[HN Gopher] State of Embedded: Q4 2025 Overview
___________________________________________________________________
State of Embedded: Q4 2025 Overview
Author : HeyMeco
Score : 119 points
Date : 2025-10-24 17:52 UTC (5 days ago)
(HTM) web link (sbcwiki.com)
(TXT) w3m dump (sbcwiki.com)
| HeyMeco wrote:
| In this format I go deep down the Mailing lists, news articles
| and more to summarise what exciting hardware has been published
| and software has been merged into Linux. Also breaking down
| rumours and developer conferences about future SoC's. Hope you
| all find it useful!
| mberger wrote:
| Can you include more prices? It would give me an idea of the
| cost even if it is in USD. What i found most annoying about my
| latest search is that it is hard to find something not named
| raspberry or Arduino for a reasonable price. I was looking for
| a simple gigabit board with usb 3 to attach a removable drive
| to. The only one i found was raspberry pi orange 3B . Nobody
| else seemed to have gigabit nic with usb 3.
| nine_k wrote:
| Why would you expect USB3 and Ethernet, fast and relatively
| expensive interfaces, to be attached to a cheap low-spec MCU?
|
| Did you consider a ready-made USB3 extender over Ethernet?
| There is a reason they cost so much ;-/
| Joel_Mckay wrote:
| Pi4 does two USB 3.0 ports, but you are right in that USB
| 3.1 is a little much for a SoM.
|
| Its a nice little SoM, in some ways it was better than the
| pi5 for hardware media encoding. =3
| rjsw wrote:
| Rockchip SoCs starting with the RK3399 can do both USB3 and
| Ethernet.
|
| The only board that I own that does both at the same time
| is the Pine64 Quartz64 that uses the RK3566. My Pinebook
| Pro doesn't have an ethernet port, Orange Pi 5 Max has
| ethernet but doesn't use the builtin controller to provide
| it.
| idatum wrote:
| I use NetBSD on a Pine64 RockPro64 and use USB3. It has
| been stable: awge0:
| flags=0x8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu
| 1500 ec_capabilities=0x1<VLAN_MTU>
| ec_enabled=0x1<VLAN_MTU> address:
| 26:80:xx:xx:xx:xx media: Ethernet
| autoselect (1000baseT full-duplex)
| Joel_Mckay wrote:
| The Raspberry PI also has an intangible value from years of
| community goodwill. And people trust that the kernel OS
| support will be around in 10 years.
|
| The NVIDIA solution is impressive... but self-immolated with
| the consumer price point (markets for government equipment
| may work.) People usually either have money or time... asking
| for both in a product is foolish.
|
| The other SoM also have a long-tail market attention problem,
| as one could spend 2 weeks tracking unstable kernel driver
| problems. Or just drop in a $35 pi, and solve the task at
| hand. =3
| fisian wrote:
| Does a Banana Pi BPi-M5 fit your specs? The banana pis have
| pretty good networking options.
| joezydeco wrote:
| If you could mirror your feed on Bluesky, that would be
| appreciated. Some of us will not use X.
| Y_Y wrote:
| Mastodon or bust!
| HeyMeco wrote:
| https://mastodon.social/@sbcwiki
| jauntywundrkind wrote:
| Lovely snapshot in time view of the small computers that drive a
| ton of product innovation.
|
| One major gap I'd really like to see covered better is power
| consumption. We've started seeing unboxing and reviews for the
| Arduino Uno Q for example, but no one seems to be breaking out a
| simple USB power meter to get some idea how much power it takes,
| headless or sitting at desktop.
|
| Pretty nice that so many chips have pretty good Linux support.
| jandrese wrote:
| I have some (somewhat older) OrangePis here that came with a
| board covering heatsink with an integrated fan because it turns
| out they really need it. The things get hot to the touch when
| working hard. I don't have any power measurement hooked up to
| them, but they have to be drawing a fair bit of juice to warm
| up that much.
| tonetegeatinst wrote:
| Does any SBC offer SFP+ or SFP28?
|
| I don't see any option to sort by network speed or network chip
| nbf_1995 wrote:
| The solidrun LX2 offers 4x SFP+, but that board is getting
| quite old at this point.
|
| There are a couple of bananapi router boards that have 1 maybe
| 2 SFP+
| Havoc wrote:
| The ones that do have pcie are generally one lane of gen 3.
