Librem 5 v2 a RISC-V Phone

When you invest in the Librem 5 v2 campaign, you don’t just buy a phone. You invest in the future of free software, free hardware and technological freedom in general.

Releasing Librem devices is becoming more and more difficult with SoC vendors locking their chips in ways that keeps being harder to work around. This makes the choice of CPUs for Librem hardware, extremely limited. Especially for a phone like the Librem 5.

The move to RISC-V is challenging at this point, but once done, it will open so many doors by removing many constraints that we currently have with proprietary SoCs.

RISC-V is definitely the future, and thanks to you, supports, we are starting to pave the way! We are on to a great start with the campaign and that is so exciting! :smiley:

The plan is that the “Founders Edition” will be a limited, 250 units, engraved production with the same specs, to thank the early backers. If the campaign exceeds its limit and gets to a stretch goal, it could be more than just the engraving. This is not entirely set in stone yet.

We want to be thankful to our backers anyway!

  1. I’m excited to see a new version of the Librem 5.
  2. I’m very happy to see all the nice upgrades.
  3. I personally have no experience with, or opinion or even knowledge of RISC-V.
  4. I sincerely hope the modem will be multi-regional, and it’s great to see the planned dual-SIM slot.
  5. I hope to see some verification of the longer battery life and cooler operating temperature.
  6. The v2 is expensive.
  7. Purism is using the same chassis, so that should facilitate production and cost savings.
  8. I’m confident it will be produced and delivered if the funding goal is reached.
  9. Current political and supply uncertainties may lead to delays (and public criticism). Again.
  10. At this time I’m not sure if I will join the crowdfunding campaign.
  11. At this time I’m not sure if I will buy the v2 unless my v1 dies.
  12. If some components of v2 prove to be swappable with the v1, I might upgrade what I can.
  13. I still hope for a less bulky, less heavy GNU/Linux phone than either v1 or v2.
  14. Purism needs to manufacture and sell an official protective case for both phones.

Incidentally, I find it a bit incongruous to see mention of those third-party abusive apps (Instagram, YouTube, TikTok, WhatsApp, etc.) in a campaign for a privacy-repecting phone, but I understand why Purism thinks they have to promote them. Sadly.

We have to remember that privacy and software freedom does not need to be a binary issue – people can, and should, take incremental steps toward these ideals. A sudden change to people’s lifestyles and habits are more likely to face resistance.

I just added 1 image on the header-images above.

Yes i am interested on support Purism Librem 5 v2 after francois-techene post.

Yes. I’ve been looking into this (baseline power usage for sleep and suspend on the K1) as well as TDP vs. the i.MX8M Quad and I don’t see a clear path for “longer battery life”. I’m certainly not an expert, but I do have my doubts.

Agreed. This time Purism is probably more experienced at factoring in their own development costs and weighing this against realistic sales numbers.

I don’t know enough about Zink ( Mesa’s OpenGL to Vulkan driver translation layer ; it sounds promising ) to know whether there will be a feasible non-proprietary GPU driver. I don’t think anyone is expecting non-proprietary GPU firmware.

I suppose it could be worse. For example: $1600 instead of $1599.
:laughing:

For power consumption, I see someone in this forum uploaded measurements:

That shows 492 mW as the lowest reading.

We don’t have Librem5 v2 measurements, since it doesn’t exist, but there are measurements for a BananaPi running the SpacemiT K1, which is the CPU in the v2, which gives a good baseline. Referenced here:

Is this doc:

That shows 159 mW.

In sum, 492 mW on ARM, 159 mW on RISC-V.

…and more importantly, is the fancy-and-photogenic(greenish) @dos is CAT noticied about L5v2 for core developments?

I hadn’t seen the “Torch datasheet” link in this forum before. Interesting.

