Broadcom, having never wanted anything to do with this market since volumes were too low, had an abundance of a CPU SKU that was good for this. So some broadcom engineers founded Raspberry Pi to use up this excess stock, essentially getting these chips for free. The original RPi blew every other SBC out of the water on price performance (and many manufacturers out of the market) because by getting the most expensive component for free, they could sell Pis for an extremely low price. It also massively expanded the market for SBCs, as hobbyists flooded in to work with RPis.
5-10 years ago, the sweetheart deal with Broadcom went away. Now Raspberry Pi has to compete with everyone else for Broadcom SoCs, and during the semiconductor shortage of 2020, Broadcom had tremendous leverage. Now, Raspberry Pi pricing is nothing special, but they still have the brand name and they have captured the community (on the back of behavior that was borderline anticompetitive).
The two pieces that have to be in place, as 'threshold' requirements are
1) The SBC exists and is available from a manufacturer
2) The same manufacturer provides an OS for that board and its associated board support package (BSP) which is drivers for all the I/O and system support functions.
The industry is full of people who went out of business because they chose Vendor A's SBC and Vendor B's OS, only to fail to deliver when it didn't work with Vendor A and Vendor B point at the other saying it was their problem. So people just don't do that any more.
What most vendors in the SBC space, prior to the introduction of the Raspberry Pi, didn't have was 20 to 30 thousand programmers writing random bits of code. What that meant was the Pi's feature set exploded rapidly, what's more there were lots of free tutorials on programming it.
In the SBC space before Pi that "Programmer Training" was one of the ways the vendor made better margins at $500/hr for a class of "up to 15 students" at our facilities.
So before, higher priced SBC + BSP, and you had to send your programmers on a road trip to the vendors facility to get the hands on training, and then you had to pay every time you made a service request.
After, cheap SBC + BSP!, a bunch of different programming videos on the web for free! Program doesn't work? Just ask the community of enthusiasts what they think!
We are not surprised a whole lot of the smaller SBC vendors closed down after that.
It will be interesting to see if the RISC-V efforts will allow for vendors to supply "50 year" SBCs (which is to say they are guaranteed to have pin-for-pin and bug-for-bug replacement SBCs for the period of 50 years.) That was a weapon system requirement back in the day that got waived for electronics because vendors couldn't "force" chip makers to keep making the chips they would need for repairs when there was no volume. But if you can define a socket/footprint and then drop in an FPGA "core" which can evolve but always appears to the circuit to be the same processor, that will be an interesting evolution point.
Before RPi became really popular and the 4 came out, I used BeagleBoard, PandaBoard, Aria G25, Gumstix, Cubieboard, i.MX devkits, and spun custom boards using Marvell, TI, Freescale (now NXP), and Qualcomm CPUs - I don't remember any of them having as good a BSP or being as easy to develop with as the RPi was five years ago. Maybe my memory is (very) faulty but the experience was leagues worse. PTSD-inducing level of worse. I wasted weeks or months on every major project shaving silicon yaks that should have been handled by the vendor (and is now handled by the RPi community).
The modern i.MX toolchain may now be comparable many years later but I've long since given up on everything else since CM4 came out in 2020.
My understanding is that they lost the sweetheart deal after they pivoted to supporting commercial, right in time for the pandemic supply crunch.
Two products.. leveraged a Yocto-based board support package (BSP); one being in AgTech, and the other being in the veterinary space. These products have followed disparate practices when leveraging the BSP for custom hardware and software.. this talk [described] the two products, how the BSP was customized and used, and the resulting consequences.*Source: have worked on both Pi based solutions and custom hardware with yocto.
It's the Arduino curse - sure, it's faster to ship Hello World with Arduino but soon you realize all your libraries were built by beginners and use delay() everywhere so you're screwed if you need two peripherals to work at once.
You use Pi for prototypes and one-offs, not where you need to ship something that's actually competitive on BoM.
Some applications are profitable at only hundreds of devices deployed. In which case, the moment you say "BoM" the hardware bill triples.
