Battery...how is it even possible?

[tomoqv]

With this topic I just want some elaboration about L5 battery performance compared to e.g. Android phones.

My partner has a Sony Xperia 10, which battery performance seems to be world class. The battery is 4500 mAh, i.e. the same as for the L5. On that battery, Sony Xperia 10 lasts for days if idle while my L5 consumes about 10-15% per hour idle. The Sony appears to endure at least a day with frequent use, whereas the L5 struggles to survive (in my case) 5 hours.

In the above cases, both phones have cellular on all the time. On my L5 I keep bluetooth off, and when I am outdoors I turn WiFi/BT off with the kill switch.

I understand that Android is much more optimized after years of development and likely it goes into some kind of hibernation to preserve battery to that extent, but still it seems to wake up almost instantly on power on or an incoming phone call. I just find it difficult to take in that it appears to be as much as 10x as power efficient than the L5.

This is not meant as a complaint, I just want some elaboration about the forces in play here and what we could expect of the L5 in a few years time.

[dcz]

The SoC including the CPU and all the heavyweight components on the Librem 5 was not made with extreme power conservation in mind. It was, however the only one we could get at a time where we could even think about competing. Getting one with better power would mean that we can’t use mainline Linux on it, and deliver the freedom we promise. Non-starter.

The SoC we use is a couple generations slower, made on older (was it 21nm?) silicon process, which is already power++. Then again, don’t underestimate the drivers, and the time and money our competitors poured in to driver development. Add applications, and Google’s data-sucking but power-saving Firebase, and we’ve got an uphill battle.

[tomoqv]

@dcz

Getting one with better power would mean that we can’t use mainline Linux on it, and deliver the freedom we promise. Non-starter.

There is no arguing about this, totally agree.

It would be interesting to know though, what the boundaries are and how much would be achievable through software optimizations. One thing that appears to be sucking a lot of power on my L5 (guessing a bit here) is the constant struggle that my phone seems to have to stay on mobile, especially when I am abroad roaming. I know it is early days for L5 hibernation, but that will help quite a bit I guess.

I am curious if we are hoping for 0-10% less power consumption on current hardware or eventually maybe in the range of 20-40%, which I guess would make the phone last up towards a day without a recharge. Respectful that there is no exact science here, but still…

[dcz]

It’s hard to tell: you have to weigh in the chances of dealbreakers like unexpectedly broken hardware (we did come across a few), and calculate the time it takes to solve the problems that can be solved 80% of the way there, versus the time to squeeze every last bit of juice.

I can say that I personally hope for 50% longer battery time with suspend and other optimizations, but don’t consider this an official position or guarantee.

[tomoqv]

@dcz

don’t consider this an official position or guarantee.

By no means, no, but your guestimate is probably better than mine

[bschubs]

I feel my android phone is not owned by me, but by Google. They have found a solution, that makes me carry their phone with their cameras and their microphone to every place I go, while I’m never sure things are turned off. And Google also lets me pay for their phone.

I can’t easily compile and install my kernel as I was used on other Linux systems. I can’t easily get a root account without loosing banking access. Most apps try to hide the directory system structure. For me android does not feel like Unix anymore.

I have the feeling, there are kernel modification, that make things run faster if android is installed. Maybe hardware modifications make cpus run
faster if android is recognized by a special access pattern. We know what is possible, if we look at VW’s diesel emissions scandal, where motors behaved different in special test scenarios.

At least Google does not document in a findable and reusable way, what has to be done to unleash the power of current phone hardware, to become more efficient, or which hardware should or should not be selected by phone builders for which reasons.

But free software will find and document this in future. Maybe not with our current librem5 hardware.

But said all that, for me a short running battery is much less important than a free phone.

And the reason why we have a short running battery is the hiding of knowledge by Google from my point of view.

[guru]

I use my L5 as my daily driver. At home I have a power supply, in my office another one, when I travel longer time, I have a power bank. In short, battery lifetime is not an issue for me.

Don’t get me wrong, I’d love longer battery time as well, but this is not my main concern.

[irvinewade]

@dcz

The SoC we use is a couple generations slower, made on older (was it 21nm?) silicon process

According to the FAQ it’s using a 28nm process.

The easiest win for a future version of the Librem 5 is a hypothetical drop-in replacement SoC from the same manufacturer that uses a smaller process node.

By contrast the Sony Xperia 10 has a SoC that is using a 14nm process (according to the internet). So ceteris paribus (hypothetically) this SoC has an approximate 2x advantage up front.

(Things never are equal though because often the smaller process is used to run the chip faster, thus giving away some of the saving in power consumption that was gained through the smaller process. Along the same lines, the smaller process may be used to pile on more transistors for better performance or more functionality etc., thus also giving away some of the power saving. In addition, sometimes - for various reasons - voltages are reduced, which reduces power consumption.)

