From Eben (who did the original RPi hardware design, don’t know if he or one of his team did this one):
Above a certain scale, chip-on-board wins over having a separate module. And Raspberry Pi 400, with a 200ku initial production run, is above that scale.
I assume “wins” here is price, but this is the entire statement so it isn’t clear from his words or the context what the actual nature of the win is, that is really just my guess.
I’m very sure the RPi 4 is much faster then PCs were when PCI was created (June ‘92 is the market date for PCI)! If I recall they were around 100Mhz at the time? Maybe 200Mhz? Also still taking multiple cycles per instruction?
I’m very sure the RPi can fill a PCI bus.
The RPi feels a little pokey because it is low on RAM, and until the RPi 4 the I/O was all basically on USB2 so “disk” was also very slow. I mean don’t get me wrong it is amazingly fast (I/O included) for $35. When I got one for a 3D printing project I was expecting it to take forever for anything it had to compile and I was surprised by how reasonable it was. It was likely helpful that I was entirely using the command line so the compiler/linker didn’t need to fight with an X11 server for RAM, but still I was expecting it to feel mid-2000s and it was actually really did just feel like a reasonably quick little Linux box I had logged into remotely.
Sadly the Pi camera (the standard quality one) was pretty disappointing, I hear the high quality one is way better, but for me I’m not sure I want to triple the cost of my little side project 
PCI I totally agree with you.
The PCIe standard is a lot faster, though… while a Pi could probably keep up with an early PCIe card from 2003, a newest standard (2019 v5) PCIe card has 64gb/s transfer on the full x16 slot… so it would be where the limit would hit, and why you would need it.
It really is one of those I am having a hard time coming up with a use case for a PCIe port… the only thing I can think of that would make sense would be lab equipment or industrial equipment with a PCIe interface, but the drivers would likely need to be custom and generally that equipment is so expensive the PC isn’t even a factor in the price… and most of the equipment I’ve run into is USB, parallel, serial, or some special bus that the manufacturer licensed from a company in Albania in 1982 which only has an adapter for PCMCIA that only works on three models of compaq laptops from the '90s… 
I want computer legos. I already have a few Raspberry Pis. I don’t need another to address a new form factor, what I need is to take my current boards and use them in new ways.
After being a long time Gumstix user (10+ years?), I feel like we’ve gone backwards with little ARM boards. At least with Gumstix the board-to-board connectors are inexpensive and easy to mount on a single-sided PCB. It doesn’t do everything an RPi4 does, but it also does some things it cannot.
And a lot of those pins have obscure stacked alternative functions, like a parallel DMA port, RGB video output, etc.
Well I have to admit I have some egg on my face. I didn’t know until now that the Pi 4 Compute actually does have a PCIe connection (and still unsure if the Pi 400 does or doesn’t).
Ultimately I would like to have a Pi that can make use of a Cuda enabled GPU. A GPU specifically so it can be swapped between projects but I understand that the Google TPU (can’t recall the name) can be used with the Pi 3 and 4.
Inexpensive computers for kids is an unqualified good thing, but I’d argue that in 2020 the $100 commodity laptop has already filled that niche. That’s retail pricing, too; We’ve come a long way from the OLPC days.
This just feels weird; Like a 2015-era SBC computer released in 2020.
Ok, but you don’t need to run at full PCIe speed to “need” it. If you need to go anywhere past PCI but not all the way to the limits of PCIe you still need PCIe. So a fast SSD could blow past PCI. So would a graphics card if you wanted to use an external one (and do all the work to drive it). A drive controller for multiple drives? I would imagine lots of things, but nothing for any of the ways I have used my RPi systems (but I did choose RPi only for projects that didn’t need high performance I/O)
I haven’t used a $100 commercial laptop, but my understanding is they are not very tough. LCD displays and hinges being weak points. A system with no hinges or LCD display would by definition do better there 
Agreed, the Pi 400 will almost certainly last longer. I just wish the foundation had released something that was competitive on more than just durability. A keyboard + trackpad design with space for an optional internal battery would have allowed for wired or wireless(via phone/tablet screen) use without adding much to the final cost.
The video server and ham radio repeater uses are certainly interesting, but I would imagine that the implied use is the cliche of using a PI to emulate classic video game systems (the Pi is capable of emulating everything from the Atari 2600 of 1977 to the Playstation 1 of 1994, plus most coin-op arcade machines of this period). There are at least two Linux distrubutions (RetroPie and Lakka) dedicated to making a Pi a retro-console
Everything I’ve seen indicates that the heat sink on the 400 is absolutely up to snuff.
IMO (other than the 400 already having it), any Pi > 2 needs at least passive cooling, and with a small heat sink even a tiny fan is a big win. Running BOINC on a Pi 4, the fan on my Pi 4 hooked to the 3.3V supply, it stays at about 52°C, so I have plenty of room for overclocking should I decide to give it a try.
Oh, that would absolutely rock. Even the cheap RTLSDR dongles are fun, but having something like that right in a cheap single-board computer would be a game-changer. I’m sure that’s more of a Qualcomm thing, though. I doubt we’ll ever see any Qualcomm-based SBCs, at least not at prices that mortals can afford, and they’d likely have the radio bits locked out or very heavily restricted.