If you ship thousands to millions, of course you're right.
And you still can! The big innovation from Raspberry Pi was making it all feel very accessible through the documentation, community, and various utilities to configure things via menus instead of by editing files.
The Raspberry Pi was rarely the best board, but it was the easiest to recommend to beginners because you could point them to volumes of documentation and community threads.
But anyone may verify from public records e.g. compare and contrast resignation notice at Companies House with subsequent RPI blog.
> by getting the most expensive component for free
I get that you're exaggerating, and you perhaps aren't trying to be deceptive by misusing the word "free," but the low markup the RPi guys initially paid to Broadcom does not explain as much as you think it does about the Pi's success. To explain why let's examine something else you're wrong about:
> You could get a lot of different kinds of boards with good Linux support and a not terrible price. Often, a processor vendor would explicitly provide support for these things because it was a good vector for selling chips.
Prior to the Pi you could get one board with good Linux support and a not terrible price from a vendor who provided support because they thought it was a good vector for selling chips. That was the BeagleBoard, from TI. I mean, the BB barely checked all those boxes: the support wasn't very good, it was kind of nonexistent compared to the support community the Raspberry Pi people created. But they sure didn't have to worry about the cost of the CPU.
So back to the original point: getting the CPU "for free" (since we're apparently just saying "free" now when we mean "at cost") wasn't a decisive advantage for the Raspberry Pi people, since the TI people had the same advantage. TI had a few other advantages as well, like a first mover advantage and relationships with everyone who sells electronics components.
TI's support was crap when compared to something like RPi which deliberately targeted newbies, and as far as I remember they didn't have a few amazing people in their community dedicated to making a whole new spin of Debian and supporting it like RPi did. And, you know, PR matters. All that stuff is what made the difference.
> anticompetitive
I guess we're just throwing words around without caring what they mean today for some reason.
Broadcom sold the chips to the Raspberry Pi organization at a significant loss, not at a low markup or at cost. They were not free, but they were close to free. The TI guys never even gave anything close to "at cost" to the BeagleBoard folks.
Also, the BeagleBoard was the most hobbyist-targeted and the one with the best PR. There were, and still are, tens of companies making SBCs, but before the RPi they would all sell you singe unit quantity. Not any more. Most of them actually had better support than TI. The NXP boards in general were and still are my favorites.
Also, providing a product at a loss so that it significantly undercuts your competition (also called "dumping") is very much anticompetitive. I don't like using the term for Raspberry Pi because they clearly weren't out to create a monopoly, but the Raspberry Pi was dumping a product.
Everything in that paragraph is “citation needed,” unfortunately, but I certainly don’t mean that in a negative way. I wasn’t aware that Broadcom lost money on the early Pi parts and I feel a little skepticism about that. You seem to know something about what TI charged the Beaglebone people, I’m curious about that as well.
It’s not that dumping parts to establish a competitive advantage is beneath Broadcom - what’s beneath Broadcom? It just seems rather… prescient of them? Unless you were talking about the Pi 3 generation or something. I’d believe that.
On the point of PR, I don’t remember Beagleboard having anything comparable to Pi’s seemingly organic enthusiasm twelve or thirteen years ago. But I guess I’m not sure I would remember.
That’s not all meant to be some kind of RPi love letter. They were great at all that community building stuff, and in my view the best at it, but in light of the IPO it’s quite a joke.
Subsequent chips were specially made for the Pi by broadcom, and supposedly they didn't have as large of margins as other customers.
For me, its beaten by the more effective claim Broadcom sold it /at/ cost.
Such as?
You perhaps aren't trying to be deceptive by misusing the word "markup," but citation please for /any/ "markup the RPi guys initially paid to Broadcom"
Regardless, any basis of your "low markup" would suffice to show markup was non-zero.
