$50 2GB Raspberry Pi 5 comes with a lower price and a tweaked, cheaper CPU
arstechnica.com
arstechnica.com
I'll gladly pay the $30 for 8GB (4x) more ram. Pretty much every time.
If it's really a low resource situation, I wouldn't bother with a Pi and would go for an ESP* system instead. Like everyone already does.
I grew up on machines chronically starved, think 16MB for Mac system 5-7, and 32MB for win98 :) (which leaked like a mofo, reboot required every 20 minutes.)
It's only a bargain if you need it. For a general purpose computer I'm with you, but the Pi5 often isn't used as a general purpose computer so if you know the software isn't going to be using it then it is pure waste.
Pi's aren't a great value for general purpose computing when cube-PCs will get you orders of magnitude more power and functionality for 2-4x the price, and people often neglect to factor in the price of a case, PSU, storage which stack up the cost rapidly.
Every generation I grab a couple of the latest model-B and run a few quick desktop tests to see where they're at subjectively (no benchmarks, just boot it up and "use" it) and Pi 5 is still orders of lower than what I expect from a computer.
Put them to work in a specific application and their reliability and great ubiquity shine; I have a couple of Pis in 3D printers, one powering a touch-screen dashboard, a couple in my server rack for Hashicorp Vault, Unifi Server, ARM build server, a half a dozen or so in home-made speaker/TTS devices, a couple in retro gaming-handhelds, one in a dedicated WiFi AP for MiTM/testing/traffic-inspection, and one with a NVMe HAT running a file-server for low-power network storage.
Each Pi serves a purpose, and I've allocated a device a specific model based on its requirements; a couple of them are probably over-powered just due to having a specific Pi spare, but I try to shuffle them around to be using the weakest/cheapest Pi available that suits the purpose.
... don't forget to add more cooling for that reliability
I don't see 3 Bs in stock in my usual places. 4s and 5s are plenty though.
Eh, not orders of magnitude more. An RP5 is like 25% of the speed of an Apple M3.
(Don't ask how I know it)
Once you are using a lot of paging, chances are performance has fallen through the floor to the point where doing anything to fix it (killing run-away processes, closing other processes (killing those that won't go willingly)) takes so long that you are fighting a losing battle. This is less of an issue with fast SD drives than it is with spinning rust (or slow SD cards / eMMC with some smaller devices) but still essentially the case.
I just have enough swap defined for some small unusual circumstances and assume that if I need more than that things are falling over anyway so OOM taking things our is not going to make it worse (if they continued to thrash I'd have to kill them anyway).
Ubuntu does this by default and has done since 18.04 or so, maybe earlier.
Only issue is, the distro they promote is a bit dodgy.
I worked with high-power EV chargers.
Each system had 4 RPi 4s. (Some due to 3rd-party hardware, some powering touch displays).
The main system started out with 4GB, but 8GB was more convenient for developing.
The systems generated a massive amount of metrics that got buffered in memory.
8GB memory is convenient when your only alternatives are SD-card and EEPROM.
Well, if that's true, have to agree raspberry pi may be better than esp32.
I'd be surprised if you can't buy a GPIO board that connects to USB for a few dollars on AliExpress.
But yeah, if you don't care about timing, that will work.
Rescue old hardware!
If someone's paying me to build something and spending that $30 saves me a day, I'm doing it.
Moving your product to an MCU, even when it would support it well (which is not the case for a lot of RPi uses) will usually be a lengthy endeavor. But it's just HN being HN, I guess.
Here's an example. A few years ago I was contracted to build a device that would input some data, do real time signal processing, and output the data in a different format. The quickest and probably cheapest way to get it done was to use a $15 analog chip to do the bulk of the processing. I based that board on an Arduino Nano (clone, of course!).
A year or two later, much faster microcontrollers like ESP32 had arduino support and were trivial to use. A single ESP32 would probably have replaced most of the other components on my board, reduced inventory required, reduced power consumption and a host of other benefits.
But I never bothered because there was no economic pressure to do so. I was selling a $300 device to a single customer who didn't care about the price, and I had better things to do with my time than redesign a board that worked perfectly.
TBH, if I wanted to, I could probably have increased the price of the current board by the same amount that redesigning with an ESP32 would have gained me.
This is one reason I avoid consumer products like the plague. I'd rather work with people to whom there's no effective difference in cost between $300 hardware and $700 hardware because the benefit to them far exceeds what I'm charging.
I guess I'm agreeing with you :-)
Look at what all the best volume manufactured custom and smart devices are based on these days.. it's ESP :)
Esphome makes life easy.
Your Raspberry Pi will have some difficulties running from battery for months while your ESP will already have a hard time driving a reasonably high resolution display for HMI
The usual HN discussion of RPi goes something like this:
"The RPi 5 has gotten too expensive, these days intel small-form-factor PCs (possible second hand) are a better choice, giving better performance and no messing around with microsd cards"
"I need GPIOs, sub-3-watt power consumption, and I don't need the performance of the most expensive RPi model"
"If you need GPIOs and low power, why not ESP32?"
