Tiger Lake - Idle - CPU Package Power too high (the higher the display resolution - the worse)

[extremecarver]

Can some people with tiger lake post their idle package power on Tiger Lake including their display resolution. Integrated Graphic cards only please.

I'm really unsatisfied with the 1165g7 power consumption on light loads and idle. While i5-8250U with applying some power saving stuff like in the guides idles at 0.3-0.6w in my Lenovo T480s with WQHD display, my lg gram 16z90p uses 1w minimum, but usually hovers around 1.5w idle package power. For simple office work, notepad++ and microsoft edge (chrome is even worse) CPU package power averages around 3-5w. Ridiculous vs my older Kaby Lake R which managed such stuff around 2-3w package power.

The only place where tiger lake is a huge improvement is video playback fullscreen. This uses about 2w for FHD to about 4w for 4k AV1 on youtube package power. HEVC264 does seem a bit more power saving than AV1 but AV1 is okay considering even 8k will playback on 2560x1600 display without getting over 4-5w package power.


Up to the latest preview windows cumulative update (3 days ago) I could not activate panel self refresh without getting constant black flashes (up to 2/3 of time screen black) - now this magically seems fixed - but the power draw does not seem to be improved at all. On Kaby lake R deactivating panel self refresh would also increase idle package power to 1-2w instead 0.3-0.6w. So maybe there is some fix in windows to detect broken hardware and not enable panel self refresh for those cases. I cannot find any way to decrease the package power. Core parking and so on didn't seem to improve it at all. Speedshift values over 180 also increase real world power use. 180 seems to be like a sweetspot.


The only real solution to drastically get power use down is decreasing resolution. At FHD I get 0.8-1.5w package power at idle. At 1280x800 I get 0.5-1w (so actually about 40-50% decrease). I would have though panel self refresh should mean resolution in idle doesn't matter much anymore.


I did have the very occassional times that the package power would drop to 0.5w - then I had more black screens. Hasn't

oh yeah - it's usually enough to move the mouse or type anything to get package power up to 3w. Scrolling fast 4-5w. And switching windows instant peaks around 10w. Cray ineffecient vs my older laptop (that didn't have IGZO screen like the lg gram definitely has, plus much smaller battery - but could last nearly as long because CPU didn't burn the battery).

 

[unclewebb]

package power

The package power data that Intel CPUs report is not measured power consumption. It is a number calculated within the CPU which is only an estimation of power consumption. The accuracy of this number is not known and can vary from one CPU to the next and from one generation to the next.

The only purpose of this data is to control the turbo boost function when the CPU is loaded. Whether this data accurately represents power consumption when a CPU is mostly idle is not important to Intel. This data was never intended to be used to compare light load power consumption between different CPU generations.

Here is a 10 core CPU running at 5000 MHz and only consuming 1.1 Watts.
This data is meaningless when a CPU is idle.

What you need to look at is C state residency time. This data is extremely accurate and you can use C state data to compare different CPUs. One poorly written driver or bloated program running in the background can prevent CPU cores from entering the low power C states. This is how modern CPUs save power. Are your cores spending over 99% of their idle time in core C7? Are the deeper package C states being used correctly? Often times one bad driver will disable the deep package C states and power consumption will go way up. Your mobile CPU should be using package C8 when idle.

Speed Shift values over 180 also increase real world power use.

That is true. A slow CPU is an inefficient CPU. The whole point of Speed Shift Technology is to get the CPU up to full speed as fast as possible so it can process background tasks as fast as possible. This allows the CPU to spend a bigger percentage of time in the low power core C7 state.

 

[extremecarver]

Yes the c-states are okay. Actually maybe a bit improved after installing Samsung NVME driver (PM981a Samsung NVME, which is a 970 Evo Plus more or less I think).
Actually I think if I don't move the mouse or anything the power is okay - Overall battery use according to batterviewinfo then drops to 1.8w with screen at lowest brighthness (and that is at least 70-80 nits, pretty bright compared to most laptops at lowest setting) or around 4.5w at brightness.
The problem is as soon as I type a bit in edge or notepad++, switch a windows or anything I have 5-7.5w overall consumption (instead of 2). Also it will first idle at 3w for 1-2minutes before dropping back to 1.8w-2w, cannot observe any change in C-states. But that 1w difference is noticeable in both throttlestop power package as well as batterviewinfo.

