Possible design changes to improve battery life

General Topics
1

There are many discussions already about battery life (e.g., Will there be a long battery life version?, [Battery Life] Impact of RAM / memory configuration + extra data). I think the consensus is that the ~5hr battery life is a major drawback for Framework laptops. This is true for me as well although I don’t yet have one: it’s the main reason why I can’t bring myself to click the “Continue to Checkout” button on the laptop I’ve configured. I’m raising a new issue to discuss how we could, perhaps through some design changes, improve the battery life.

Edit—the 5hr figure was taken from anecdotal reports from users, and may only be relevant for users of Linux that hasn’t been properly tuned; Windows users and some Linux users have claimed ~10hrs; nevertheless, battery life is a frequently discussed topic so I still think it’s worth brainstorming possible improvements

Some people have suggested getting an external USB-PD power bank to supplement the built-in battery. This is a good suggestion, but I’m curious what kinds of gains we can get through various design changes, because one reason I’d want a thin and light laptop is to not lug around extra weight. Note: I don’t do hardware design so forgive me if any suggestions are naive.

For instance, LPDDR4x memory uses ~half the power of DDR4, but (as I understand) it must be soldered on to get a reliable connection, which is somewhat antithetical to the promise of a repairable laptop. My question here is: how does the lower-voltage memory translate into real-world battery life improvements? Would it gain 15 minutes on an average load? 2 hours? If the latter, perhaps that’s a compromise on repairability that I could accept. Or could we have 1 soldered LPDDR4x stick with a second open DDR4 slot for expandability, and maybe the second could be off/low-power until it’s needed, or only on when the laptop is plugged in, etc.?

Some possible areas of improvement:

  • LPDDR memory (see above)
  • More efficient CPU (e.g., Intel Alder Lake, AMD, or ARM)
  • Underclocking or dynamic scaling of CPU (reduced wattage in CPU, reduced use of fan)
  • Underclocking memory during low-load (is this possible?)
  • OS-level performance tuning (I’m concerned about Linux, here)
  • Adaptable refresh rate of screen
  • Piezoelectric keyboard (only half-serious)

Does anybody know how the above translate into minutes of battery life, or at least to the overall power draw of the laptop? Ideally the laptop would last an average work day (8 hours) plus a little extra, so say 10 hours. So if the 55Wh battery is lasting ~5 hours on average load, it’s pulling 11w? So we’d have to get that to an average of ~6w for 10hr battery life?

I don’t expect we’ll soon see a new mainboard to accommodate other CPUs or memory configurations, but it might be useful to think about where to focus efforts.

4 Likes
2

First off, you can’t use lpdd4x and ddr4 on the same system

Personally I’m not against the use of lpddr4x as the failure rates are low and so long as replacement parts are available and schematics provided-this isn’t an impedance to repairability

Dynamic clocking of memory based upon load is part of the ddr5 improvement so it is possible but will require a new revision of the board

Undervolting is no longer possible with Tiger Lake, Intel locked that down, yes I think it’s stupid as it provides no benefit to users or protection

CPUs already race to idle at as low a power state as possible so that’s not a concern

Talking about adapting refresh rates? Thats basically promotion by Apple, good luck not getting sued-and it’s something that would need to be developed on Windows or Linux, not a firmware or hardware revision, beyond a display free sync capable

Edit: I’ve heard about significant power saving switching to the SK Hynix gold SSD

2 Likes
3

I’m not sure where you’re getting that number from. Speaking for myself, I’m getting 8-10 hours of casual web browsing, light coding, etc, which matches experiences of reviewers and other users.

The only folks I’ve seen report numbers in that range have been running Linux and haven’t successfully tuned the OS to improve performance.

5 Likes
4

Great, thanks. I was hoping someone with more knowledge would fill in these gaps. And, as mentioned, I agree that considering LPDDR4x (or similar) memory a non-repairable (or not-trivially-repairable) part of the mainboard might be acceptable. But I’d want to see the numbers (i.e., minutes gained, on average) before making such a compromise.

Do AMD/ARM block undervolting? Again, I’m thinking of changes that might require a new mainboard, not just tweaks to the current model. Although that would be an expensive upgrade down the road, so I’d want to have some estimate of the benefits.

So you’re saying the technique is Apple IP? Does “basically promotion” mean it doesn’t actually help? In any case, if it’s a software change instead of hardware/firmware, then I guess it’s not worth discussing.

Good. That’s something that we can do now. Thanks for sharing!

Maybe I’m only looking at the numbers for Linux as I’m a full-time Linux user. But that’s good to know you’re getting 8-10 hours. At least for a Linux user it’s good to know what’s possible with some tuning!

And, sorry, as a new user I could only provide 2 links. Here’s a comment by a user claiming to get up to 6 hours on idle, and 2-3 hours of moderate/heavy use. There are a few other threads here with similar numbers (for Linux users), and also some on the Reddit community (this post has some helpful information). Then again, nrp claimed here to get 10.5 hours on one benchmark (presumably on Windows). There are some threads here about Linux tuning, so maybe I should consider those when estimating battery life under Linux.