| some n100 boards might have enough of you stick storage on data
| jpm_sd wrote:
| Nothing about NXP? They are a pretty big player in this space,
| and mostly do a great job contributing upstream to the kernel,
| but I guess they don't have a hobby-level entry point readily
| available.
|
| I've used NXP-based embedded Linux hardware from Toradex,
| Gateworks and EmbeddedTS in previous projects, lots of vendors
| out there.
|
| https://www.toradex.com/
|
| https://www.gateworks.com/
|
| https://www.embeddedts.com/
|
| https://variscite.com/
|
| https://www.ibase.com.tw/en
|
| https://www.phytec.com/
|
| https://www.ezurio.com/
| nickpeterson wrote:
| Anything new happening on the risc-v side of things? Last I heard
| there were some micro-itx boards at around rpi4 speed but haven't
| heard much since.
| rcarmo wrote:
| Huh. It mentions neither Intel nor AMD, and I am seeing _a lot_
| of N100/N150 industrial SBCs right now (outside the hobbyist
| space). Intel in particular seems to be replaying their approach
| of flooding the market with Atom chips and making N-series
| available to OEMs in volume and relatively low price.
|
| Full marks for Rockchip coverage, yes, they're filling in the
| gaps below the RK3588, although some of those chips aren't that
| interesting in terms of power budget and have apparently low
| yields.
| Sponge5 wrote:
| it's right there in the first section:
| highlighting notable advancements in ARM-based Systems-on-Chip
| (SoCs) and their increasing competitiveness against traditional
| x86 platforms.
| bluGill wrote:
| Not discussing x86 doesn't highlight their competitiveness.
| It might be a useful article for some, but for most of us the
| N100 is a better option overall for everything we might look
| at a SoC for. YMMV of course, I haven't seen a N100 SoC (I've
| also never looked), but complete N100 systems that are ready
| to work are similar prices to an ARM SoC after you buy the
| non-optional extras like case, disk, and power supply.
|
| It also misses the other end - many things people think of
| SoC for could be done with a ESP32 or other micro controller
| for less cost, and this might be a better option.
|
| I'm not completely faulting them, you have to set limits
| someplace. However the limits they have make this summary
| less useful for most people who will read it.
| wpm wrote:
| The N100/150's are great.
| BolexNOLA wrote:
| Got an N150 running elementary as a basic plex server and I
| couldn't be happier. About to shift it to Jellyfin this
| winter just need to find time and do it lol
| 0cf8612b2e1e wrote:
| I picked up a no name brand N100 for like $150, came with a
| (pirated key at this price point?) Win 11 pro license and
| am blown away at how capable it is. Could easily be a
| desktop machine for web browsing and office work. Would
| undoubtedly struggle with heavy electron apps, but really
| impressive. I even tried to run a few games on it, and it
| still handled everything surprisingly well.
|
| It makes me look at my way over specced workstation (for
| significantly more than $150) wondering why I am burning so
| much electricity daily. Should use a micro desktop for the
| web and just remote into my beefy workstation as required.
| theothertimcook wrote:
| I daily'd an n100 box for a year for browsing, admin and
| light media. Had a second running Plex headless.
|
| Using stripped down versions of win iot, they were
| fantastic.
|
| The oem keys are legit fyi.
| zokier wrote:
| My biggest frustration with Intel Twin Lake (2025 Nxxx
| series) is its anemic PCIe configuration. 9 lanes is pretty
| restrictive, but the real kicker is them being only PCIe 3.0
| in a SoC released in 2025(!).
|
| It is not directly comparable, but AMD 8840U and V3C14 both
| give you 20 PCI 4.0 lanes.
|
| Of course it is true that ARM SoCs are usually even worse in
| this regard.
| wpm wrote:
| TBF that's usually my frustration with _any_ consumer level
| chip. I get it takes die space and power, but even high-end
| non-HEDT platforms still cap out at just enough PCIe lanes
| for the common loadout of "One GPU at x16, One SSD at x4".
|
| Which, fine, I get it that most people don't need or want
| more than that, but I shouldn't have to jump up 3x in power
| draw and heat and 5x in cost to go to HEDT to get _far_
| more lanes than I actually need. I 'm lookin for enough
| PCIe for a couple of SSDs, maybe a third, and I dunno,
| _more than one goddamn slot_ to be wired to my CPU without
| compromises? Every ATX motherboard for consumers brags
| about all their slots but they 're all only x4, squeezed
| through the PCH's _4_ lanes. Just because I have a high
| bandwidth capture card, a 10 /25/40/100Gb NIC, _and_ a GPU
| doesn 't mean I want to only give one of them far more
| bandwidth than it needs to the CPU and compromise on the
| rest.
