DFI GHF51: AMD Ryzen R1000 on a Raspberry Pi Sized Board
dfi.com
dfi.com
However, no NVMe, no regular PCIe, no even a single SATA port...Lack of IO is a serious dealbreaker to me. Feels like it's wasting the CPU's capabilities. I know there must be compromises for the RPi size, but the size itself is already compromised because it requires a cooling system, at least a big heatsink, on the back of the board. This company is hiding it.
I would go for an Odroid H2 for now.
USB-C ports are flimsy and I've had many just snap right off the board, especially given the massively long cantilevers that most USB-C cables are. They also massively increase the effective footprint of the board when plugged in.
Or at the least, have the USB-C connector but provide something else as an option. I know the Pi can be powered off the GPIO, but it's not ideal as it bypasses all the power filtering circuitry.
The pi form factor is fine, but 2-pin Molex-type connector in parallel with the incoming 5V would be a great addition. Hell, even just leave pads or holes and I'll solder in a connector myself. RPi's are being used more and more in embedded applications. It would be nice if the Foundation acknowledged that and made it a bit easier to work with.
However, I disagree that the Molex connector is inherently superior to USB-C. I am abundantly familiar with the incongruity of a wall wart in an embedded application, but all you need for the power supply is 5V delivered over a USB-C connector. I understand an appeal for a latching or screw-secured connector, but a pin header connector is neither of those. USB-C doesn't easily fall out, it's just another connector.
Another answer is to hack a USB 2.0 C-C cable [1] in half, pull back the shield, heat shrink over the shield and 30 AWG data wires, crimp ferrules on the 26 AWG power wires, and run those to my screw terminals of the existing embedded 5V power bus.
Molex has very sturdy through-hole power connectors as well as plugs that are lower profile and won't torque off the connector as easily. 2-pin Mini-Fit Jr. would work pretty well.
Or XT30 even. I love XT30 connectors.
https://www.amazon.com/JSER-Type-C-Locking-Connector-Standar...
Issue solved.
I would strongly hesitate to recommend the Pi for anything situation where 99% reliability (or above) was required unless I understood the domain back to front and could cover all the failure cases such as the expected operating environment ranges (temperature, power quality, EMI etc...).
"Industrial" spans a wide range. The problem I run into is that when I mention using Arduino or pi in industrial applications, people immediately jump to thoughts of extremely harsh environments and safety-critical applications. There are plenty of existing solutions to those problems.
Where the Pi shines is in "light industrial" tasks that a standard PC would be unsuited for (cost, size, I/O, power consumption, etc), but the environmental requirements aren't stressful. Those can be anything from counting rebar loaded onto a rack or providing help to truckers at a loading dock.
Even so, I have lots of small devices (arduino clones, Raspberry Pi) in machine control applications and they've been working very well.
Moreover, Raspberry Pis were never intended to be used in industry, and obviously are not designed as such. They _happen_ to work in many use cases, yes, but they lack so much: reliability engineering, failure testing, ruggedness requirements, industry certifications, QA, I could go on.
And, do you really want to be the guy who gets the call because failure of a $35 part stopped the production line at 2 a.m.?
There is no lack of industrial computing gear on the market, even stuff that runs Linux. It will cost an order of magnitude more for the same computing power, but it is more rugged, integrates well with existing systems, and is sold by a vendor that you can actually get on the phone and hold accountable for some things. And most importantly, putting it into production probably won't get you fired.
Sometimes you don't need all that stuff. Sometimes it's as simple as using a neural net to count the number of people in a room at a given time, or get a sensor to publish data online, or provide a networking shim for a RS232 device that fits in a small space.
There are also many use cases where having 10 working backup copies of a $35 device on a shelf that can be swapped in under a minute is a much better reliability guarantee than 1 $2000 industrial certified, QA certified, FDA approved, UL listed, GDPR compliant device whose tech support is in some other timezone, takes holidays, and isn't 24/7.
It's just that the easiest thing that can fail on a raspberry pi in my experience is the stupid micro-B/USB-C connector, so fixing that, which would cost pennies to put a slightly more rugged molex or other connector on the PCB, would be a really nice addition. The rest of the Pi is actually pretty reliable in my experience.
> stopped the production line at 2 a.m.?
Production lines aren't the only use case for Raspberry Pis in industry. Many production lines also don't run at 2am.
https://www.udoo.org/udoo-bolt/
Alternately, you could buy the Asrock 5x5 barebones kit.
https://www.tomshardware.com/news/asrock-am4-motherboard-ryz...
So if somebody needs some odd port setup, it's then carried on to boards sold on mass market.
Taiwanese used that "dual track" trick for a long time. Have B2B clients pay for RnD, and tooling, and then make money for free on it in OEM side of your business.
Also providing a USB header on a high-speed expansion header is standard in the 96 Boards spec I think. https://www.96boards.org/specifications/
I've done the hack version and just super glued the start of a cable coming from a keyboard to a case so that the actual connection wasn't stressed.
It might be better if they made it easier but it is a problem that can be solved.
If the target is full support with standard distros, then one needs to go for x86 SBCs. As far as I know, the only (relatively) cheap one is the ODROID H2.
Once the 5.5 kernel gets wider support I suspect this won't be an issue anymore. Ubuntu already supports ARM and with kernel support on the way it's only a matter of time.
>in RPis, the GPU is actually the main processor.
huh?
While that might be a fair view, I think it deserves further qualification as to not just cause extra confusion.
Can you clarify what you mean by this statement? It doesn't really fit well with my understanding of the RPI design, other than the fact that it's a SoC and the GPU and CPU are on the same chip.
Defining "main processor" is certainly arguable, as there isn't a unique way of viewing the architecture, but on the other hand, the idea that ARM boards are "very custom" still stands.
Some informations:
- https://raspberrypi.stackexchange.com/questions/10489/how-do...
- https://raspberrypi.stackexchange.com/questions/14862/why-do...
this block diagram makes no sense
USB 3.1 Gen 2 output made up from ... a Gen 1 and an USB 2.0 link? That's not possible you need to feed Gen 2 signals into the socket to get Gen 2 signals. While the USB standard is a mess, this is crystal clear: Gen 1 is 5gpbs and Gen 2 is 10gbps.
http://linuxgizmos.com/raspberry-pi-sized-sbc-is-smallest-ry...
Anyone know?
Only thing I can imagine is the key variants for M.2 are more complicated.
M.2 limit to only storage.
Edit: Link: https://www.scan.co.uk/products/intel-8265ng-dual-band-m2-22...
Sigh. Hopefully somebody else will do this properly.