My comment was based on the data I collected from available manuals and other sources for the i.MX8M Quad SoC vs the SpacemiT K1:

Power State / Workload NXP i.MX 8M Quad SpacemiT K1 (SoC Main Control) SpacemiT K1 (Full Board / SBC)
Deep Sleep / Suspend-to-RAM 77 – 105 mW (DSM) ~32.9 mW (Sleep) ~159 mW (BPI-F3 Sleep)
Suspend-to-Idle / Low-Power Standby ~397 mW (167 MHz Idle) Not separately published Not separately published
OS Desktop / Active Idle ~796 mW (800 MHz Idle) ~1.36 W (Bianbu Desktop) ~2.92 – 3.30 W (Desktop / Idle)
Audio / Local Video Playback 466 – 534 mW (Audio) / ~1.8 W (A/V) ~1.83 W (Local Video) ~4.24 W (Local Video)
Online Video / AI Workload ~1.82 W (Stream) / ~1.87 W (VPU) ~2.87 W (Online) / ~3.13 W (AI) ~5.50 W (Online) / ~6.11 W (AI)
Heavy CPU / Multi-Load Peak 1.31 – 1.51 W (CPU) / ~3.35 W (Peak) ~3.56 W (CPU Heavy) ~6.16 – 8.50 W (Board Peak)

Here is NXP’s power doc: https://www.nxp.com/docs/en/nxp/application-notes/AN12118.pdf
Some of the data for the K1 was from a place similar to yours: BPI-F3 SpacemiT K1chip BPI-F3 DEB1 scenario power consumption data - BPI-F3/BPI-CM6(SpacemiT K1) - banana pi single board computer open source project official forum BPI team

There are several of different power modes … and I know that the NXP SoC had some errata in regard to some of the sleep states. I’m still not sure what to expect for relative battery usage between v1 and v2. i.e. Should we compare “Deep Sleep” lines … or should we compare “Suspend-to-Idle on NXP” with “Deep Sleep” on the K1 ??? Perhaps the Librem 5 v1 isn’t going into deep sleep because it can’t wake the cellular modem???

V2 will better for battery and temperature than V1 but not much as 2 hours more. So we will gain 2 hours and risc-v.
Librem 5 v2 main feature is RISC-V only, it is not BAT or TEMP or WAYD, CPU, GPU.

I remember that @dos enabled a thing on STR to wakeup more faster - modem, which i was unhappy, because battery save is much import than speed to me.

I think that V1 can be potentially better it is just that thanks to linux-and-gnu mainline the generic-config is WEAK.
Gnu Dawn will ship powersave-and-tempsave on active-mode but accidentally.

I asked @dos why the v1 soc suck too much power on s.t.r and dos didnt mind responding…disappointingly.

PULL IT FOR FREE. MIRROR
post from amosbatto: 31594/41

Purism’s hardware choices in its next smartphone, the Librem 5 v2 (L5v2), certainly surprised me. Purism is taking a very different path than I expected, and I can see both advantages and disadvantages to its approach. Purism’s decision to focus on the RISC-V architecture should lead to greater freedom in the long term, and get rid of the nasty DDR RAM training blob at boot that other recent processors (RK35xx, POWER 10, i.MX 8M, i.MX 95) are requiring. However, supporting the SpacemiT K1 processor is going to require quite a bit of dev work on Purism’s part, and the performance of the K1 is underwhelming compared to the SoC’s used in current and upcoming Linux smartphones: MNT Vector, Dawndrums Divine D., Mecha Comet, Commodore Callback 8020, Furilabs FLX1s+, Jolla Phone (2026) and Volla Phone Plinius.

I had never heard of SpacemiT before, so at first I wondered why Purism selected a startup chipmaker that was founded in November 2021, when it could have used an SoC from an established ARM chipmaker like Rockchip or NXP. According to its Pitchbook profile, SpacemiT has 130 employees and is based in Hangzhou, Zhejiang Province, China. It has 22 investors, including Legend Holding (which owns Lenovo), China Internet Investment Fund, Shangqi Capital and ABC Financial Assets Investment, plus loans from the Agricultural Bank of China. SpacemiT focuses is producing RISC-V chips that support on-chip AI, and it just raised $86 million in capital for its future RISC-V server chip.

