to disabled. What does demotion and undemotion settings do anyway?

Also noticed this:

For mx Dell XPS 15 9570 HwInfo shows for "system agent power" a really low value:

Where on the N100 "system agent power" never goes below 1.8w:

It seems something is blocking the "system agent" to go into sleep mode on the N100, and from the total of 2W package power idle train, 99% is the system agent in this case.

 

[unclewebb]

another screenshot with HwInfo @unclewebb It seems to show the same weird data for the N100, with showing C6 and C7 at the same/close similar values for the cores.

ThrottleStop and HWiNFO are reading the same data from the same registers that they always have. I guess that confirms that Intel changed something C state related in the N100 design. It might be documented somewhere but I have not gone looking for what Intel might have changed.

Demotion / Undemotion, whats that all about?

Demotion and Undemotion controls how quickly a CPU transitions up and down between different C states. There is not a huge difference whether these options are checked or not checked in ThrottleStop. Some users blame some in game stuttering on these settings. I have never done any thorough testing of them.

Your BIOS has access to an overwhelming number of options. I have never played around with an unlocked BIOS like that so I do not know if any of those settings are useful or not.

It seems something is blocking the "system agent" to go into sleep mode on the N100

That is possible. HWiNFO does not report System Agent power for my 10th Gen desktop so I have nothing to compare to.

Intel CPUs only report estimated power consumption. This is not the same as actual power consumption. No one knows the accuracy of these estimates, especially when a CPU is idle. Trying to compare estimated data between two totally different CPUs is probably not a good idea.

 

[mkdr]

@unclewebb Yes. The bios of this Chuwi Herobox N100 is 100 % unlocked. I have basically access to everything, not something Ive seen before myself. There are millions of more settings I could change not shon on the screenshots in countless of sub menus. And I mostly shoudlnt change any of them without risking of bricking. The only option I wont have / see is undervolt lock. It seems to be missing sadly. I was hoping I could set it to disable and have undervolt access to the N100. I could set CFG lock disabled though to change the clock behavior for turbo boost through TS.

Weird, for my 8th gen 8750h cpu it also show system agent. System agent is just the chipset /PCH I think. Wonder if there is a way under Windows to debug and see what is blocking it going into low power. My theory is it could be the M2 SSD maybe not supporting low power mode.

Oh I am sure the values are accurate. If it shows 2w or 0.5w thats a huge difference for a fanless little box like this, causing a lot of not needed heat. I can see the same values on my XPS when it wont enter C8 for package / system agent.

 

[tmkhiem]

Where on the N100 "system agent power" never goes below 1.8w:

You're lucky you can get the power usage so low (your IA cores don't average over 2W). I tried tweaking many things but this is likely the best I can attain:

About N100 and C8, I have found some relevant links here, not sure if they're of any help:

• https://www.reddit.com/r/MiniPCs/comments/1975n2n/_/l4a2zr6 this guy claims he reached C8/C10 (not sure if that's package or core)
• https://forums.unraid.net/topic/147...0-or-i3-12100/?do=findComment&comment=1331319 he said his system "can reach Package C8 or C10 even", although it's a little blurry in the context if that's about N100, or 12100
• https://www.cnx-software.com/2024/0...son-features-benchmarks-price/#comment-617220 he said: "With lan, the CPU package goes only to C8" and the CPU is N100.

Apart from these links, I realized it's a weird reality that not many people seem to care about C8/C10 and N100. However, @unclewebb, if you need to test ThrottleStop on N100, I believe I can manage to run some tests on my N100 laptop.

 

[mkdr]

@tmkhiem no im not lucky, your OS isnt tweak properly it seems and more stuff is running on your Windows in the background. Didnt you notice "that guy" is me? I dont trust these comments with a screenshot. most people on the internet dont know what theyre talking about. no screenshot so far from an N100 showing C8. 2.5w is also the same I get with c3, so that speaks for not c8. like I said. my 8750h reaches 0.5w.

