bit of a tangent perhaps, not really even relevant to the article: i am really very hopeful that we will finally see the 'democratization' (is that the term?) of intergrated circuits, with projects like Google's SkyWater PDK and eFabless involvement, perhaps being a nascent seed that hopefully leads the way.
it really does feel to me, like the final frontier of old-style closed-source "I have $20M capital and you don't, so screw you, we make the decisions around here, and no we aren't publishing the datasheets or firmware"-world, i dare say the final holdout.
imagine having an actual open source raspberry pi. You can read back the github sha id from a control register of the repo checkin that was used to cook the masks that were used to etch your chip, the gdsii masks and netlists are just a build artifact, the verilog is all there. nobody asks for the HDMI controller datasheets anymore, because its just a github link away to see where the control registers live. Questions regarding the 3D accelerator are answered with a quick perusal of the verilog.
its so obvious that this is the future destination - atleast, to me. its just a shame that its not around right now. that there is all this politics and heavy beurocracy that stands in the way, understandably so - this stuff is super expensive, and the investors want a secure return on their investment. but imagine how cool that would be. to have chips become as commonplace as github projects.
Cool project, but over time I understand the appeal of the raspberry pi less and less. It seems like a compromised device with perpetually terrible retail availability and bait-and-switch pricing.
An x86 mini PC from the likes of Beelink is going to be way better at the computer stuff. Or choose some other microcontroller like an Arduino, or maybe just add GPIO to your PC.
> Other than PI, what PC actually drives its own GPIO?
They used to be able to using parallel ports. When the controller was on the southbridge and accessible via DMA, you could get hard realtime even on Windows using device drivers and interrupts (pre-Vista, which changed the security model). There was software like Mach3 that could control a whole CNC off of a parallel port and lots of hardware companies used Wintel instead of SBCs or rolling their own PCB. San Francisco BART still uses parallel ports to control their trains and they have to hunt down for replacement parts on eBay [1].
Nowadays a PCIe GPIO card is at least a few hundred dollars and you're unlikely to get hard realtime guarantees unless it supports uploading custom firmware. Sharing functionality with the most common printer interface was a boon for PC GPIO but that's gone now.
Back in the late 80s before ubiquitous digital audio, someone published a design for a sound card that used the parallel port and a super simple circuit with just some resistors. I found a similar recent one:
No, the controller used to be part of the motherboard chipset in a way that allowed DMA access to GPIO registers (so the CPU didn’t need to actively bit bang the port, drivers could just fill a buffer and offload it to the hardware) and interrupts that had priority over all other processes.
Intel N100 and N95 MiniPCs are cheap as chips, the last one I bought was $60 and had 8GB of RAM and 256GB of NVMe SSD, plus it included the power adapter and two ethernet ports.
Mind sharing which one you bought and from where? I was looking around at a few and they seemed to start at around $120 shipped with no RAM nor storage.
It's because beelink is building up a name that makes them able to market their products at a premium price. Many other brands of these are much cheaper. My 8/256 one was 120 and that included an nvme. The pi with adapter comes to about the same and does not include storage and is also much slower than an N100.
Not really! I found one for $129 at 8GB RAM/256GB SSD. To get an equivalent RPI you start with $80 for the device, get up to $100 for the power supply and case, get up to $114 to add the m.2 hat, and then you still have to buy your SSD.
The n100 is comparable to a pi5 performance - up to 2x for certain workloads. Where are you getting 5x from?
Most people are buying pi for the form factor - I don’t think I’ve seen an n100 that isn’t more than 3x the size of a pi in a case - and the smaller size ones come with a hefty premium.
I think some people cherry pick a certain benchmark or reference point and build up the N100 as some sort of monster chip from that.
It's faster, yes. Also less efficient. But in many tiny builds it's only marginally faster because it has to be thermally constrained. Otherwise you require a larger power supply and bigger fan/heatsink.
Also, some configurations end up slower due to cheap shortcuts manufacturers make in their tiny builds.
The N100 is great, but it's not as great as I think some people think, for all uses and purposes!
Some of this won’t apply to the 400 because it’s in a case but…
They’re cheap. If you want to play around with another OS or something you don’t have to risk your main computer. You can also run them 24 seven even if your main computer is a laptop that you turn off or move from location to location. The Pi can just sit there. Also great if you’re gonna put it somewhere it might get damaged like outside for a project. It’s something happens you won’t lose too much money.
They are a known quantity. Sure there are lots of little x86 boxes but everyone knows exactly what’s on a Pi 7B++. That means pre-configured images for lots of Linux and BSD flavors and probably other things. Plus you can write your own OS!
Lots of expansion stuff is available too in the form of hats, on top of the built in I/O.
All the above gives it a big community. If you need help with a project or want to find out how to do something there are people and parts ready to help you.
Of course it’s quite efficient. I have a tiny GMKTek computer but it uses quite a bit more power IIRC. Need to run it off a battery or solar power? That’s very useful.