|
| I need like, 32 lanes. Not 64 in a Threadripper. Not
| 128828182 or whatever comes on a brand new $10,000k EPYC, I
| need 32, and don't waste em on a bunch fo flaky USB crap I
| can't use. Don't give me a fucking oculink port. I want to
| use the slots. That i have. In my computer. That I can
| screw the card into securely and power without an
| additional power supply. I can't be the only one.
| npodbielski wrote:
| I looked briefly at one PC like that. 15w of power looks
| great but from other hand 2 nvme disk 6tb Max total does not
| looks that great anymore. You have to have some other PC or
| Nas for backup at the least.
| nicce wrote:
| Depends on the purpose. Maybe not the best for very big NAS
| with 10G ethernet requirement but suitable for many other
| purposes. Maybe you don't need that much space or maybe you
| are just running git server.
| tonymet wrote:
| the AMD and intel mini PCs are a fabulous value and the best
| solution for hobbyist / homelab
| bangaladore wrote:
| Title critique:
|
| RaspberryPi is not an SoC vendor. They take proprietary SOCs from
| Broadcom, use proprietary firmware and build a product around it.
| They obviously upstream what they can, but they fundamentally are
| a system integrator, not an SoC vendor.
| wolrah wrote:
| Pis use Broadcom SoCs, not Qualcomm.
| bangaladore wrote:
| Yes, thank you for the correction.
| klelatti wrote:
| Not to disagree with your critique but just to note that they
| are a microcontroller vendor with the RP2040 and RP2350.
|
| I do wonder if there is a long term wish to break their
| dependency on Broadcom but I suspect creating an SoC for the
| main Pi series is probably in the 'too difficult' category.
| bangaladore wrote:
| Yup, I really like their microcontrollers (which quite
| frankly nowadays what's the difference between an SOC and
| some of these more capable MCUs), but my comment was
| specifically relating to their "Linux" capable boards.
| klelatti wrote:
| You might like the AMP hour podcast that had the team
| behind the RP2350 on. Lots of interesting stuff on what was
| involved in making it. One of my favourite podcast episodes
| this year.
|
| https://theamphour.com/687-the-rp2350-with-the-raspberry-
| pi-...
| 8cvor6j844qw_d6 wrote:
| I have dabbled in model edge processing like Efficientnet and
| YOLO on Hailo platform.
|
| Raspberry Pi and its AI HAT+ seems to be the most accessible,
| often others can easily pick it up the basics and get up and
| running without much trouble even without experience.
|
| I wonder if there is any alternative? Raspberry Pi 5 + the 26
| TOPS HAT+ is not cheap.
| justin66 wrote:
| > I wonder if there is any alternative? Raspberry Pi 5 + the 26
| TOPS HAT+ is not cheap.
|
| Combined they're like $250? Not expensive.
|
| On the other hand, at that price maybe you ought to get a
| Jetson Orin Nano.
| babl-yc wrote:
| What Raspberry Pi is to Broadcom (developer-friendly SBCs),
| Beagleboard is to TI.
|
| It's a slightly different approach -- Beagleboard is a non-
| profit and emphasizes openly purchasable components. But
| similar in that it is the cheapest way to tinker with SoCs from
| that vendor.
|
| BeagleY-AI has 4 TOPS for ~$70. AI inference tooling is still
| improving but I've been working on it here:
| https://docs.beagleboard.org/boards/beagley/ai/demos/using-e...
| wolrah wrote:
| Something I'd be very interested in seeing summarized is the
| current state of fully open source software on SoCs and SBCs. I
| hate how common the situation described in the nVidia section
| where SoCs that require vendor kernels get abandoned on ancient
| software, so it would be very useful to know what SoCs are
| supported to a useful level by mainline kernels.
|
| I feel like there are three tiers of support that most people
| would be interested in:
|
| 1. Usable for headless appliances (serial console or
| unaccelerated graphics, wired networking, storage, USB)
|
| 2. Usable for interactive use (accelerated graphics, WiFi/BT)
|
| 3. Fully supported (all major hardware works)
| NewJazz wrote:
| I agree having this info more plainly available would be great,
| but in the interim interested folks can take a look at
| postmarketos device support, collabora blogs, and the pine64
| wiki to get some of this info.
| zokier wrote:
| Are there any recent ARM SoCs that are fully supported by
| mainline kernel and don't have weird proprietary blobby boot
| chain?