(ETA:)
And more ports than a MacBook!
Seriously, though, bringing out the 40-pin GPIO is awesome (and certainly in keeping with the spirit of Raspberry Pi). In a way, the BBC Micro has come full circle.
I always thought it unfortunate that more of the British 1980s computer scene didn’t make it across the pond. The Beeb had to be reworked for NTSC and was too expensive to survive in the market, Amstrad and non-BBC Acorn were nowhere to be found, and even the American computers popular there like the Atari ST and Amiga were not too common here.
Yeah, keyboard + trackpad (or stick mouse, or one of the rollerballs) would have been useful to more people I think, and as you say having space inside for a battery would have been nice, although maybe more important bringing the pins out for the LCD interface would have been great (as opposed to needing to use a LCD display that takes HDMI!).
Still I think, and this is a little odd coming from a “build something with it!” foundation, the Pi 400 is intended as much more of a finished product then the rest of the Pi line of products.
I’m not really in love with that keyboard, so to me it isn’t a great loss, if I wanted to make a “Pi laptop” I would use a different keyboard anyway, one that fits into this nice wooden box I already have…and spec’ed out for it…and realized I would build and never ever use because I have a really nice Mac laptop already, and an iMac. What do I need a homemade laptop for? I’m all for building things, but I like them to do a job after…
I see the appeal of having an “out of the box” keyboard/computer hybrid. Whatever I think having this on the market to bring more people to STEM is a great thing.
It’s cool, isn’t it? It gave me a little burst of joy to see so much easily accessible IO - a user port on steroids! This makes a really nice teaching/learning platform, where you can really see the direct link your code and some physical response (lights, motors, …).
Ah, the BBC Micro was great. Fond memories. I recently tried to control a circuit using the digital IO port on an Agilent vector network analyser. The instrument costs as much as a car, but the interface is terrible. You have to control it using Visual Basic, and nothing is clear. I eventually decided it would be faster to manually measure all 4096 states, gave up and triggered the reading from a BeagleBone instead.
Given that Gumstix are still for sale(and, alas, still cost what they did when I was lusting after them as an impecunious student); I’m not sure that it’s a regression in ARM SBCs so much as the emergence of a different market segment that pretty much didn’t exist back when it was Gumstix(and possibly some friends I’m not aware of) and absolutely stupidly priced dev boards exclusively aimed at people buying them with the company’s money in preparation for their own large run of something based on the same SoC(any of the ones you bought directly from the chip vendor, had a list price north of $2k; but at least came with a crippled version of their compiler/dev environment!).
Gotta love Gumstix density; don’t so much have to love how little you can connect to one without a breakout board that has a couple of these involved: I fully admit to being a distinctly tepid solderer(and those are clearly intended to be pasted and reflowed); but ‘easy’ is a relative term.
The rPi-and-imitators are, indeed, compromised in that they don’t necessarily bring out as many interfaces as their SoCs provide(and their lineage with either Broadcom’s slightly oddball set-top-box SoC or various cheap Chinese tablet SoCs means that the onboard peripherals aren’t always the most robust); but for someone of my mediocre skill the fact that what they do bring out is either to a reasonably standard connector for that interface or to a big, fat, 40-pin header is a bit of a boon.
The part that I do find interesting, and which may be indicative of a shift in the ARM SBC market, not sure where I’d find marketshare information, is that the Pi has apparently found enough success among people who are willing to actually design carrier boards and solder nontrivial connectors and such that it has been worth it to them to design embedded versions specifically for that purpose.
It’s those product lines that are stepping more directly into the territory of the Gumstix and similar. Not sure if there was just enough room for everyone as the number of devices expected to have a full computer onboard has exploded, or whether broad popularity and dev boards costing a pittance has really helped them; but if there is a place where it is rPi vs. the traditionalists it’s in their designed-to-be-embedded product lines; not the kids-and-n00bs-friendly boards that have full sized connectors and similar inflexible but friendly features.
I remember reading somewhere that at least one manufacturer was using Pi 3 Compute Modules (which used a DIMM form factor) in their commercial displays.
I find the new Compute Module to be very interesting. It would make a nice core for building a tinkerer’s laptop, and even with the developer’s carrier board there are a lot of interesting possibilities since it breaks out that PCIe interface.
pro tip: use solder paste and clean up mistakes with solder wick. they go down pretty fast, almost like magic once you’ve prepped your board for it. even better is if you have a template for the paste, but it’s not strictly necessary.
what is hard with that layout if is you’re making your own PCBs on with laser printer iron-on and etch method. You will have trouble routing every fine trace on a single sided board. but at least you don’t have to drill (I break drills like every 100-200 holes, it’s expensive for me). For a small project simply don’t route every pin to something, make your PCB very specific for what you need. If you only need some serial console, GPIOs or some I2S pins (audio), then it’s less of a traffic jam trying to route the traces around.
But what to do with all your other Pis?
