Will Linux phones get decent processors in the future?

As someone who attempted to use my Librem 5 for my primary personal cell phone for about two months, I can tell you from actual experience that it felt significantly slower than working from a normie phone. Gnome Web started quicker than Firefox but wasn’t as compatible with websites. I think Firefox was compatible with all websites I went to but was extremely slow to start and somewhat slow to run. I don’t know if further optimization could make it start and run faster, but it seems like a more powerful (and perhaps cooler) CPU would have run Firefox better.

6 Likes

Thank you, Amosbatto for the good analysis.

Has anyone tried VNC from Librem 5 to a desktop computer and how was the performance?

Marketability, for one.

While you or I may not need a fast CPU or jigawatts of RAM on a phone, I saw a lot of responses to the L5 from mainstream buyers along the lines of “Not enough RAM for a phone in 2020” or “CPU is too slow”.

Whether or not they even need the horsepower is another thing too, but bigger numbers go a long way towards making people buy stuff in my opinion.

Not me (no phone yet) but I can’t wait to ssh -X from my PC to my phone and have Chatty on my desktop through my phone.

I mean there are lot cool features that Librem 5 still need enabled just better than more cpus onto L5, increase cpus mean more Hot y Battery hungry.
With current L5 CPUs we can have a faster convergence if we put air cooler to L5 when L5 is running on convergence.

Like Librem 5 need a dedicated LIBREM-5-DOCK to keep cooler the device to ingrease cpu speed on Convergence. I making one.
L5 it good designed to pasive sink hot, but still need vent on.

2 Likes

But why do users need access to their software? It’s not like they’d know how to use it or have use for it. That is what your post sounds like to me. It is a very Apple mindset. The same mindset steve jobs had.
To counter, why would I need a server to remote into when I could have more than enough processing power on the go in my pocket. As someone whos generally doing IT support related tasks in the field remoting into a server at my house is useless when I need to fix a computer that won’t boot. This phone has the ability to replace my entire keychain of random recovery drives I am forced to lug around. But according to people like you we don’t need a more powerful CPU and it’s okay that it will take longer to complete a task.

Why the future?

One the great benefits of linux was that it allowed folks to refurbish old computers, speed them up and give them new life after windoze bloat. Too bad there isn’t a medium to flash old 'droids with a linux O/S.

Having used both the L5USA and PinePhone, I do think that their processors are limiting the phones. With the PinePhone, the RAM and eMMC is very slow and it takes a long time to load apps. (The slow RAM is due to the A64 and I suspect that PINE64 hobbled the interface speed to the eMMC because they werent sure that the A64 could handle a faster eMMC.) It takes about 10 seconds just to load the configuration app (GNOME settings) in Phosh. I really notice how sluggish the interface is even after the apps are loaded.

With the L5USA, it loads apps in about half the time as the PinePhone, but my $175 Android phone with a Snapdragon 660 loads apps in about a third of the time that it takes the L5USA and the interface is much snappier. Some of this difference can be solved with software optimization. When GTK 4 gets hardware acceleration from the GPU that should help speed up the interface, and image encoding can be made faster, so it doesn’t take so long to snap a photo.

However, the bigger problem in my opinion is the energy inefficiency of the processor. Because of the aluminum frame, you can feel the heat from the processor in the L5 and it also shortens the battery life, which is the biggest limitation of the L5 in my opinion.

The Purism devs will eventually figure out the power management issues, so the L5 only needs to be charged once per day, but the lack of a competitive processor still effects the ability to market Linux phones. For most people, their phone is their camera/camcorder, and that requires a decent processor. Linux phones’ killer feature at this point is convergence, and it is hard to market that feature with the current processors. When a phone costs $150, people will buy it as an experiment and not worry if it gets stuck in a drawer, but when it costs $1299, they need to be able to justify that expenditure, and it is hard to do that when their Android/iOS phone has a processor that is 10 times faster.