Purism says that it is planning on designing its own RISC-V SoC in the future, so it makes sense for the company to start pivoting to RISC-V with the Librem 5 v2 in order to gain experience with the architecture, before it attempts to design an SoC with the architecture. It is vitally important that Purism makes this pivot to RISC-V, because there will probably never be another chip using the x86, Power or ARM architectures, where it is possible to run a 100% POSS operating system. Intel’s Nehalem introduced in 2008 made it impossible to remove the Intel Management Engine (IME), since it would shut down after 30 minute without it. AMD’s Platform Security Processor (PSP) introduced in 2013 is even worse, since we don’t know how to disable it with software. The IBM POWER 10 introduced in Sept. 2021 required proprietary blobs from Synopsys to initialize its DDR4/5 RAM and PCI express 5.0. The last ARM SoC that can be run without proprietary blobs is Rockchip RK3399, which was released in April 2016, but it took 3 years for the community to get rid of its blobs.

SpacemiT appears to me to be better than all the other ARM and RISC-V chipmakers in terms of open access and software freedom. SpacemiT publishes all its documentation and code for its processors and doesn’t require signing an NDA to get the documentation, like Qualcomm, Rockchip, Allwinner, Amlogic, Broadcom and Nvidia. Unlike those companies, SpacemiT also has a public forum where anyone can ask questions. SpacemiT describes itself this way on its Github page:

Upstream-First Engineering
We actively contribute to open-source foundation projects. Both K1 and K3 are supported in mainline Linux, standard GCC, LLVM, and Binutils. No proprietary toolchains, no obsolete custom forks — standard distribution packages from Debian, Ubuntu, and Fedora work out of the box.

SpacemiT works with RISCstar (which commits to mainstream Linux), OpenEuler’s RISC-V group and Canonical to get its hardware supported upstream. SpacemiT provides an SDK for its processors on the following 7 platforms: Debian, Bianbu (SpacemiT’s Ubuntu-based RISC-V distro), ROS 2, OpenWrt, Buildroot, OpenHarmony and Eco-System.

NXP’s i.MX platform is generally regarded as the best in terms of freedom, but SpacemiT appears to be just as good as NXP in terms of access to its documentation, code and its forum, and you don’t have to login to its platform to get access. Intel and AMD stand above all the other chipmakers in terms of getting their hardware supported upstream, but I would place SpacemiT above all the other ARM and RISC-V chip companies due to its goal that its hardware will be able to run in standard Debian, Ubuntu and Fedora. It still requires a non-standard device tree and blobs that aren’t in mainline Linux in some areas like camera support, but it is still admirable in its ambition to support the major desktop Linux distros.

Even though I am disappointed that Purism didn’t choose to use the Rockchip RK3588S for its next phone, it makes sense to me why Purism would want to work with SpacemiT over Rockchip. Rockchip has removed its open source documentation from its website, and it doesn’t appear that interested in good mainline Linux support, since outside organizations such as Collabora and Ideas On Board have done most of the work to get its RK35xx processors supported in mainline Linux.

The SpacemiT K1 SOC is advertised as supporting 50K DMIPS (Dhrystone million of instructions per second) of CPU computing power. SpacemiT claims that its X60, which is a low-power, RVA22, 256-bit RVV 1.0 core with an 8-stage, dual-issue, in-order pipeline, offers 30% better CPU computing power and 20% better power efficiency than ARM’s low-power Cortex-A55 core. The K1’s AI capability with 2.0 TOPS is limited, but I doubt that there will be much use for the NPU in the Librem 5 v2, since mobile Linux also needs the software to do things like speech, image and OCR recognition.