@tomoqv

the forces in play here

Unless you have suspend-to-ram (STR) enabled in the Librem 5 then you are basically running the device continuously. However I think there are still gremlins with STR i.e. “experimental”. I couldn’t tell whether you are or are not using STR.

If you are not using STR then for an apples-to-apples comparison, you would need to keep the other phone awake all the time (other than the screen).

Another force at work is the separate, dedicated peripherals e.g. modem card and WiFi card. As they are separate in the Librem 5, their power saving modes need to be controlled separately and their power consumption optimisation needs to be tackled separately, and of course they consume power. By contrast, a typical iPhone or Android phone will have a SoC that includes much of the functionality of those cards (at the cost of course of removing the possibility of ever doing an in situ upgrade on those phones).

You will note discussion elsewhere in this forum about someone who replaced the WiFi card with one that is apparently more power-consumption friendly.

So, as you started out by saying, mainstream phones have an advantage in the time and money that they have spent so far on development.

The real comparison though is not between the Librem 5 and another phone but between the Librem 5 and what you need.

[dcz]

@irvinewade

SoC has an approximate 2x advantage up front.

Power consumption is not proportional to process node, though. I don’t know the exact relationship, but smaller processes have more leak current, or something like that? Anyway, most recent process nodes have plateaued in power despite getting smaller. So the “2x” advantage should be taken with a pile of salt.

[irvinewade]

@dcz

but smaller processes have more leak current, or something like that?

That is my understanding - and a reason to drop the voltage. Putting that aside, it isn’t an exact proportionality. It is complicated. Since 28nm is fairly expansive these days, maybe approximately proportional is fair. (If and when an appropriate i.MX is released, maybe the datasheet will give something more concrete.)

[Hristo]

@dcz

The SoC we use is a couple generations slower

As I wrote in another post earlier, Librem 5 is trying to kill too many flies with one stroke and this leads to many customer groups being semi-happy instead of one customer group being fully happy.
It is what Germans call the “eierlegende Wollmilchsau” - the Pig that lays eggs, gives milk, and gives wool.

There are around 4 different target groups that could have been 4 different products and for some of these different Products Purism could have opted for a SoC from Qualcomm or whatever instead of developing from scratch with an older generation CPU. That would have sacrified on some awesome bells and whistles but could have delivered better experience to some of the potential users.
Some people just want a working Linux smartphone Mint style as an alternative to Android and iOS. Not everybody wants to to be completely blob free. Some people just want it to work and to be a little bit more trustworthy than Android or iOS. Raspberry Pi is not blob free and is a total success. And maybe this could have been a better ground for attracting people to develop apps for mobile Linux.

Another approach would have been to go for an ever slower CPU and to not make Librem 5 a general purpose computer, but a reasonably secure mobile phone with reasonable battery life. If security is the main focus, then the user is not getting thousands of apps and features but just the bare minimum.

Librem 5 Fir Batch

I am just thinking for a longer time that maybe the main problem of Librem 5 is that there are several very different use cases and Librem 5 tries to satisfy way too many user groups at the same time leading to many people disappointed and to splitting the highly limited resources to work on too many topics at the same time. A 4G mobile phone that is more trustable than the mobile phones on the market. An Android phone where you have real control over the camera, microphone, and wireless commu…

[irvinewade]

@Hristo

eierlegende Wollmilchsau

“Challenge accepted”, says the genetic engineer.

[dcz]

@Hristo

Not everybody wants to to be completely blob free. Some people just want it to work and to be a little bit more trustworthy than Android or iOS.

And those will choose from one of a hundred projects like LineageOS, Jolla, or the freed Android flavor of the day. Purism going there would be just an also-ran, and I would not have joined it, seeing no value.

@Hristo

Raspberry Pi is not blob free and is a total success.

You mean: therefore, only a partial success. It’ll be total when it’s blob-free. But I’m not seeing any chance of that, so it’s a dead end for this purpose.

@Hristo

Another approach would have been to go for an ever slower CPU and to not make Librem 5 a general purpose computer, but a reasonably secure mobile phone with reasonable battery life.

That’s what the L5 is. We get the general purpose computer for free.

[Hristo]

@dcz

That’s what the L5 is. We get the general purpose computer for free.

This is a very interesting statement. I don’t think the users perceive Librem 5 like this.
This would mean as priorities:

1. MUST HAVE: Stable LTE modem connection (as the focus is then in being a phone and not a computer).
2. MUST HAVE: Reasonable battery life.
3. MUST HAVE: Possibilities for encrypted phone calls.
4. MUST HAVE: Possibilities for encrypted messaging.
5. Nice to have: Possibilities for encrypted messaging to popular messaging apps like Signal/Viber/Telegram.
6. Nice to have: E-Mail client (with PGP support working with the Smart Card).