6 Likes

I know a lot of effort went into adding support for the m1 chip in linux with no help from apple. How hard would it be to have a similar effort for one of the snapdragon processors?

edit: Also mobian had a build for the oneplus 6. So there appears to be some effort on the processor that runs

High specs are also contributing towards virtualization options. And as we all know sandboxing and virtualization is the mother of security. So for a security focused phone it is not a bad thing to be able to run VMs.

1 Like

6T on Mobian is taking ages to charge and battery discharges within 3 hours just by browsing the internet. Mobian needs optimization for Oneplus and nowhere near ready to be a daily driver.

2 Likes

Have to admit though, Gnome apps and FF on 6T running smooth, some do not scale automatically but there’s a workaround for that. I doubt there’s continuous support and commits for 6T to last on Mobian, hoping to be proven wrong soon-ish and to see devs get the 6T work horse becoming an affordable alternative. However it still remains a de-googled smartphone, no mainline so it isn’t fair to do a comparison against L5.

Fullay agreed amosbatto. The mx8 might have decent enough performance for me, but it was never designed to be energy efficient. And it shows. So we can have fast, short-run time and hot like the mx8 or the rockchips (and I am doubtful that suspend will be a silver bullet), or ultra slow like the pinephone.

3 Likes

Let’s hope that happens. Apple, Samsung and Huawei only make chips for their own devices, and I doubt that they are interested in making Linux phones. That leaves Qualcomm, MediaTek and UNISOC. Of those three, Qualcomm has the best history of being willing to work with the Linux community by publishing the source code of its Android kernels on the Aurora Forum, but it doesn’t share any info about its hardware without making people sign an NDA. MediaTek and UNISOC routinely violate the GPL 2, because they don’t publish the source code for their Android kernels. UNISOC is also partially owned by the Chinese government, so I doubt that a company like Purism would even want to use its processors.

The global smartphone market is about 1.4 billion units per year, so it is easy to see why the chip companies aren’t interested in the tiny Linux phone market. We basically have to hope that companies that make tablets, automotive head up displays (HUDs) and drive processors, TV setup boxes, object and voice recognition devices, etc. will convince Qualcomm or MediaTek to support Linux, because they have the kind of volume to convince these companies. With the advent of electric vehicles and autonomous vehicles, I think we have an opportunity, because they need HUDs and drive processors, which are both powerful, but also energy efficient.

Historically, it has taken a couple years after a Snapdragon release to get support for the chip to be added to mainline Linux. If a company was dedicated to the task, it probably could happen much much faster. It is a risk for Linux phone makers, because if they wait until there is good Linux support, Qualcomm may stop producing the chip, which is what happened to the Snapdragon 800 that F(x)tec wanted to use in the X1 Pro, and they had to switch to another Snapdragon chip.

From what I understand, the Linux support for the M1 happened so fast, because Corellium (a company who specializes with ARM device virtualization) spent years studying how the A-series processors work, and they applied that to the M1. If Apple didn’t block the bootloader for the A-series processors, it would be possible to run Linux on iPhones and iPads.

4 Likes

There is preliminary work to get Linux on the Shift6mq (Qualcomm® Snapdragon 845), at least for the postmarketOS distro. So there will be more powerful options in the future, even if they don’t have kill switches

3 Likes

… and might require blobs?

1 Like

It has been possible for a number of years to install Ubuntu Touch, Sailfish OS and postmarketOS on select Android phone models. If they use a Linux kernel, they usually don’t fully support the hardware, as is the case with the Shift6mq. If they use an Android kernel and drivers with libhybris, there is a better chance that all the hardware can be supported, but the makers of mobile processors usually only release one Android kernel for each chip and never upgrade it.

In the case of the SHIFT6mq, its Snapdragon 845 processor was announced by Qualcomm in December 2017 and the first phones shipped with the processor in Feb. 2018 with the Linux/Android 4.9 kernel. Qualcomm has never upgraded the kernel, and will stop supplying security updates in Jan. 2023, so the SHIFT6mq will have an effective lifespan of 2.5 years (Jun 2020 - Jan 2023).