However, the available benchmark scores for the K1 aren’t that impressive. The Geekbench 6 single-core CPU scores for the K1 are a tad worse than the i.MX 8M Quad in the Librem 5 (my measurements), but its multi-core scores are better, which is to be expected since the K1 has twice as many cores as the i.MX 8M Quad. Considering that the K1 is designed to run 8 cores at 1.5 - 2.0 GHz and fabbed on a 22nm TSMC planar process (versus 4 cores at 1.5 GHz and 28nm for the i.MX 8M Quad), I would expect better scores for the K1. Geekbench only recently added support for RVV 1.0, and I suspect that it isn’t as well optimized for the RISC-V architecture as for ARM and x86, so I wouldn’t trust the Geekbench numbers too much.

Comparing available SoC’s for Linux phones

SoC Process (nm) Specs GB6 single GB6 multi Phones
Allwinner A64 40 4x 1.2GHz Cortex-A53, Mali-400 MP2 109 278 PinePhone
Rockchip RK3566 22 4x 2.0GHz Cortex-A55, Mali-G52 MP2 140 450 rumored PinePhone 2
Rockchip RK3399 28 2x 1.8GHz Cortex-A72, 4x 1.4GHz Cortex-A53, Mali-T860 278 708 PinePhone Pro
NXP i.MX 8M Quad 28 4x 1.5GHz Cortex-A53, GC7000 Lite 146 477 Librem 5, Liberty
SpacemiT K1 22 8x 1.6GHz X60 RISC-V, IMG BXE-2-32 134 583 Librem 5 v2
SpacemiT K3 12 8x 2.6GHz X100 RISC-V, PowerVR BXM-4-64 350 1738 15-30 W TDP
NXP i.MX 8M Plus 14 4x 1.8GHz Cortex-A53, GC7000 UltraLite Mecha Comet
NXP i.MX 95 16 6x 2.0GHz Cortex-A55, Mali-G310 350 1300 Mecha Comet
Rockchip RK3588S 8 4x 2.4GHz Cortex-A76, 4x 1.8GHz Cortex-A55, Mali-G610 MP4 730 2250 Liberux NEXX, Dawndrums Divine D., possible MNT Vector
UNISOC Tiger T606 12 2x 1.6GHz Cortex-A75, 6x 1.6GHz Cortex-A55, Mali-G57 MP1 375 1350 Jolla C2
MediaTek Helio G81 12 2x 2.0GHz Cortex-A75, 6x 1.7GHz Cortex-A55 cores, Mali-G52 MC2 432 1416 Commodore Callback 8020
MediaTek Dimensity 900 6 2x 2.4GHz Cortex-A78, 6x 2.0GHz Cortex-A55, Mali-G68 MP4 1003 2423 Furilabs FLX1/s/s+
MediaTek Dimensity 6300 6 2x 2.4GHz Cortex-A76, 6x 2.0GHz Cortex-A55, Mali-G57 MC2 780 1940 Brax BraX3 (Ubuntu Touch port stalled)
MediaTek Dimensity 7050 6 2x 2.6GHz Cortex-A78, 6x 2.0GHz Cortex-A55, Mali-G68 MC4 970 2330 Volla Phone Quintus
MediaTek Dimensity 7100 4 4x 2.4GHz Cortex-A78, 4x 2.0GHz Cortex-A55, Mali-G610 MC2 995 2900 Jolla Phone (2026)
MediaTek Dimensity 7300 4 4x 2.5GHz Cortex-A78, 4x 2.0GHz Cortex-A55, Mali-G615 MC2 1040 3000 Volla Phone Plinius
Qualcomm QCM4490 4 2x 2.4GHz Cortex-A78, 6x 2.0GHz Cortex-A55, Adreno 613 900 2318
Qualcomm QCM6490 6 4x 2.7GHz Cortex-A78, 4x 1.95GHz Cortex-A55, Adreno 643L 950 2850 MNT Vector, Fairphone 5, SHIFTphone 8
Qualcomm QCM8850 4 1x 3.2GHz Gold-Plus, 4x 2.8GHz Gold, 3x 2.0GHz Silver, Adreno 740 1100 3300 likely MNT Vector
Qualcomm Snapdragon 4 gen 4 4 2x 2.3GHz Cortex-A78, 6x 2.0GHz Cortex-A55, Adreno 613 920 2210
Qualcomm Snapdragon 6 gen 4 4 1x 2.3GHz Cortex-A720, 3x 2.2GHz Cortex-A720, 4x 1.8GHz Cortex-A520, Adreno 810 994 3082
Qualcomm Snapdragon 7 gen 4 4 1x 2.8GHz Cortex-A720, 4x 2.4GHz Cortex-A720, 3x1.84GHz Cortex-A520, Adreno 722 1150 3700
Qualcomm Snapdragon 8 gen 5 Elite 3 2x 3.8GHz Oryon Prime, 6x 3.32GHz Oryon Performance, Adreno 829 or 840 2400 8700