That chip is ancient, better get a ESP32-C3 or ESP32-C2
The price is already uncompetitive for the hardware so might as well throw in all the bells and whistles, since anyone who's buying these is doing so for the software support ecosystem anyway.
But why? SBC are not supposed to be used as a general purpose computer. They're supposed to be used for some single task. If this task requires 2 GB RAM, why would you pay more for unused RAM? That makes little sense.
Of course if you're building a personal computer, home server or something like that with unspecified tasks, that makes sense.
Why not? We have smartphones with ridiculously powerful specifications. I use Termux for software development. No reason we can't have computers in the same form factor.
I'm planning to buy a clockwork pi uconsole specifically because it's a handheld Linux computer with a physical keyboard.
Funny sentiment on a site called "hacker" news. Use them for whatever you like, the more creative, the better.
However, there are plenty of people using Pis in the maker space and embedded world where the device will just run one piece of software for the rest of its life, and often 2GB will be enough. This device is for those types of use cases, and offer substantial cost savings for them.
Assuming prices for this type of device rise at the same rate as CPI.
In practice, the inflation rate for most sectors is very different to the CPI average, so these “adjusted for inflation” numbers are pretty meaningless.
The Pi 1 A+ is currently ~£20 including taxes: https://www.rapidonline.com/raspberry-pi-model-a-75-1131?src...
Disposable Income (inflation-adjusted or proportionate to whole income) tends to go up and down with a different pattern to inflation (or housing which makes up most of inflation most of the time) but represents more the difference in cost to you for a consumer good (especially elastic ones). The picture is much less rosy but represents the median person this way, not tech workers.
That's a good point that people often ignore. It's especially relevant when talking about technology, where pricing tends to fall, not rise. Pi feel like they should cost like $20 now considering they make them at way higher numbers than in the past and that the technology isn't anything particularly unique.
You do have the launch Pi feature set available in the Pi Zero which is much less than 20 USD.
Wasn’t there once a 5$ version?
The Zero was $5, the Zero W was $10 and the Zero W 2 is $15.
If you need the RAM then you're very quickly approaching N100 prices, and the N100 offerings give you way more expandability and performance.
I just don't get who the RPi5 is for, except maybe companies using them in embedded applications.
There are more performant and efficient alternatives, but for many hobbyists, prototypers, startups, and students (or rather, their teachers) it's not worth the extra time/headaches/hassle to start adapting an rpi solution to non-rpi hardware.
What I said was, I don't get the RPi5 value proposition. It's fairly expensive so not just something you throw at simple tasks like an NTP server without consideration.
If you need grunt and RAM, well, why not just pay a bit more and get tons more of that? You get more PCIe, you get way more RAM, CPU and GPU, in a form factor that's not that much larger. And it's x86 so you're not facing the kernel support issues like OrangePi and friends.
In my experience, there's very little overlap between "need hardware GPIO pins" and "need 8+GB of RAM and/or lots of compute".
Totally in agreement with you though, even as a huge fan of Raspberry Pi, the Pi 5s are overkill for the tasks they're capable of, and overpriced and underperforming in the tasks a mini-PC can do for 2-4x the price (maybe less once you factor in a PSU, case, storage etc).
Yea I included that under GPU, but spelling it out is a good point. A major step back. I use some of my Pi4s with a camera module for other uses, and rely on hardware encoding. The N100 can encode 4+ 1080p streams in hardware without breaking a sweat.
> maybe less once you factor in a PSU, case, storage etc
Most certainly. For a mini-pc application you'll need special USB power supplies, since the Pi5 doesn't fully support USB-PD (it won't accept >5V), so the USB-PD supplies you got for your phone, laptop etc probably won't cut it. And to use M.2 storage you need a separate HAT. And you want a case for it, and some cooling. Between those three, price for a 8GB RPi5 comes very close to a N100 box.
Now you mention it; I recently put together a Pi5 setup for low-power file-storage (portable), ~£80 for the Pi, ~£20 for the PSU, £5 for the cooler, £12 for a decent MicroSD, £16 for the NVMe HAT and £60 for an SSD, no case (total: £133), compared to the MinisForum Ryzen cube-PC I bought for a low-wattage server which cost ~£230, has 16 core Ryzen Laptop CPU, 32GB RAM, 1TB SSD, PSU, nice case, etc; It's almost as portable, and orders of magnitude more powerful. Ok there's almost 100% price difference, but you get way, way, way more than 100% more of a computer for the price.
The cube PC has really nice build quality, 32GiB SKHynix RAM, an extra PCIe slot, space for an additional DATA 3x4K display outputs, 2.5Gb networking, plenty of USB-C ports (don't remember which speeds) and a few USB-A (3.x) ports, active cooling, and although an anti-feature for me, came with Windows 11 out of the box (I blew it away and installed Debian) on a 1TB Kioxia NVMe drive.
I use it as a headless server but it's a powerful enough system that I'd absolutely use it as a desktop replacement for day-to-day work (development).
But sure, if that's a concern then the Pis or clones might be a better option.