| cogman10 wrote:
| RK stuff I believe is the closest. I believe it still has a
| binary blob for the GPU but generally a lot of the other
| stuff is mainline.
|
| I've not seen any other ARM provider come close to mainline
| support.
| mfuzzey wrote:
| NXP i.MX, TI, STM32MP
| msgilligan wrote:
| What are the most recent and/or highest performance of
| these?
|
| Update: And mainline support and lack of proprietary boot
| blobs are two separate criteria. I've heard that NXP offers
| the former but not the latter.
| msgilligan wrote:
| Well, a 5-year-old chip may not count as "recent" but the
| RK-3588 boot chain is "almost fully open-source" [1]. And it
| seems like it took a major amount of effort (from Collabora,
| others) to get it this far. I don't know of any equivalent or
| newer chips that are "more" open, but would love to hear if
| there are.
|
| [1] https://www.collabora.com/news-and-
| blog/blog/2024/02/21/almo...
| aurintex wrote:
| And the NPU part is also in the mainline kernel
| https://blog.tomeuvizoso.net/2025/07/rockchip-npu-
| update-6-w...
| zokier wrote:
| Yeah, Collabora has done good job on rk chips, but even
| then there are still missing bits and pieces. At least for
| rk3588 they are documented nicely:
| https://gitlab.collabora.com/hardware-
| enablement/rockchip-35...
| robotnikman wrote:
| I'm guessing this is why, as others have already mentioned,
| there seems to be an increase in use of x86 processors like the
| N100. You don't need to worry about stuff like specific vendor
| kernels.
| mschuster91 wrote:
| yeah, x86 BIOS/UEFI plus VESA provides for a not-great but
| decently usable foundation. Easy enough to hand-code a hello
| world in assembly - good luck trying to do that one on any
| ARM system.
| tliltocatl wrote:
| > any ARM system
|
| If you are talking bare metal rather than Linux support,
| many M-core MCUs are easy, and some of these (e. g.
| STM32H7) actually have usable 1995-level desktop
| performance sans MMU, ergo enough for many things that
| aren't web browsing. It's A-cores that's closed, because
| vendors have zero incentives to open them up and because
| the whole thing is a heap of Synopsys modules ducktaped
| together - and Synopsys has an less incentive to open up.
| And then of course, there are GPUs, that's not well
| supported even on x86. Video out - yes, you get it with
| UEFI GOP, but usually no multihead. You also get it on many
| ARM SoCs - the video output generator is sometimes
| documented, it's accelerator/compute that's universally
| closed.
| megous wrote:
| It's always possible to just go to dts directory in latest
| Linux kernel, and just look for what's enabled on SoC or board
| level, for a quick apprisal of level of mainline support.
|
| You will not catch all the details this way, drivers may be
| incomplete, or things may not be fully integrated to work
| together really well, but then that's usually fixable,
| especially if datasheets are available.
| yjftsjthsd-h wrote:
| Thanks for covering the software support side of things; that's
| always been the worst part of the ecosystem, and annoyingly hard
| to get info on.
|
| On which note: Oh, wow, the Radxa Dragon Q6A looks great!
| Mainline support, good hardware, good price. Once it's back in
| stock I may have to buy one:)
| topspin wrote:
| > Radxa Dragon Q6A looks great
|
| Indeed. With 16GB RAM, NVME, integrated GPU, 1Gbps ethernet and
| Wi-Fi, it's basically the mean 2017-2018 laptop. External Wi-Fi
| antenna connectors (plural) is eye opening. Too bad it's
| Quectel AIC8800.
| bertm wrote:
| This misses the NVIDIA Jetson/Orin/Thor line of GPU-enabled edge
| computers which are more comparable with Raspberry Pis. The
| Jetson Orin Nano Super is $250. Another feature which we find
| useful is the native NVMe drive support. For the price, this
| blows RPis out of the water for imaging applications.
|
| https://www.nvidia.com/en-us/autonomous-machines/embedded-sy...
| Havoc wrote:
| The dragon one sounds interesting. mainline support, cheap price
| point and npu/gpu would be great
| asadm wrote:
| I am working with Rockchip rv1106 which has 1 TOPS NPU but also
| some cool DSP features which has helped me do very interesting
| things.
|
| For my project I am making a SLAM/VIO using it, see video:
| https://x.com/_asadmemon/status/1977737626951041225
| klawed wrote:
| Surprisingly no mention of ESP soc's. Mane esp boards easily
| replace both arduino and raspberry pi zero projects with a single
| board.