Until the clone boards started to show up wasn’t it also basically the easiest and cheapest way to get a modern ARM computer at all?
I’m not sure it’s a silver bullet for anything. But it’s a decent option for a lot of things especially.
Not sure I see that, aside from during the chip shortage when CM4 in particular were like hen's teeth. Aside from post-release shortages, I've always been able to get whichever model I need, at a non-scalped price, typically with next day delivery.
As someone who has purchased a half dozen SBCs over the years but who has also switched to mini PCs the other thing that you start to miss is having a real BIOS so you can install a vanilla operating system. With the SBCs, there is often a single Linux distro that is supported, and which rapidly stops getting updated. Even using stuff like Armbian, it's challenging to use an SBC that is more than a couple years old, and I've had bad experiences where the NIC stops working after a software update.
I think this post is a good example of why people buy rpis: network effects. Are people writing articles like this about the competition? Especially if you're in to modding, hacking, etc the community around the product is an important consideration.
> Indeed. I think for most of us, seeing the pads there, but unpopulated, was a giant head-scratcher. The Pi 500 would've been more of a slam-dunk win with the slot in place, even if empty.
Hearing that coming out of a HAM radio enthusiast whose father is a broadcast technician is a little strange.
The most likely explanation is either RF interference; NVMe drives suck down enough power to substantially self-heat during big transfers and I bet that causes all sorts of interesting RF problems, especially given the Pi isn't sold in a case.
Or power sag issues. The Pis have always been infamous for their terrible power supply designs, which seem to be intentional to give resellers a way to bend people over. It's the old "every phone you buy has a different plug and you need $100 in AC and phone adapters" nonsense we used to have to deal with to pad retailer margins.
I remember Motorola designed one of the Nexus phones to require a proprietary fast-charge protocol so that Verizon, who had retail exclusivity for a bit, would get to bend people over for the stupidly expensive charger. Verizon also got Google to permanently fuse the bootloader so installing an alternative OS was impossible. On a nexus phone.
What percentage of buyers would ever use an M.2 slot? How many people would avoid buying it because it lacked the M.2?
I suspect the return would never match the parts cost. While there must have been dev time spent designing it, it’s basically free to print on the PCB right? And it’s not like they didn’t have the free space.
Interesting RPi thinks it might be useful when reusing the board in future designs. I’d be curious if those ever come.
It's cool that it's been solved but I'm thinking having a 3.3v bench power supply breaks the practicality of the PI500. /s
Yeah yeah, just get a converter. I think the real reason the Pi Foundation didn't make the slot available is because they don't want to have to support the influx of broken cases when hordes of people tear open a product that isn't meant to be opened.
I'd say it would be more expensive to make it like that? There's already enough complaints about how 'expensive' the Pi line has become (I don't think they are in reality taking inflation into account, FTR), so adding a few £ would mean they would be more expensive again.
In addition, there's not that many chances for performance upgrades - there's only been 2 generations in the current form factor, and that could change again depending on where the Pi6 line goes?
> I'd say it would be more expensive to make it like that?
Only marginally. And they probabky lost a lot if sales because of that. From the very start, a pi 400 with 4GB of ram was a non starter for me in 2020, and a 8GB one is the same for me in 2024. However if they had said we guarantee we will support this board for 8 years with new compute modules, I may have bought one and moved the cm4 to a different headless compute module board when I would have upgraded it.
And that may have opened a market had they decided to create a standard motherboard form factor for arm cpu modules in a keyboard.
42 comments
[ 3.6 ms ] story [ 116 ms ] threadit really does feel to me, like the final frontier of old-style closed-source "I have $20M capital and you don't, so screw you, we make the decisions around here, and no we aren't publishing the datasheets or firmware"-world, i dare say the final holdout.
imagine having an actual open source raspberry pi. You can read back the github sha id from a control register of the repo checkin that was used to cook the masks that were used to etch your chip, the gdsii masks and netlists are just a build artifact, the verilog is all there. nobody asks for the HDMI controller datasheets anymore, because its just a github link away to see where the control registers live. Questions regarding the 3D accelerator are answered with a quick perusal of the verilog.
its so obvious that this is the future destination - atleast, to me. its just a shame that its not around right now. that there is all this politics and heavy beurocracy that stands in the way, understandably so - this stuff is super expensive, and the investors want a secure return on their investment. but imagine how cool that would be. to have chips become as commonplace as github projects.
An x86 mini PC from the likes of Beelink is going to be way better at the computer stuff. Or choose some other microcontroller like an Arduino, or maybe just add GPIO to your PC.
By this I assume you mean “dangle an Arduino off the side”. Other than PI, what PC actually drives its own GPIO?
They used to be able to using parallel ports. When the controller was on the southbridge and accessible via DMA, you could get hard realtime even on Windows using device drivers and interrupts (pre-Vista, which changed the security model). There was software like Mach3 that could control a whole CNC off of a parallel port and lots of hardware companies used Wintel instead of SBCs or rolling their own PCB. San Francisco BART still uses parallel ports to control their trains and they have to hunt down for replacement parts on eBay [1].