In such scenario topics like browsing, music, photos, navigation, and other apps, are not a priority at all. Just a nice to have bonus. Even furthermore, maybe they are bloating the system and drain resources and battery.

@dcz

You mean: therefore, only a partial success.

Well, they sold over million units despite the best models being close to impossible to buy for a very long time, so I will subjectively count this as a success.

@dcz

And those will choose from one of a hundred projects like LineageOS, Jolla, or the freed Android flavor of the day. Purism going there would be just an also-ran, and I would not have joined it, seeing no value.

The market lacks an Android with kill switches to be honest.
One major problem with Android and iOS is that it is difficult to trust them to not track you/to be offline. A simple option to kill stuff like microphone, camera, WiFi/BT, LTE, GPS would be a benefit to Android users without bringing the issues of a new OS.

The AOSP flavors are not exactly the same as a a Linux project like PureOS, but you are right that there is Postmarket OS on the market that caters for this target to some degree.

So to sum it up. Maybe a very large portion of the Librem 5 users don’t use Librem 5 as it is meant to be as a product and then wonder about the results.

[tomoqv]

@Hristo

Another approach would have been to go for an ever slower CPU

What about underclocking the L5 CPU, is it possible (albeit maybe not advisable)? Could that be part of the hibernation or would it just make it slower to wake up from suspend?

[dcz]

@Hristo

This would mean as priorities:

I’m not the one who chose the priorities, but having a limited phone-only device limits the kind of people who would want to use it. I doubt that the relationship between features and users is linear, but on which side of linear it falls I can’t answer.

I do know that lots of people want photos, and I think not having browsing would cut the market so badly that I’m not even sure it would have been sustainable. Banking apps are already a problem. Like it or not, we’re building a mobile computer cause the goal is to reach the masses without dying on the way there.

@Hristo

One major problem with Android and iOS is that it is difficult to trust them to not track you/to be offline.

You don’t get that problem according to most of those projects.

@Hristo

maybe they are bloating the system and drain resources and battery.

When they aren’t running, no, they aren’t. If they are using a scarse resource, it’s developer time.

@Hristo

they sold over million units […], so I will subjectively count this as a success.

You can sell a million units of many things. I’m at Purism because I want to sell a million units of a blobless phone, and only then is it a success.

[Privacy2]

@dcz

You can sell a million units of many things. I’m at Purism because I want to sell a million units of a blobless phone, and only then is it a success.

Of course we are all aware that the Librem 5 does have blobs. The current cellular modem(s) requires a blob. Regardless of whether it is resident to the card, the Wifi card (Redpine; now owned by Silicon Labs) requires a blob. And although there hasn’t been much discussion about this lately, I think that the Librem 5 requires a blob to boot ( https://puri.sm/posts/librem5-solving-the-first-fsf-ryf-hurdle/ ; the blob is required and the point of the article is a technicality about whether the blob is isolated away from the main CPU).

For most people, the real point is whether or not (and to what degree) the blobs can be isolated and whether the blobs restrict the owner in regard to the FOSS components.

[dcz]

I never said the Librem 5 is a success by this measure yet. But, unlike the RPi, where the blob runs on the main CPU and Linux on a secondary processor, the L5 is pushing the blobby status quo, and has a good chance to make a mark in history as a stepping stone to a blob-free future.

[Privacy2]

@dcz

I never said the Librem 5 is a success by this measure yet. But, unlike the RPi, where the blob runs on the main CPU and Linux on a secondary processor, the L5 is pushing the blobby status quo, and has a good chance to make a mark in history as a stepping stone to a blob-free future

Great! I was just trying to clarify for others.

While the Raspberry Pi Foundation isn’t all that interested in being blobless, it’s worth knowing that there are efforts for blobless boots on the Raspberry Pi ( https://github.com/librerpi ).

Also, and I may be wrong on this — and given your work on millipixels you would know — but I was under the impression that libcamera started as FOSS effort on the Raspberry Pi to avoid blobs. The underground (non-RaspberryPiFoundation) efforts toward decreasing blobs on the Raspberry should be celebrated and should be recognized that this is related to the fact that there are millions of inexpensive Raspberry Pi devices that people want to Free up. The same could be said about a lot of improvements in FOSS video driver technology.

[Christal]

Hi Privacy2,

you have to think about that in iterations. Before Blobs, we had Algorithms or Drivers… we or some one else had to trust.

Now its about binary Blobs, or soc’s.

With the view by Users, its about running and using some thing with my Computer or Phone…

they do not care about an app or data, just about some working Program or Device. Which is not fine.

About Power usage, its some kind of the gold standard. We have to optimize Code or Power usage for our bereaves. Capitalism got that done with the cheapest price (do not ask how its done!) and some kind of product.

P.s. Sorry for my language issues. Keep on your good work.