| bityard wrote:
| Which ESP chips run Linux?
| synergy20 wrote:
| This could be improved a lot IMHO.
|
| NXP's IMX6/8 family is dominant in the market, really should have
| presence there.
|
| And TI's ARM series, still popular in beagleboard family and used
| widely in the field.
|
| And Intel's N100/N150, though I'm unsure if they're still "SBC"
| boards.
|
| Allwinner, ESP32 are also major players from China.
|
| Now apparently Qualcomm is entering the fight with new chips on
| Arduino UNO Q.
|
| NXP and TI are both open source friendly, unlike broadcom. My
| first choice will be NXP.
|
| Never heard about CIX.
| adboc wrote:
| The biggest problem for me with Raspberry Pi alternatives is the
| Linux support. Given the fact that getting mainline support is
| hard & long, I like the approach RPi has taken, to fork the
| kernel, add support for most parts of the SoC, upstream it and
| rebase the fork regularly. And I can still use my RPi model B
| with the latest OS they release. For example, I bought
| RK3588-based board and the officially supported kernel version is
| 6.1. I know that Collabora and other people are working hard to
| have upstream support [1], but it will take time until all IPs
| are covered. Is there any alternative that has Linux support
| comparable to Raspberry Pi?
|
| [1] https://gitlab.collabora.com/hardware-
| enablement/rockchip-35...
| bmurphy1976 wrote:
| Since they're all RPI alternatives anyway and you don't get the
| ecosystem benefits, you should try an Intel N100. I switched my
| personal services over to one of those a couple years ago, and
| it's a great bang-for-your-buck small server. Being an Intel
| chip, stock Ubuntu _just works_. I 've had no compatibility
| issues.
| adboc wrote:
| N100 indeed looks like a good alternative. I own one
| N100-based mini PC and I see there are some N100-based SBC as
| well. x86-like support for ARM/RISC-V SoC would be a miracle
| ;-).
| bmurphy1976 wrote:
| Yeah, I love it. Losing access to the RPI ecosystem addons
| kind of sucks, but I found I don't really use them anyway.
| I think you can get a USB GPIO if you really need that, but
| personally I've moved more towards N100 for services, ESP32
| for devices.
| taffronaut wrote:
| I'm not sure that the RK3688 and a big chunk of the article spent
| on its specs belongs in a "State of Embedded: Q4 2025 Overview"
| given that it's due sometime in 2026. I'm sure it's going to be
| great but I suggest it belongs to a future state.
|
| On the other hand, CIX have been putting actual Arm v9 hardware
| in developers' hands for some time.
| HeyMeco wrote:
| First of all thank you everybody for reading and commenting! I
| want to address some of the comments I got regarding the article:
|
| - xyz vendor is not covered: For this there is a simple answer, I
| don't have the hardware so I can't make a fair judgement when
| reporting about it. I don't want to be another reporter of basic
| benchmark scores & 1080P Youtube playback but actually show off
| the hardware capabilties with the right software. Hopefully this
| will be possible with more in the future as this project grows
|
| - This is only one sector of the space why don't you have micro
| controllers or x86: While I want to cover all aspects of the
| space I am not an embedded engineer. I started during covid with
| the goal to replace my X86 homelab server with an ARM one to save
| power and got deeper into the rabbit hole until I ended up
| maintaining some boards and doing some Debian/Ubuntu based bring
| ups in my freetime. This led to me wanting to have one place to
| share my findings along the way and document things that might
| leave one stranded in the world of Yocto/U-Boot/Linux
| Kernel/Device-Trees/etc. and I created sbcwiki.com, not only for
| me to share my findings but for others to contribute with simple
| markdown files to the GitHub repo too.
| sheepybloke wrote:
| A lot of the chips have started including NPUs. How are
| applications supposed to access that acceleration now with
| embedded Linux? Does linux handle this for you, or do you need to
| leverage some specific drivers like CUDA?
| dmitrygr wrote:
| vendor-specific drivers are the norm. Usually on an obsolete
| kernel version which never gets updated.
| HeyMeco wrote:
| Ideally things are going to adapt the Mesa Teflon framework.
| See here: https://docs.mesa3d.org/teflon.html
|
| Available for NXP IMX8M, Amlogic A311D and RK3588
___________________________________________________________________
(page generated 2025-10-29 23:01 UTC)