Nowadays a PCIe GPIO card is at least a few hundred dollars and you're unlikely to get hard realtime guarantees unless it supports uploading custom firmware. Sharing functionality with the most common printer interface was a boon for PC GPIO but that's gone now.
[1] https://news.ycombinator.com/item?id=32884814
http://jubatian.com/articles/building-an-lpt-dac/
USB simply doesn’t work like that.
And a minimum of 5x more expensive, and more like 10x in most cases.
I have several Pis doing various work, some 3Bs, some Zero Ws, some 4B.
The 5 just does not make sense to me and I don't see myself buying it unless the price radically changes.
[1]: https://news.ycombinator.com/item?id=42365555
Plus, the Beelink can be upgraded to 16GB of RAM.
Meanwhile the n100 is like 5x processing power and is natively supported by any software since it’s x86
Most people are buying pi for the form factor - I don’t think I’ve seen an n100 that isn’t more than 3x the size of a pi in a case - and the smaller size ones come with a hefty premium.
It's faster, yes. Also less efficient. But in many tiny builds it's only marginally faster because it has to be thermally constrained. Otherwise you require a larger power supply and bigger fan/heatsink.
Also, some configurations end up slower due to cheap shortcuts manufacturers make in their tiny builds.
The N100 is great, but it's not as great as I think some people think, for all uses and purposes!
N100 have a significantly faster ssd, gpu and faster ram.
You can get Radha x4 for similar size too
They’re cheap. If you want to play around with another OS or something you don’t have to risk your main computer. You can also run them 24 seven even if your main computer is a laptop that you turn off or move from location to location. The Pi can just sit there. Also great if you’re gonna put it somewhere it might get damaged like outside for a project. It’s something happens you won’t lose too much money.
They are a known quantity. Sure there are lots of little x86 boxes but everyone knows exactly what’s on a Pi 7B++. That means pre-configured images for lots of Linux and BSD flavors and probably other things. Plus you can write your own OS!
Lots of expansion stuff is available too in the form of hats, on top of the built in I/O.
All the above gives it a big community. If you need help with a project or want to find out how to do something there are people and parts ready to help you.
Of course it’s quite efficient. I have a tiny GMKTek computer but it uses quite a bit more power IIRC. Need to run it off a battery or solar power? That’s very useful.
Until the clone boards started to show up wasn’t it also basically the easiest and cheapest way to get a modern ARM computer at all?
I’m not sure it’s a silver bullet for anything. But it’s a decent option for a lot of things especially.
Not sure I see that, aside from during the chip shortage when CM4 in particular were like hen's teeth. Aside from post-release shortages, I've always been able to get whichever model I need, at a non-scalped price, typically with next day delivery.
Hearing that coming out of a HAM radio enthusiast whose father is a broadcast technician is a little strange.
The most likely explanation is either RF interference; NVMe drives suck down enough power to substantially self-heat during big transfers and I bet that causes all sorts of interesting RF problems, especially given the Pi isn't sold in a case.
Or power sag issues. The Pis have always been infamous for their terrible power supply designs, which seem to be intentional to give resellers a way to bend people over. It's the old "every phone you buy has a different plug and you need $100 in AC and phone adapters" nonsense we used to have to deal with to pad retailer margins.
I remember Motorola designed one of the Nexus phones to require a proprietary fast-charge protocol so that Verizon, who had retail exclusivity for a bit, would get to bend people over for the stupidly expensive charger. Verizon also got Google to permanently fuse the bootloader so installing an alternative OS was impossible. On a nexus phone.
How does heat cause RF issues?
What percentage of buyers would ever use an M.2 slot? How many people would avoid buying it because it lacked the M.2?
I suspect the return would never match the parts cost. While there must have been dev time spent designing it, it’s basically free to print on the PCB right? And it’s not like they didn’t have the free space.
Interesting RPi thinks it might be useful when reusing the board in future designs. I’d be curious if those ever come.
Heat throttling might be an issue but it's still going to be way faster than a microSD.
Yeah yeah, just get a converter. I think the real reason the Pi Foundation didn't make the slot available is because they don't want to have to support the influx of broken cases when hordes of people tear open a product that isn't meant to be opened.
(either that or leave an unpopulated footprint for a varta battery to leak all over the pcb)
I was interested in the concept of a pi400 but never bought one because there was only one spec and there was no way I could update it in the future.
In addition, there's not that many chances for performance upgrades - there's only been 2 generations in the current form factor, and that could change again depending on where the Pi6 line goes?
Only marginally. And they probabky lost a lot if sales because of that. From the very start, a pi 400 with 4GB of ram was a non starter for me in 2020, and a 8GB one is the same for me in 2024. However if they had said we guarantee we will support this board for 8 years with new compute modules, I may have bought one and moved the cm4 to a different headless compute module board when I would have upgraded it.
And that may have opened a market had they decided to create a standard motherboard form factor for arm cpu modules in a keyboard.