Initial support for the Snapdragon 845 was added to mainline Linux 4.18 in June 2018, but 4 years later there is still no support for the GPS and suspend and the cellular modem is only partially supported. If Shiftphone had decided to release the phone with Linux, it would have used the Android drivers+libhybris, because figuring out how to get the GPS, suspend and cellular modem to work with standard Linux would probably take years of work. Therefore, Shiftphone would have been stuck with Android drivers which would have only been supported for 2.5 years.

We need the chip makers like Qualcomm to support Linux, because phone makers aren’t going to ship Linux phones without support from the chip makers. Yes, we can install postmarketOS on 11 different phones with the Snapdragon 845 processor, but none of them are functional phones, so it is meaningless unless you want a portable WiFi device without suspend and GPS.

My hope is that someone will build a Linux phone that has a Snapdragon processor with mainline Linux support, and add separate cellular modem/GNSS and WiFi/Bluetooth chips.

4 Likes

What are your thoughts on this (nonexisting) thing?
https://libre-soc.org/

There are three key issues at hand here. One is performance, one is thermals, the third is battery life, in use but also in idle (without suspend).
Performance can be approximated by the number and kind of CPU and GPU cores and the memory bandwidth the SoC can work with. Regarding the other two, node size is often seen as the holy grail – which it may not be:

While the AllWinner A64 (slow cores, poor GPU, poor memory bandwith), is made in 40 nm, the TI OMAP 4430 (in the mainline supported Droid4) is made in an at first glimpse worse 45 nm node, and has comparatively amazing battery life—it was designed for phones. It also runs cool, which matters a lot for phone designs.

With NXP iMX8M, AllWInner A64, Amlogic S922X, Rockchip RK3399 and also the RK3566 and RK3588 we have chips that weren’t designed for phones. I’ve seen reports that the Rockchip RK3566 is supposed to run cool without a fan or heatsink, the same applies to the StarFive JH7110 BTW, which both might land in future PINE64 phone hardware (I just hope that PINE64 will not name their RISC-V powered devices PineBook/PinePhone, because otherwise support threads will be hell).

The best chip with decent mainline Linux support we currently have is the Snapdragon 845; the best supported devices right now are the Xiaomi PocoPhone F1 and the OnePlus 6 (with postmarketOS; the latter seems much nicer, but is harmed by OnePluses proprietary charging crap, which requires you to put the device into EDL mode to charge at somewhat decent speeds). I expect the SHIFT6mq to overtake these two soon, but at least according to the postmarketOS wiki it’s not the best supported device yet.

All that said, I don’t really see massive issues on the hardware side. The PinePhone can work as a daily driver, if only the software is light enough—a surprisingly large number of “daily driving” users use Sxmo because it is light enough, other light candidates are Maemo Leste and Nemo Mobile, which likely mainly need additional contributors to make their software fully viable sooner. For me, speeds are okay with Phosh too—at least on DanctNIX and postmarketOS. You just need to breathe when in a hurry while waiting for Firefox to open, then it’s workable.

With the Librem 5, performance with Phosh is fine—the main issues are idle battery life and heat, which both are somewhat workable when you turn off that Redpine WiFi whenever you can.

We also should not focus too hard on the hardware, as it’s a war won by scale, something we’ll likely never have. It’s no surprise Apple is ahead (and with it TSMC): Tim Cook is a genius at scaling businesses, and they did so with their own silicon by also massively investing into the best foundry they could work with (TSMC). Industry giants like Intel and or (Android manufacturer) Samsung can’t keep up; and Linux Phones never be able to keep up either in terms of performance, battery life or the best camera.

If we manage to have “good enough” devices (a cool running RK3566 PinePhone 2 may be just that, and I hope Purism will manage to reasonably iterate on the Librem 5, too), and we manage to tell a story around privacy, digital wellbeing (I think it’s a feature to not have native clients for TikTok or Instagram) and sustainability (being able to use devices way longer), then that’s IMHO good enough.

6 Likes