The K1 has 1MB of L2 cache and no L3 cache. Each of its two 4-core clusters uses a separate 512MB L2 cache, which I suspect is limiting its benchmark scores. In comparison, the i.MX 8M Quad has twice as much L2 cache per core, and the two clusters in the K1 can’t share the workload as easily since they don’t have a common cache.

The K1 supports up to 16GB LPDDR4x, so it is a much better processor than the i.MX 8M Quad, which is limited to 4GB LPDDR4 for high-memory tasks like processing video, opening dozens of browser tabs, using a Waydroid container to run Android apps, and running an external display to use the phone as a convergent PC. However, i.MX 8M Quad’s memory is faster, supporting up to 3200 MT/s (12.8 GB/s), compared to the K1 that supports 2666 MT/s (10.67 GB/s), despite using the newer LPDDR4x standard.

I understand the economic constraints that a small chipmaker like SpacemiT operates under (see my previous article), but it is hard to get excited about a processor fabbed on 22nm planar node that only has low-power cores. In comparison to the 2 in-order instructions per cycle in the X60 core, ARM’s high-performance Cortex-X cores can fetch up to 5 instructions per cycle and feature out-of-order execution, twice as much L1 cache and 4 to 16 times more L2 cache per core, plus L3 cache.

The real advantage that I see for the K1 is its energy efficiency. The K1 uses Dynamic Voltage and Frequency Scaling (DVFS) between 0.6V and 1.05V, so power draw can be scaled, according to the SpacemiT K1 Datasheet. I suspect that the reason why Purism says it needs 9 months to work on energy managment is that it needs to work on the software support for the K1’s DVFS, and I suspect that dev work will reach all the way up to Phoc and Phosh. The i.MX 8M Quad uses up to 3.4 watts at 1.5GHz, whereas the K1 uses up to 3 watts at 1.5-1.6 GHz and up to 5 watts at 1.8-2.0GHz. I assume that Purism will only run the K1 at a max of 1.6 GHz, like many of the SBC’s that use the chip with passive cooling. In comparison, the SpacemiT K3 processor is quite a bit more powerful and advertized to process 130k DIMPS, but its thermal design power is between 15 and 30 watts, so it consumes way too much power for a phone.

The other advantage I see in using the K1 is better software freedom in the long term. All the recent Rockchip and NXP i.MX chips contain Synopsys DesignWare DDR4/5 RAM, which requires a proprietary blob to train the memory timing at boot. The K1 also has a RAM initialization blob, but it can be replaced by oreboot’s POSS initialization, so it is now possible to boot the K1 without blobs.

The K1 contains Imagination’s BXE-2-32 GPU, which supports Vulkan 1.3, OpenGL ES 1.1/3.2 and OpenCL 3.0. It scores 126 on Geekbench 7’s Vulkan Benchmark, which is roughly equivalent to the Adreno 506 GPU used in the Snapdragon 625, 626, 632 and 450, released in 2016-7. It appears to be better than the GC7000Lite GPU in the L5v1, but I’m not willing to pay for a Geekbench 7 license in order to find out how much better.