5

@Michael_Goodman

Ditto, and I am doing video and audio editing as well. On windows… And am very happy with the laptop in all ways.

1 Like
6

@ImaxinarDM thanks for another data point! I’m feeling a bit more encouraged that 5hrs is not a hardware limitation and that it should be possible to do much better on Linux, too.

1 Like
7

To clarify, I’m running Debian testing full-time on my machine. I’m running the 5.14 kernel using an old wifi card to avoid the AX210 issues, which means I’m able to run with PSR enabled on the display. I’m using tlp with some tuning (I’ve posted details elsewhere so I won’t repeat them here).

On a full charge powertop reports an estimated 12-14 hours idle with the display brightness turned all the way down, and 8-ish hours under more realistic conditions.

1 Like
8

I certainly hope Linux gets better at it.

Considering how much Windows is a bloated steaming pile… I just use it & put up with it because I have to. But on some things it is better - for now.

9

Let me clarify that, I did very poorly texting out an answer

It’s more accurately called ProMotion or Pro-Motion than promotion

I am certain that it mitigates high refresh rates consuming greater battery but given that it’s likely a patented technique by Apple, yes I think an attempt to duplicate it would result in litigation and it does require certain display technology, it isn’t just software

AMD doesn’t at the moment but I don’t know about how you would even do such a thing on ARM

It’s entirely possible Intel will reverse their position and restore that capability with Alder Lake

The problem is that it’s variable per user, much like overclocking, it relies upon the silicon lottery, some samples will tolerate lower voltages than others

I think ultimately a new revision of the board is required to implement hardware level changes that can increase battery life

For intel at least, Alder Lake’s efficiency cores are supposed to have the power of a full desktop Skylake core at I think 1/4 the power draw? I’m eagerly awaiting benchmarks to test the efficiency of Alder Lake

DDR5 is looking more important than ever, I had a poll a while back gushing user interest in a switch to DDR5 but I think it will be necessary

I don’t know exact or roundabout figures for failure rates but I bet @Louis_Rossmann does

1 Like
10

I had to dig around but I think you were referring to this article? Debian on Framework | The “B” Ark

11

That’s the one! I should go back and update it, as I did make a change to switch to the schedutil cpu governor, but other than that, that post reflects my current setup.

2 Likes
12

Microsoft is working on it, fortunately. The feature Microsoft is working on is not exactly the same as ProMotion, but it would reduce power consumption. No FreeSync required, but the hardware would need to support DRR.

I really want this to be true, but I can’t find more info about it. Could you point me to a source?

I’m using this and would recommend it. It seems to run pretty cool.

1 Like
13

Unless I misheard the info, you can find it on a video Linus did just recently on DDR5

Edit: I think I did mishear, I re watched the video and it’s never mentioned

So I have no idea where I came up with that idea

2 Likes
14

Knipsel
Knipsel1506×1165 118 KB

I have my Framework laptop now for 4 months and my battery life is now 4,5, hours on a full charge, I’m using power saver mode(at 100% already) and I have my screen brightness on just 30%. This does not seem quite normal to me. above you can see what I’m running, which shouldn’t be so harsh on the battery.
Do other people have the same problem?

15

I recall that while power saver mode does carry out some power saving changes…it wasn’t enough for my case. What I also had to do was to limit the PL1 and PL2 TDP to something like 7 watts. This 7 watts was high enough for me to watch videos, do some emailing…with no noticeable degradation to the perceived responsiveness, and yet, low enough that will allow for longer battery runtime.

Note that changing TDP value does not necessarily slow down ‘all’ task execution…but rather, it depends on the instructions that the task is trying to execution. Some instructions are less / more power hungry.

1 Like
16

I don t yet have the fw, reading the forum

1 having modules that don t draw 1,5watt hours just sitting there

2stop providing western digital ssd that are know rlto draw twice the power idle…
3 maybe experiment with turning off turvoboost on 12th gen.

The review of fw gives 7hours in benchmark. And usually its lower 1 to 2hours for irs class.
Undervolt isntemoved from 12th gen because of a security issue. Anyhow the cpu adapts its voltage depending on the frequency it is…call it speed step shift boost.

17

Hey @Second_Coming is there any source you can provide me how to do this? I’m not that experienced into changing these kind of settings.

18

Can I add something random like changing the tapered edge of the Framework base to

PXL_20220831_033718814
PXL_20220831_0337188141000×750 136 KB

something like the macbook pro flat shape

2021-macbook-pro-ports
2021-macbook-pro-ports1952×1156 182 KB

Or even the same contour as how the back of the Framework base where the hinge is.

I would think the sides would need to have the angled and tapered for the speakers to work.

Not sure if this will give more volumetric space to fit a slightly bigger battery.

1 Like
19

Here:

  • Ignore anything around undervolting. (It’s not possible on 11th and 12th gen Framework laptops (limitation brought on by Intel))
1 Like