...though given this is HN I would have to concede low could be zero! :)
For example, I owned an ECS Liva 2/32 that I got for $100 AR and a pair of ECS Liva X 2/32 that I got for $125 each AR. These used standard N2807 and N2808 Bay Trail-D processors (Silvermont), meaning they are OoO, run standard x86 binaries and distributions, come with storage onboard (which was far more reliable than the early days of fullsize SD cards), a proper adapter, a case, and USB 3.0, for essentially the same price you paid for a Raspberry Pi 1B once you factored in all the little extras.
Obviously if you want to do GPIO and stuff, the RPi is a better option, but that also competes with micros, as mentioned. And a lot of people ended up using it for some variety of non-GPIO IOT thing attaching some USB device to wifi/ethernet, or as a NAS, etc, and the Pi was terrible at these. It had massive reliability problems (unthinkable today) and I went from incredibly excited to bouncing off it hard and learning a lesson about the right tool for the job, and I think I wasn't alone.
Part of the problem was the decision to use USB 2.0 as a system bus. Everything including network and disk all hung off a single half-duplex USB 2.0 high-speed bus. And the Rpi firmware had a bug in the USB stack which dropped frames under load, so actually this was unreliable and could lead to corruption all on its own. It took over 2 years after launch for the foundation to fix this fundamental bug in the literal system bus of their hardware (and thanks to broadcom's closed documentation/blobs, they were the only ones who could see the info to fix it).
https://github.com/raspberrypi/firmware/issues/19
On top of that, people forget about the non-micro-SD card slot. The fullsize slot just let the card cantilever out into space... but the thermoplastic used to make SD cards actually will soften enough from the heat of the rpi over time to warp away from the contacts and lose connection. Even just stepping up to eMMC (not the most reliable) was a massive step upwards because of how bad the Pi was at not melting its SD cards.
And the power adapter situation has always been dire - most USB chargers were utterly unable to provide consistent enough power to avoid brownouts and corruption, and (while you could PXE boot) the hardware was not even capable of PXE booting without at least a helper SD card to load from. Today, it's dire once again, with the Pi 5 drawing 5A @ 5V from the adapter, which essentially no adapter on the market is capable of delivering.
The 2B and later models (especially 4B) are enormously faster, the move to OoO cores instead of literal a single-core A57 or whatever improved performance a ton. The change to USB 3.0 (5gbps full-duplex vs 480mbps half-duplex) and getting the USB frame drop bug fixed helped a ton (the pi couldn't even serve a usb hard drive at full speed). And the 4B moved to armv8 which has a drastically clearer standard for binaries than the armv7 situation, which was an utter mess at the start (softfp binaries were a thing for a bit with the early 1B releases iirc!). Admittedly Pi was better than any other arm device but the Liva just ran debian or ubuntu.
I actually think it was an utter failure as a product. I cannot imagine having to maintain a whole computer lab of these things melting (softening+warping) the SD cards, can't even PXE boot unassisted, can't get good adapters? Let alone the situation with firmware problems/driver quality and needing (at the time) weird binaries with a non-standardized bootloader etc. Why would you buy that instead of the Liva X for the same finished price?
The pivot to "these are for makers!" was a pivot, and then they pivoted again into commercial. Like good for them but it's always been a messy product with an uncertain target market.
It was always borne largely on the backs of enthusiasts shoving it into various home-server and IOT usage, most of which didn't actually need a GPIO. And the GPIO thing was always better served by the ESP8266 in most situations, and that ecosystem really matured almost in parallel with the Rpi. So sure, while the Pi got more suitable for actual GPIO hobbyism over time, so did everything else too. I regard the whole thing as fundamentally having been overhyped and a waste of time by people who (in most circumstances) really should just have bought a booksize or an off-lease OEM USFF desktop.
That was me. Got an old Core2Duo desktop for $10 at a surplus sale, that became the fileserver I was trying to build, which instantly ended the reliability struggle I'd been having (rpi lasted an average of 1-2 months MTBF in my homeserver usage) then I got a couple booksize and nettops and ended up with NUCs etc which are just a far better fit for what I'm trying to do.
To each their own, but again, I think really very few people are interested in them for the GPIO stuff, it's the "$35 for a little computer" thing that draws people in, but that's an impedence mismatch to expectations, that's not why you should buy a rpi/if that's your use-case then there are better options.