Mesa’s PowerVR free driver only suppports Imagination’s BXM-4-64 and BXS-4-64, so currently a userspace proprietary blob is required to use the Imagination BXE-2-32 GPU. Purism projects that it will take 6 months of dev work to add Mesa support for the BXE-2-32 GPU. The PowerVR driver currently supports Vulkan 1.2 and Imagination is working on adding Vulkan 1.3 and 1.4 support in the future. OpenGL ES 2.0 support is available through Zink translation from Vulkan, and OpenGL ES 3.1 will be added in the near future through Zink. The PowerVR driver lacks Gallium3D support, so it can’t use Mesa’s Rusticl free OpenCL library.

The K1 uses the ARM Mali-V52 VPU for encoding and decoding video, which uses the free mvx_v4l2 kernel driver to create V4L2 memory-to-memory (M2M) device nodes (such as /dev/video0 for decoding and /dev/video0 for encoding), but the driver relies on proprietary firmware. In the FAQ for the Librem 5 v2, Purism mentions using a firmware jail, so this firmware presumably will be included in that jail. On the Librem 5 v1, the firmware is pulled from an SPI NOR flash memory chip and stored into a initramfs file system so it can be loaded into a component at boot, and I assume that the VPU’s firmware blob will be loaded in the same way on the Librem 5 v2.

The drawback to the K1 in terms of software freedom is using the camera. SpacemiT provides a POSS driver for the K1’s Image Signal Processor (ISP), but it relies on 6 proprietary shared library files to do the 3A processing (auto-exposure, auto-white-balance, auto-focus), face-detection and ISP pipeline control, and these files run in userspace. Purism doesn’t mention hiring anyone to work on the camera drivers in the L5v2 announcement, so I’m guessing that the L5v2 probably won’t use the hardware Image Signal Processor and will have to process images and video in software which will be slower and make the phone run hotter. At any rate, the audio processing will have to be done in software since the K1 lacks a hardware aDSP.

The good news is that at least the K1’s VPU can handle the video encoding into H.264, H.265, VP8 and VP9 formats (up to 4K at 30Hz), so the L5v2 will be a big improvement over the L5v1 that lacks a hardware video encoder. The higher speed microSD card in the L5v2 will also help a lot in terms of storing recorded images and video, because people who use their cameras a lot are going to run out of space in the 128GB eMMC pretty quickly.

The Librem 5 v2 won’t be that great of a phone for convergence. Its CPU needs some high-performance cores and the BXE-2-32 isn’t that powerful of a GPU. In my opinion, 8GB RAM and 128GB eMMC is not enough for a capable convergent phone. People who want to use the L5v2 as a desktop PC are going to have to store a lot of files on a microSD card which will be significan’t slower than the eMMC, because 128GB isn’t enough storage space for most people. According to the Gsmarena database, 93% of new phone models (519 of 558 models) introduced in 2026 offer 256GB or more of storage, so there is no reason for Purism to offer such limited storage space, when it is marketing the L5v2 as a convergent phone and asking a very premium price of $1599 for the phone. Given that the Banana Pi BPI-F3 SBC with the K1 processor costs $135 for 16GB RAM and 128GB eMMC, I think that Purism should have included 16GB RAM and 256GB of Flash memory in the L5v2 in order to be able to sell it as a convergent device.

In terms of video out, the K1 supports HDMI, but doesn’t support DisplayPort. Purism says that the Librem 5 v2 will support DisplayPort Alt Mode over USB-C, so presumably an extra HDMI-to-DisplayPort conversion chip will be added to the circuit board. The Mali-V52 VPU in the K1 can decode 4K video at 60Hz, but its video out is limited to HDMI 1.4, whose maximum data throughput is 10.2 Gbps, which means a max resolution of 4096×2160 at 24Hz or 3840×2160 at 30Hz. In comparison, the i.MX 8M Quad in the Librem 5 v1 supports 4K video out at 60Hz.

Purism’s decision to use the same case and screen for the L5v2 as the L5v1 will presumably save some development costs for the phone, and hopefully Purism will be able to offer an upgrade kit in the future for the people who own the L5v1. Given the amount of electronic waste (e-waste) in the world, it would be nice if Purism can sell upgrade boards. As far as I know, SHIFT is the only phone manufacturer that has ever offered upgrade boards for their phones (SHIFTphone 5.1 → 5.2 → 5.3), and it no longer offers that option, so Purism could stand out as the only phone maker in the world to offer hardware upgrades.