I also notice if you set windows power profile from balanced to low energy it gives another 0.3w . with balanced it is areound 2.5w with low energy around 2.0w. still too much, it should reach 0.5w. my guess is the culprit could be the SSD, it is not an NVMe SSD might be it is running all the time in no energy saving mode. I dont think the issue is the Intel Wifi adapter.

like I said, system agent is the issue here, and that is the bridge for PCIe and so on, something is blocking it. wonder if you can debug somehow on Windows whats blocking it.

system agent should in theory go down as low as 0.3w or so, thats the missing 2w here which is too much.

 

[ShrimpBrime]

You'll never visibly see C8 and higher because you'd have the PC in at least hibernate if not totally asleep. So no screen display.

Perhaps you'd catch the reading when coming out of sleep.

C-states in bios are not configurable on your particular board. Auto is what you get or disable obviously.

 

[mkdr]

@tmkhiem See thats what I meant with people dont know what theyre talking about : - ) And just spread nonsense over the internet.

Thats on mx XPS 15 9570 with display on obviously, pc on obviously.

And it can even reach C10 while being ON with display ON (4k resolution btw) but on battery, im just too lazy to make a screenshot because it is in docking station. Did you even read this thread? @ShrimpBrime Nope, you didnt.

I was so kind to make the proof for it right now, here we go : - )

You also dont seem to know what hibernate is : - ) That is total power off state same as shut down with zero power usage.

 

[ShrimpBrime]

@tmkhiem See thats what I meant with people dont know what theyre talking about : - )

View attachment 349448
Thats on mx XPS 15 9570 with display on obviously, pc on obviously.

And it can even reach C10 while being ON with display ON but on battery, im just too lazy to make a screenshot. Did you even read this thread? @ShrimpBrime

No, I don't need to read the thread to understand that the board may not even be equipped with C8 and higher in the first place based on past basic knowledge you can't always rely on software built more than a decade ago and not even tested on the particular hardware you're using and was even stated as such by the author himself.

What was it you where saying about not knowing what you're talking about???

Further more, you're not using a 12th gen N100 cpu. Your screen shut is null and void to the thread at that point.

Package C8 + display in PSR or powered, all VRs at PS4 + crystal clock off. The processor will enter Package C10 when: All IA cores in C10 + Processor Graphic cores in RC6. The platform components/devices allow proper LTR for entering Package C10.

Package C8. All IA cores in C8 or deeper. Display in PSR or powered off1. All VRs at PS4 . Crystal clock off.

 

[mkdr]

@ShrimpBrime : - )

 

[ShrimpBrime]

@ShrimpBrime : - )

Package C8 + display in PSR or powered, all VRs at PS4 + crystal clock off. The processor will enter Package C10 when: All IA cores in C10 + Processor Graphic cores in RC6. The platform components/devices allow proper LTR for entering Package C10.

Package C8. All IA cores in C8 or deeper. Display in PSR or powered off1. All VRs at PS4 . Crystal clock off.

Package C-States - 001 - ID:655258 | 12th Generation Intel® Core™ Processors Datasheet, Volume 1 of 2

 

[unclewebb]

Intel® 64 and IA-32 Architectures Software Developer's Manual
Volume 4: Model-specific Registers

Intel® 64 and IA-32 Architectures Software Developer Manuals

These manuals describe the architecture and programming environment of the Intel® 64 and IA-32 architectures.

www.intel.com

The Alder Lake-N series N100 have a CPUID Signature of 06_BEH. Intel's March 2024 publicly available documentation includes information for the 06_BAH, 06_B7H and the 06_BFH but there is no information available for the 06_BEH. Alder Lake-N was released in Q1 2023 but Intel has not yet released updated documentation for it.

The 4th Gen Haswell processors introduced counters to keep track of package C8, C9 and C10 residency data. The location of this data has been consistent from 4th Gen through 14th Gen for Intel desktop and mobile CPUs. Both ThrottleStop and HWiNFO seem to be reporting this data correctly.

Alder Lake-N is the odd ball. Intel must have changed the location or meaning of some of the C state counters and has not yet publicly documented what they did. Intel needs to update their documentation before ThrottleStop can be updated to correctly report C state activity of the Alder Lake-N processors.