In comparison my T480s never goes below 3w overall consumption with dimmest displaz. However for surfing / writing text it only uses 5-6w overall. Also the screen not being igzo needs much more power for usable brightness. At dimmest setting it's already hard to see in really dark room. So while the T480s on the i5-8250U only needs around 2-3w additional for the workload. The tiger lake needs 3-5;5w for the same work (yes not video - when it comes to video tiger lake leaves the Kaby lake R in the dust, as it has much better hardware decoding and is really efficient there). So yes - I believe you the power package isn't too important, however the actual battery consumption shows that the tiger lake is way less efficient for my light work. Surfing with Chrome was even worse - the Kaby lake makes not much of a difference if Chrome or Edge, the tiger lake would consume around 8-10w with Chrome, slashing the battery.

To me it seems that the lg gram 16 with tiger lake has a really efficient platform, just the CPU hinders it from reaching Macbook M1 comparable battery life (well for video it's actually better thanks to the 80wh battery).


I also tried for 20 minutes just surfing with 1280x800 screen resolution, and power consumption was dramatically better. On the Kaby Lake R this change was barely noticeable (1280x720)

 

[unclewebb]

Yes the c-states are okay.

What is your definition of okay? When idle, are your CPU cores spending 99% of their time in C7 or are the cores only averaging 75% in C7? Is package C8 being used when idle?

I had a laptop once that had a poorly written track pad driver. Simply placing a finger on the track pad was enough to send CPU usage sky rocketing on the first thread. This is how programmers used to write drivers back in the 1990s. The track pad was being sampled hundreds of millions of times per second in an endless loop. It took a long time before I was able to come up with a driver that did not have this issue. You might have a similar issue with the keyboard driver.

Do you have any new Chrome plugins or do you let Chrome run constantly in the background? These features can suck power.

 

[extremecarver]

yeah - quite a few chrome plugins. But then the chrome plugins are identical in edge, and identical on both laptops. Just typing or moving the mouse is kinda okay, the worst is if I switch between two windows.
Actually it's about 50% of the time in C10, a lot in C8, and a bit in C2.
I have about 96-98% in C7 or bigger. The screenshot is using sniping tool which will ramp up tiny bit just before taking it. Also the CPU runs quite a bit hotter than the one in my T480s - and needs fan more often (yes no undervolt possible of course) - however than the fan on the LG gram is not good at all. It can only cool about 22-23watts vs maybe 35 on my T480s (I say maybe because the T480s is locked to 28W, and cannot consume more, but then only reaches 70-75 degrees at full fan speed).

 

[unclewebb]

Great to see that C9 and C10 are actually being used. Manufacturers enable those two C states so rarely that I was beginning to wonder if ThrottleStop was reporting those correctly. They look OK.

the CPU runs quite a bit hotter

Manufacturers are getting really cheap with their heatsinks. They seem to think that consumers are OK with lots of fan noise and sky high temperatures up into the 90°C range. I love my old dinosaur laptop. It is silent when lightly loaded and I can use it on my lap without it feeling like I am trying to hold on to an Easy Bake oven.

no undervolt possible

That is a big problem. Intel's default voltage curves are usually way too high. I read a review the other day of an 11400 and they were able to reduce the CPU voltage by -0.3V. This made a huge difference and dropped power consumption almost in half. The CPU still ran 100% reliably so there was no need for all that extra voltage. With Intel blocking off CPU voltage control on their 11th Gen U and G7 series, we will never know what could have been possible.