I personally would have prefered a phone based on the Rockchip RK3588S SoC, but I can see the logic of switching to a RISC-V processor given Purism’s long-term plans, and SpacemiT really is a good company that deserves to be supported. Rochchip has removed its open source documentation from its website. It continues to update its published kernel code for its processors and answer Collabora’s questions, but it isn’t as open or helpful to the Linux community as it used to be.

For people who want a powerful convergent phone to use as a desktop PC, the L5v2 is probably not the best choice. I recommend keeping an eye on the upcoming MNT Vector, which will have a replaceable SoC, RAM and eMMC on a 200-pin SO-DIMM card. MNT says that it is designed to use the Qualcomm Qualcomm QCS6490 and QCS8850 processors, but it has also tested the RK3588 on it, so that may also be an option for people who want to avoid most of the blobs. It will be a long shot for the Tunisia-based Dawndrums to bring the Divine D. to market, but it also promises to be a good convergent phone, based on the RK3588S.

The three areas where I foresee the L5v2 improving over the L5v1 is in terms of energy efficiency, hardware acceleration with Vulkan support and its ability to record video, which are all areas where the L5v1 falls short in my opinion. If Purism can work on the energy management with voltage scaling with the K1, the L5v2 would require less energy and run much cooler, so it should solve the problem with the current Librem 5 being a space heater with short battery life. If Purism can work on the PowerVR driver to add support for the BXE-2-32 GPU and get Vulkan hardware acceleration added throughout Phosh and the GTK4 applications, then the L5v2’s interface should be much faster with fewer CPU cycles, which will dramatically the responsiveness of the software. Purism is also promising that L5v2 will be able to handle higher resolution video than the L5v1, which I assume is due to the K1’s ability to do hardware video encoding. Purism says that it will offer a 5G modem upgrade in the future, but I’m really hoping that they can solve some of the connection problems with the current 4G BM818 modem.

Maybe the K1 processor isn’t the dramatic improvement like I was anticipating with the RK3588S, but I suspect that most people who are currently using the L5v1 will be content with the upgrade that the v5v2 offers as long as they aren’t planning on using the L5 as a convergent desktop PC. However, I do think that Purism should bump the RAM up to 16GB and the storage up to 256GB, because it is asking people to pay a very premium price at $1599. Even if Purism can’t provide a powerful processor, more memory and storage will provide a much better desktop experience and allow the L5v2 to be better marketed as a convergent phone.

I like Purism’s vision and admire the company’s ambition, and I encourage others to pre-order the phone, because the work Purism is doing is important and worth supporting. The L5v2 will be the first RISC-V phone on the market, and Purism is paving the way for other companies to follow, just like it did with the L5v1. I don’t foresee the situation getting any better with ARM, Power and x86 processors, so RISC-V is the best hope that we have for being able run hardware in the future with 100% free software. This appears to be the first step for Purism to eventually design its own RISC-V SoC which presumably won’t have an blobs, so I really want to support that work.

Nonetheless, I don’t think I can afford to support the L5v2 at a price tag of $1599. That is OK, but I hope that there are enough people in the world who are willing to put down the money to get the L5v2 into production, because this phone will break a lot of new ground

I’m less concerned about Flatpak support than I am about AppImage support.

If the operating system can run AppImages I could probably forego flatpak apps entirely, and it would save a lot of space on the device.

Is there a way to get some truly epic power savings we have been missing via risc-v or is that not a there-yet thing? The biggest downside to our phones is the battery life and I dont think we will see this with rockchip or i.MX 8M and for sure not allwinner.
I think we also need to do some LLM work at breaking out the OS of the modem cards like they did on the pinephone but more-so to hopefully undervolt underclock and get some power savings other ways, less burner phone cook when we make a phonecall.