I also like your perspective on the bigger picture here.
Maybe there was a market of cheap and friendly to use sbc. But calling it hobbyst is an stretch.
I switched to the cheaper Intel mini pcs when RPi supplies were short, and scalping made it much more expensive. A whole PC with more ram, case, psu and faster storage that was faster for under $200 vs $150+ for an 8GB RPi 4 board only was a no brainer.
The Raspberry Pi 5's base model does start at $60 but its specs are too different for a comparison to be meaningful.
[EDIT] Oops, I hadn't realized the 4B with 1 GB was discontinued. So the starting price of the 4B would be $45 for the 2 GB version.
The pi is fun but honestly for pi hole or similar you might as well buy the all in one x86. For media streaming definitely buy the all in one x86. For gpio stuff ok the pi is reasonable but even then if you want to make a product rather than a home automation once off you’d go a different route completely.
Of course, the Pi is also facing competition from higher end microcontroller based solutions. People seem to forget that there was a time when hobbyists bought the Pi for "Internet of Things" like projects, both due to its cost and size. Then came the ESP8266 and ESP32 and development boards that packaged both a microcontroller and network interface.
Most makers I know have pivoted to ESP32 during this time as it was good enough and actually available. Probably would have happened sooner or later though.
When the first raspberry was introduced, it was really really hard to interface an electronics project with the internet. Arduinos were really dumb at the time. That's why the raspberry was so ideal. But most electronics projects didn't really need a whole linux distro running on it. It was just that there was no other option.
The ESPs introduced a totally new class that can cover most of the usecases of electronics projects that the raspberry originally aimed at. They support wifi, bluetooth, pretty serious processing, enough for most connected projects. They totally ate the raspberry's lunch in the embedded market. Of course the RP2040 aims for this too but IMO it's kinda too little too late, the ESP32 is already so well established and has the biggest community.
At the other end of course the PC pricing came down and the intel N100's and the like eat its lunchon the other side.
In terms of driver support these intel systems are the easiest to work with in Linux. Very mainstream and well established drivers.
Requirements may vary.
And I've got no regrets about using a Pi 4 a router. It's sitting on the shelf next to me, and has been trouble-free for over four years now.
You may not care but I do.
I suppose a Pi 4 is fast enough to route 1 Gbps even through a usb network card?
[I do have one "sitting on a shelf" as a home Minecraft server btw. I hate the usb ssd hanging off it.]
I upgraded my home LAN to 10gb/2.5gb early last year, and the RPi can't really keep up with it anymore. I'm using an N305 based mini-pc with 4 ethernet ports for my router now... exceeds my needs and runs great. My home server is an AMD 5900HX based mini-pc with also works very well (ProxMox, Docker, etc).
I mean: There's not much to it. LAN port, WAN port, MicroSD storage, and a (completely optional and of dubious merit) hardware RTC that sits on the GPIO header. I'm using the bog-standard Pi Foundation plastic case.
It runs OpenWRT, which itself is very light-weight -- it's more-or-less just enough Linux to set up routing functions using a GUI, and it's meant to be able to fit in the tiny little bit of flash that consumer router hardware comes with.
IIRC, the status at the time when I put it together was that it could route packets and perform traffic shaping at ~1Gbps even with a USB NIC. It certainly did just fine when I had a 400Mbps pipe.
I'm willing to pay a little more for a respected brand with a little more QA involved (and less hassle to me as a developer).
It bugs me to no end how much a small (4-drive) nas can cost, compared to much more capable mini-pc prices.
Options with mainline linux kernel drivers:
- MCP2221A (i2c + 4x GPIO)
- CP2112 (i2c + GPIO)
Options without kernel support:
- FT232H
- Arduino nano and clones
- Raspberry Pi Pico running some interesting firmware
- ESP32 running something like ESPHome (completely separate from host)
I've chosen the Pico for now and forked the u2if project for firmware and host support[1]. I also put together a generic ESPHome-compatible protocol server in python to tie my widgets to Home Assistant[2].