Do you really need to check the Speed Shift EPP box in ThrottleStop? I usually just let Windows 10 manage this. The highest EPP value that Windows uses is something like 154 depending on the laptop and how the power slider in the system tray is set. I do not think there is a huge difference between that and using ThrottleStop to set EPP to 180. You might end up with a conflict where Windows and ThrottleStop are writing different values to the same register. Watch the Speed Shift EPP value in the FIVR monitoring table to make sure this issue is not going on.

 

[extremecarver]

thanks for the note about speedshift. Well I do notice if I set lower values it runs a lot faster - and on battery I like to limit it a bit because tiger lake is inefficient at high speeds. It runs hot and doesn't work much faster... Singlecore is okay up to 4400 mhz or so, all core turbo just should never run above 3400mhz due too the limited cooling. On AC unlimited single core and also 3400 mhz for all core turbo will provide best results as otherwise it's instant prochoot...

Actually speedshift does work for sure - because at 180 surfing the cpu temps are about 3-4 degrees lower than with it disabled.. So I guess it also saves a bit of power. CPU temp maybe the best indicator besides looking at batterviewinfo stats.

The lg gram went for super silent and inefficient cooling. But yes it burns your lap and is not efficient as the hot air is removed to the back. Depending on the screen angle the exhaust can be fully blocked - especially if on lap as fully open screen = the air exit is blocked. Intake is from below... Really a bit stupid design - but likely much easier and leaves more space for ports vs side exhaust. And yes they went for light is everything...

oh yes - I'm using the standard/default microsoft keyboard and mouse drivers. LG doesn't offer any separate drivers and I'm clueless who is the manufacturer of the trackpad. Cannot find any name that I can do something with the hwinfo or device manager (so maybe neither Elan or Synaptic). So I don't think I have a way to install a separate driver.
For the trackpad it resorts to microsoft precision drivers as I have all those options.

Oh but you are correct about windows overwriting my speedshift values according to FIVR. strangely though in FIVR the values from windows are shown, the value I enter in Throttlestop is actually used as is clear when running any benchmark or demanding task.

And about C9 and C10 state - It is only used if panel self refresh is activated! With it inactive the max state is C8. Strangely though my laptop often flashes black when using panel self refresh - sometimes as bad as half of the time screen off. I do not know however what causes this problem. Yesterday it never happened, today on boot first thing I had to do was deactivate it because not usable. Now playing around with Throttlestop values and activating it was fine again.

Actually I think maybe C9 is broken - it would be great if I find a way to disable C9 while keeping C10 (if just for debugging) without disabling panel self refresh of course which also disables C10...

Edit again> well comes another reboot - and C9 and C10 are never used plus power package and heat definitely higher - it seems like luck whether or not the C10 is used, and how much power this laptop draws.... I'm not sure this is a software or hardware problem...

Edit: after working for 4 hours without the C9 and C10 states because they did not activate. 30% battery in 4 hours - or 7.5% per hour (so about 6w) with 80% screen brightness. That was of course with some minutes away from the laptop, but never screen switching off. With C9 and C10 active I guess it would have been 0.5w lower per hour. I still have no clue what prevents this state or makes the CPU go into the C10 state (as C9 is rarely used).

 

[extremecarver]

I think I know now about C10. It can only be activated by deactivating panel self refresh - then reactivating panel self refresh. How long it stays active I haven't really found out. I know if I suspend or hibernate the laptop - then afterwards I need to deactivate/activate panel self refresh to activate C10 again. Only with C10 active I can get low idle power.
I will have to find a way how to deactivate/activate panel self refresh without opening intel graphic center. With C10 activate - the laptop runs much colder in general. Even though it will take about 2-3 seconds after any interaction with mouse or keyboard to get into C10 instead of C8.

I seem to have the same problem as reported here - and it is also about 1w difference measured with batterviewinfo. Idling at 2w vs 3w with lowest screen brightness, or 3w vs 4w with 80% screen brightness.
www.reddit.com/r/Dell/comments/eb6vad/xps_7390_2in1_wont_got_to_c10_state_anymore_after/