I currently own two RISC-V boards (Milk-V Jupiter 2 with the SpacemiT K3, and the Milk-V Titan) so I happen to know the situation around the SpacemiT upstreaming process already. They have a README.md containing a list of all projects they are sending patches to. One of them is the Linux kernel, for which they have a special Wiki page. Specifically for the SpacemiT K1, most of the upstreaming has been done in kernel 7.2, with only a few (but important!) drivers left like Display, GPU and VPU, that still need to be upstreamed. I recommend bookmarking the Linux wiki page and re-visiting it regularly, as it gets updated every few weeks. SpacemiT also offers a monthly upstream progress update on Reddit.

Additionally, Canonical collaborates with SpacemiT for supporting the K1 and K3 on Ubuntu. That means that partner-built images should no longer be necessary, but can just use the Canonical-built images, with only a few patches on top that have not made it in time in the mainline kernel. This is already officially possible with the K3, and I tried it myself on my Milk-V Jupiter 2 with Ubuntu 26.10 and kernel 7.3. While this seems to be unrelated to Purism, it is relevant in the sense that there is already official support without a vendor-specific kernel, meaning that work for supporting the SpacemiT K1 by both SpacemiT and Canonical will benefit the Librem 5 v2 as well, since a vendor kernel is mostly no longer necessary.

Back to the Librem 5 v2, the work above lays the groundwork for the SpacemiT K1, but since the patches have been tested on desktop boards, they might eventually not include all battery and efficiency improvements necessary for a battery-powered device like a smartphone. Still, like Purism mentions on the crowdfunding page, further work needs to be done in addition to the kernel.

Personally, I’m genuinely very excited about Purism choosing a RISC-V processor for their next smartphone. I was very happy to fund their crowdfunding campaign as fast as I could. I think 1599$ is a lot, but since Purism tries to upstream all the work, I view it rather as an investment in open source, Linux smartphones, and the RISC-V ecosystem as a whole, and I think this is something worth funding.

I should note that starting with 25.10 and on an ongoing basis (e.g. 26.04, …) that Ubuntu is only supporting RVA23 and later (i.e. not supporting RVA22 for the K1). At least that is my understanding. i.e. For Ubuntu LTS, the K1 support stops with 24.04.

Aside: It seems like all of the SBCs that support RVA23 are built on the SpacemiT K3. The K3 looks very interesting to me.

K1 still ship old 2D ASIC Desing while K3 ship new 3D ASIC Desing(like FIR), so yes K3 is better but no much for a phone, so K3s should better for a phone.

Very true! According to Canonical’s announcement, they will support the K1 officially on Ubuntu 24.04 (but not on 26.04 like the K3, as the K1 uses RVA22 instead of RVA23). The situation is similar for Bianbu Linux, which is SpacemiT’s Linux distribution based on Ubuntu. For example, Bianbu 4.0 is based on Ubuntu 26.04 and thus does not work on the K1. However, SpacemiT still supports Bianbu 2.3, based on Ubuntu 24.04.1.

In case anyone is interested, here is SpacemiT’s official download page, which points to downloads on this page, which also includes a Debian build. A list of all projects they are providing download for can be found on this page.

My point is: yes, there might be some work left for the SpacemiT K1, and some further work for it to work properly on a smartphone (battery optimizations, performance improvements, etc). And while I agree that choosing an ARM-based CPU would have been the easier choice, I think choosing RISC-V is definitely the right decision. The SpacemiT K1 is already mostly supported by the mainline Linux kernel, and boards using it already exist since almost 2.5 years. I mean it in a positive way, as this means that Purism can concentrate on properly supporting it on a smartphone, choosing a battle-tested CPU (at the cost of it being almost 2.5 years old).

In conclusion, I think that choosing RISC-V was definitely the right decision for the Librem 5 v2.

Thanks! Your posts have been very informative.

I don’t think it was the right decision for them – I don’t think it will work out well for them, but I could easily be wrong since I don’t know much about RISC-V.

However, I like the decision because it gives me the impetus to investigate RISC-V and available SBCs.