Battery life tests with RK3588

MNT Pocket Reform
1

I ran a first battery test tonight after installing the RK3588 module with the device just idling and running btop (and background stuff like syncthing but with no meaningful activity). Just booted with a fully charged battery, launched btop and let it sit. Taking screenshots every once in a while.

screenshot-2025-05-14-01-00-40
screenshot-2025-05-14-01-00-401920×1200 268 KB

Uptime was 4:04h.

Power usage was oscillating between 3.7 and 7.5W the whole time, mostly staying at around 4.3W

Temperature was pretty stable at 45-50°C. Much cooler than the i.MX8M+ which was usually at 75°C the whole time.

Shortly after this one the Pocket rebooted and went into a weird loop of trying to mount the SSD and failing at that so I plugged it in again (it was fine after that).

Tomorrow I’ll see how it performs in real-world use at university. And when I find time I’ll also do one test of just letting a long youtube video play.

9 Likes
2

How’d you measure the power usage? Is there a btop option for it, or…?

3

Btop displays that on the top right next to the battery percentage bar

1 Like
4

Thanks for sharing. I had similar results (just had irssi instead of btop), maybe closer to 5 hours, with scaling governor set to powersave. What was the screen brightness in your tests?

5

Screen brightness was around 30% as I did this indoors at night.

I did not set any particular power saving though.

1 Like
6

How do you actually do that?

7

You’ll need linux-cpupower package and then just run sudo cpupower frequency-set -g powersave.

1 Like
8

I wrote down some notes in this for reference:

9

I also wrote a little helper script to automatically switch governors depending on what the current powersource is:

1 Like
10

D’oh! Thank you! I had just missed that entirely.

11

Today I tested battery life on RK3588 while playing a 1080p fullscreen video on Youtube in Firefox, connected to a close-by access point with strong signal.

IMG_5790
IMG_57901920×2446 396 KB

CPU governor was set to powersave, screen brightness was at around 30%. Audio was turned off (muted) system-wide.

The device rebooted with 0% battery on the OLED display after 3h07m25s.

7 Likes
12

About temperature:

The RK3588 is pretty consistently running 20-30°C cooler than the i.MX8M+.

I used to have temperatures around 75°C constantly and independently of actual load, now it’s between 45° (idle) and 55°C (video playback).

Makes the Pocket much more comfortable to handle and I’m less worried about having in my bag for a moment while it’s running.

6 Likes
13

It’s been a few days of regular use now, mostly at university reading papers, writing, email, and web browsing (with some pretty heavy web apps like WhatsApp) and for writing on my couch.

This is of course highly subjective but I think I’m getting a little more battery life out of the RK3588 than I did with the i.MX8M+, maybe half an hour more for a total of about 4h, which makes sense to me given the lower heat output. After all, computers are electric heaters that happen to do math, only some to more math for any given amount of heat produced.

I carry a USB powerbank on most days anyway since I would need far more battery life to get through the hole day with no charging, but I need to use it less often now.

I was worried that the RK3588 would drain the battery faster but that’s absolutely not the case. It’s faster AND uses less power. That’s really cool.

11 Likes
14

This is a really encouraging thread that has me really looking forward to my RK3588 upgrade as well. Still looking to see if we can get a bluetooth solution as well but I’m not worried. The RK is really going to make the Pocket a little wonder.

2 Likes
15

Hi,

now that I also got a Pocket with rk3588 I also did some armchair “benchmarks”. I only tested each scenario twice (to weed out bad systematic errors and i kept the better of both results) and my results are in minutes not seconds. Unless somebody convinces me of the utility of it, I am not planning to perform more thorough tests because your actual runtime will vary wildly depending on how you use the unit, whether you have wifi on and how much you transfer, how much you utilize your SSD, your display brightness, keyboard backlight on or off etc…

I wanted to check the default installation so my test setup involves the latest system image tag sysimage-v5-20250603 loaded on an SSD and using the new default desktop: gnome. The only changes I did to the configuration were setting these things in /etc/UPower/UPower.conf:

AllowRiskyCriticalPowerAction=true
CriticalPowerAction=Ignore

Without the above, gnome (via upower) will shut down the system when the sysctl thinks that the battery is at 0% which might not necessarily mean it’s really empty. I wanted the system to run until the battery protection boards cut the power and the system switches off hard (it actually reboots and fails to boot because the batteries are flat).

1. Fullscreen YouTube with full display brightness

I played this fireplace video which is the same video that Jeff Geerling used in his review of the classic reform and tried to do the same test: play the video in fullscreen in firefox while connected to the internet via wifi. According to the YouTube interface, the video was displayed in 720p@30. It lasts less long than Esther’s probably because I had my display on full brightness. Runtime: 02:26 h

2. Actually watching a movie

Above test is to replicate a similar setup in the review by Jeff. But personally I would not watch videos via the YouTube web interface. Instead lets check the scenario where you just want to watch a movie in 1080p@60 h264 and to do so you use clapper because it can use the system’s hardware acceleration. Keyboard backlight is off as it would distract from the movie in a dark room. Note, that compared to the previous test the machine lasts longer despite the video quality being better both in resolution as well as in framerate. Runtime: 02:47 h

3. Trying to conserve as much energy as possible during use

I wanted to find out how long I can actually use the Pocket while trying to be as conservative with my energy use as possible. Scenario: I’m on a trip and there are no power plugs in the train but I still want to be a bit productive. So I reduced the display backlight to minimum and turned off the keyboard backlight. Then I opened a terminal and ran htop to simulate some very light editor workload like writing documentation or reading text files. Runtime: 4:56 h

4. I forgot my Pocket in the pocket

The Pocket cannot (yet) suspend so to not loose my work I put it into my pocket and it can happen that it gets forgotten there. How long will it survive? This test has the screen and keyboard backlight switched off and the gnome desktop is just idling (why is this gnome-shell process continuously consuming 4% CPU doing nothing??). Runtime: 05:23 h

5. Maximum CPU stress

How long does the battery last if I try to drain it? The stress utility keeps the system busy (I used stress --cpu 8 --io 4 --vm 2 --vm-bytes 128 no SSD is installed so no --hdd), display is on full brightness and keyboard backlight is on. Runtime: 02:04 h

6. sway

Okay, who am I kidding, I’m not going to use gnome. :grin: So lets perform test 3 above but with sway instead of gnome (same terminal emulator). Runtime: 4:59

Conclusion

I was surprised to see how flexible the RK3588 is. If you need it, it gives you high performance at the cost of battery lifez. If you don’t need the performance though, you easily get twice the battery runtime.

I want to do a more thorough test comparing the different SoMs against each other in the classic Reform. If you have ideas about specific scenarios you’d like me to test, please write me a PM or reach me via josch@debian.org (don’t hijack this thread which is about RK3588 Pocket battery life), thanks!

4 Likes
16

Does your Pocket have an SSD in it? Mine does and I’m sure that affects battery life too.

17

yes it does. The OS was running off the sd card. The test matrix is indeed huge.

1 Like