A system that's appropriately designed for low power draw should be notably less than that.
I've got an N305 laptop that I'm using as a home server that idles at 4-5W.
https://jellyfin.org/docs/general/administration/hardware-ac...
It's an open-source 3D printer design, you want video encoding for your camera stream on the printer.
For projects intended to be built by ordinary people, a Raspberry Pi is a very accessible, reliable and well supported device with a track record, it's a no-brainer.
Your software might be targeting ARM, Pi boards are likely lower-power consumption than an N100 board, your application may not need the compute of an N100 (My Voron 2.4 is using about 20-30% CPU on a Pi4, while streaming (hardware encoding) a 720p camera feed, and printing at 400mm/s); the N100 launched last year; many projects built around Raspberry Pis are much, much older than that. There's _plenty_ of reasons to use a Pi over the N100. Also, N100 isn't an SBC it's a CPU, so who are you buying the finished board from, and what can you expect from them in terms of software and support? We know very well what to expect from RP at this point.
I'm not saying RPi is the answer to everything, but the N100 isn't the answer either.
[0] https://www.vorondesign.com/
P.S. To whoever downvoted; use your words.
That said, there are still plenty of legitimate reasons to do hardware video encoding, as opposed to _transcoding_, which probably is what you're thinking of.
But from what I understand Pi 5 still doesn’t have any HW encoder? So if your task relies on that you it basically loses its main advantage against the N100, lower power usage since you get it for free with Intel.
Pi4 _does_ do hardware video encoding, so that is still a valid option for many applications; RP has a good record, decent board design, great software support and a huge community; if we compare with Radxa and their X4 for example; their track record is one of poor board design, poor software support and a miniscule community.
My personal opinion, and it's just that, my opinion based on my experience;
Unless I'm looking at an "embedded" task (board needs to go _inside_ a product), I wouldn't use an SBC; If I'm running Plex/Jellyfin/etc, I'd use a mini-PC or "bigger"; something with a better board design (there are questions about the board design of the Radxa X4 with N100 given its power envelope), proper cooling, better storage and GPU etc.
Your writeup is making me somewhat regret-im-advance this purchase. https://www.jeffgeerling.com/blog/2024/radxa-x4-sbc-unites-i...
I was interesting in looking at the Radxa Zero 3 after one of your videos; for an application where I need the Zero form-factor but more power than a Zero 2W, unfortunately a quick look at the Radxa forums left me with the impression I'd spend half of my time fighting with poor software and general board issues, and there's only so many rabbit-holes I have the time and energy to go down these days.
Knowing you've tested a good selection of boards, can you think of anything in that FF that runs mainline Linux, or something close?
With the popularity, one would assume we'd be able to get a CPU that's not a decade behind smartphones. Mine is now gathering dust; it's inefficient for the power it consumes.
I am not so sure about that unless maybe you mod them to extract everything from the original casing and add heatsinks and possibly a fan.
Also from a safety point of view, you'd need to make sure those can still be powered up without their batteries as I wouldn't trust a +10y old smartphone battery on a device that is running unattended 24h/365d/year.
one is the primary database, another runs everything else in docker containers.
price-wise, though, Orange Pi is unbeatable. the Raspberry Pi has no business being more expensive than an Intel NUC
Plus the Banana Pi M4 has eMMC.
Seems like an odd situation. Did Broadcom want to sell the BCM2712 to other customers and insisted on the more expensive design?
I had a Pi 4 home server -- described in the blog post below -- that ran for several years with maybe 2 or 3 unplanned outages in all that time.
I am currently using a Raspberry Pi Zero W as a wireless security camera. It has been running continuously for over a year without a single crash.
Price no longer match the reality, if you haven't already, look elsewhere, much better alternatives exist, much cheaper
... but without the software support and documentation side, be it upstream kernel support, libraries for interfacing with stuff like acceleration or GPIOs or whatever.
The biggest advantage of the RPi ecosystem is the community. Unless you're dealing with absolutely cutting edge stuff involving PCIe, chances are very high that someone else has already seen the problems you are facing and documented solutions.
This is a very cynical take on future events, but I wont be surprised if it goes this way, as was there actually a need to IPO? We already had a massive shift away from hobbyists to industrial/commercial clients, before the IPO, so I don't see it getting better for hobbyist/enthusiasts. Costs will increase more, as they will see how their industrial clients make money from using a Pi in manufacturing process, and so the shareholders will want more of that slice from their clients.
The problem is that the competition in embedded is generally an utter pile of shit. Good luck even talking to some vendor if you're not willing to order 100k or more and don't have inside connections, BSPs (bootloader, kernel, userland libraries) are generally years out of date forked versions that barely build, and serious documentation is often behind NDAs...
RPi has a lot of room to sink before it gets threatened by the competition.
chip pricing is the crux of the matter here.
The equivalent of to the ESP32 would be the Rasperry Pi Pico 1 / 2 / W. They start at $4, which is a fair price.
ESP 32 S3 can now run Linux and for most tasks it's much better than a Pi and dirt cheap (about 1 USD)