You can roll your own "RGB temperature sensing modem" and get/set values with `bash` and `expect` over ttyUSB, no one PWM controls RGB LEDs on a gaming keyboard straight from a task tray app. Trying to wrestle GPIO problems that way is going to be unnecessarily hard.
Currently it's running in "one light string per PIO" mode but there's an inaccessible mode where one PIO can do up to 8 strings simultaneously. The pins have to be physically adjacent, though, so it's a bit more difficult to set up in a generalized way.
Clarification please. I thought inaccessible meant cannot do :)
Is it? I can buy one for $35 at my local Microcenter, and I can see that Adafruit is selling them for $35 as well. [0] It's still listed as a $35 part on the official Raspberry Pi product page. [1]
[0] https://www.adafruit.com/product/4295
[1] https://www.raspberrypi.com/products/raspberry-pi-4-model-b/
The power brick got more expensive due to the increased power use. Also if you want to make use of the full power, you need a cooler. But you can also do without cooler, because it just gracefully slows down when overheated.
I think the Pi drove the price of lower end computers down so far that people completely lost their sense of perspective.
Hey now. Proud owner of a 256MB 1B here. Although they upgraded the baseline spec right after that first production run.
For those unfamiliar with them, Digi-Key is a electronic components supplier to manufacturers but they also sell to individuals. Their stock count should be accurate.
Are you talking about the Pi5 4GB which is $60? We are talking about the Pi4 2GB which is $45. For many things, the cheaper, older version is fine.
The Raspberry Pi Zero 2 W is available right now for $15 with a 1 GHz 64 bit quad core CPU and 512 MB RAM.
So it seems to me that you can get a Raspberry Pi SBC today, that has higher or equal specs in every regard, for a lower cost than the original. Am I missing something?
Looking at https://github.com/raspberrypi/linux, I scrolled a few pages of commits and didn't even see one from outside the Raspberry Pi org. I'm picking through their merged PRs and it looks like maybe there are a couple, it's hard to tell.
But it looks to me like they just do a great job supporting their own product?
It’s just supporting whoever needs a RPi 1st generation. It’s the same reason some older models of things cost the same or more, it’s basically fishing for the “we need the exact same hardware” customer.
The Pi Zero models are also kind of priced to move to make a statement of “look how much power we can stuff in a $5 board”; IIRC they were at first sold at a small loss to hit that $5 price point.
They’ve got a full product line now, including systems that are comparable in cost but much more capable than the original Raspberry Pi.
This is one of the main advantages for hobbyists.
I am surprised by all of the comments here that assume Raspberry Pi has great upstream support. It’s amazing that people just assumed their boards were working great with upstream kernels. Raspberry Pi has a history of doing nonstandard things that serve their community but are actually a little bit quirky when it comes to normal embedded Linux.
0: yeah, I know you can run it in theory; I mean in a usable way.
They kinda work if the system is not mission critical. Just have them self reboot every 4 hours :)
Raspberry Pi 5 isn’t as well supported in mainline. You’re still going to be using their kernel if you want all the features, just like many other module these days.
> The only reason they don't cost $500 or more
$500 is a huge exaggeration. There are numerous modules and small boards well under that price that come with good support, including many with full x86-64 CPUs.
I think it’s more correct to say that the boards are approaching equilibrium with other boards and modules in price, not that they’re secretly some premium $500 product sold at a discount for reasons. Nobody would be buying Raspberry Pi anything at $250, let alone $500.
It's not a great choice if you're hoping to productize from it, for various reasons.
this is sort of sad. I remember reading advice given to the raspberry folks early on to keep things inexpensive, not cave into feature creep.
I guess now the price creeps upwards and the original target market has been left behind.
The only reason they don't cost $500 or more is because the foundation needs the hobbyist market to write and support the open source BSP, without which the RPi would be just another poorly supported also-ran in an already crowded market. With how well supported mainline Linux is on the Pi, EEs would be willing to pay a lot more.