Pine64 and Radxa's new Pi CM4-compatible boards | Hacker News Reader

Pine64 and Radxa's new Pi CM4-compatible boards

jeffgeerling.com

106 pointsgeerlingguy82 comments

> The Pi, despite its warts, has two things going for it: a company that devotes a lot of time to testing, documentation, and bug fixes, and an active and broad community [...]

True, but lets not forget another non trivial one: advertising. I'm lucky if I can spot a new board discussed in a few sites, while every time a new PI is launched, the news resonates like someone set their feet on Mars. If manufacturers want to build a community and have more developers/ports/bugfixes etc. they should allocate some resources where it's appropriate. Not easy since the PI is pushed by Broadcom themselves, I know.

Also, I would avoid the term "clones", which gives a bad impression of the products. There's no such thing as a Raspberry PI clone, unless having similarly placed connectors in some models counts as cloning. The hardware is (thankfully) vastly different. Also, Broadcom doesn't sell the PI processors to anyone else ensuring that cloning them is impossible, as is building any product using the same technology, short of hosting it in the form of a compute module, which may not be the ideal/cheaper solution in many contexts. Not a big issue since there are much more powerful platforms out there, but few know about them for the above reason.

Personally, if I see an interesting board, my 1st and 2nd stops to see if and how well they are supported are the Armbian and DietPI boards and forums (other suggestions highly welcome!).

https://www.armbian.com/download/?device_support=Supported

https://dietpi.com/#download

> advertising. I'm lucky if I can spot a new board discussed in a few sites, while every time a new PI is launched, the news resonates like someone set their feet on Mars.

That's not advertising, that's marketing and name recognition.

Yep, but that doesn't change the point made.

Do you know of a board with a gpu, that is a bit better, than what the Pi offers?

Does it have to be ARM based? Most old laptop (you probably already own) can do everything a Pi can, but better at negligible energy cost. If you want to stream, you can remote control an old android for instance. If you want something similar with good graphics, Jetson Nano is good.

Well, size and battery life is important, when you want to use it as the heart of a robot for example.

And yes, a android device (with broken and deactivated screen) is maybe the way to go for small size mobile computing power.

What does "heart of a robot" mean? If it just needs to be remote controlled you can do that with simpler hardware like arduino. Do you need computer vision? Jetson has good support for that. Do you want it to be for edge computing? It just has to relay info to a server.

What is your project? The Jetson is probably the best choice. Saying GPU is very vague, there are GPUs that are good at decoding video for instance that might not be useful.

No concrete project as of now, but different ideas around semi-autonomous robots. And since they require heavy computation sometimes, I would like to offload that to the gpu if possible.

Jetson sounds interesting.

But mobile phones offer many sensors, computing power and a battery all packed together, so I might start with that. Or rather, have a smartphone for computation and remote connection and a arduino for directly controlling the servos. Or a lego mindstorm. Or both.

You’ll want a breakout board to connect it to motors, or see other projects. It’s possible to make a separate mobile platform for the sensors on the phone and have it gather info.

You’re going to want to look for NPUs and probably not GPUs. If you don’t have any idea check this out for inspiration. https://arxiv.org/abs/2004.10934 The DJI founders wanted engineers to be seen as hero’s instead of sports stars. He sponsors engineers to complete here, the prize is their dream job. https://www.robomaster.com/en-US

Well, I have seen and done some projects already, too. And tons of ideas for new ones, but am currently occupied with something else. What I am doing right now is looking what is out there.

So I am quite open about plattforms, but actually I would prefer simplicity. But powerful.

Something that has the potential to run WebGPU in a decent way (once that becomes stable).

You definitely want the Jetson. Great support, decent hardware, CUDA, Linux, etc. I was called a shill for being sad that ARM and Nvidia didn’t merge but I think this is the best quality device, and has tons of support.

Eh. Sometimes there are legitimate technical wins even for things that are otherwise problematic.

The TMo/Sprint merger has had issues, for instance, and I personally have been less favorable to it as time went on, but it didn't mean I wasn't excited at the time to see one more death knell for CDMA.

Also Nvidia has been pretty open-source friendly with their ARM platforms in addition to their being pretty powerful in their own right, and that doesn't have to be mutually exclusive with the concerns it would raise if they controlled the whole ecosystem of ARM platforms.

CDMA has died since LTE. Sprint worked on hybrid CDMA/GSM since they needed SIM cards.

>Also Nvidia has been pretty open-source friendly with their ARM platforms in addition to their being pretty powerful in their own right, and that doesn't have to be mutually exclusive with the concerns it would raise if they controlled the whole ecosystem of ARM platforms.

Yes that is true, surprisingly so since fuck you nvidia. Intel and AMD have x86 rights and GPUs and can make more integrated devices, while nvidia can't. Since it didn't go through, they better be good at RISC-V!

One thing that makes RK3566 notable is that its GPU (Mali G52) is supposed to have decent open source driver support: https://www.collabora.com/news-and-blog/news-and-events/panf...

As a Pinebook Pro user I can confirm that the panfrost drivers make it a vastly more usable platform ever since that team released them. Its not perfect, but its better than nothing.

With the Compute Modules, Raspberry Pi wanted to enter industrial/professional markets.

However, in these markets, availability on short notice is a fundamental requirement. The CMs appear mostly unobtainable, and it's not clear that the Pi foundation will culturally be able to remedy this in the long term ; their strength is elsewhere.

I’d say consistency and long term availability would be the fundamental requirements. The Pi foundation has always seemed to have that part figured out.

I don’t think we should necessary judge availability of any part based on the current environment.

I am not referring to the current environment, from what I remember of the last few years, Pi parts have not been consistently available. Or am I misrepresenting?

Until late 2020, maybe early 2021, I could always find the Pi model I needed from _somewhere_. And usually the local Micro Center had at least partial inventory of all the range of Pi models.

Even midway into 2021, I could find at least a couple of each model at Micro Center, and more than half the SKUs would be in stock at most online retailers.

As 2021 progressed, though, it became impossible to even find the less popular varieties in stock, to the point I haven't seen a Pi 4, CM3+, or CM4 in stock at my local Micro Center since September. I'm glad I picked up a couple spares and placed orders for a few 8GB modules in January—otherwise I'd still be waiting indefinitely.

Pis are only one of many devices that are basically screwed by supply chain issues this year. Eben Upton mentioned that they're producing more than ever this year, but they've sold through and are still going to be fighting supply issues until 2023 [1].

[1] https://www.theregister.com/2021/08/20/raspberry_pi_supply_e...

But, beyond the short term, what do you reckon of the availability over the last 5 years?

Over the last 5 years, availability was often spotty for the quarter or two post-launch, but stock always caught up. In fact, I remember in late 2020, the Micro Center near me had an end cap with over 100 Pi Zero W's hanging on it, a bunch of Pi 3 B+s below, about 30 Pi 4 model B's in a case (you had to ask for them), and about 25 CM3+'s on a shelf.

Those were heady times; the pandemic (demand for hobby computers spiking) then the parts shortage were a 1-2 punch and now it seems only the Pi 400 can be reliably found in stock, along with some expensive kits that bundle a ton of things with the Pi 3 B+ or Pi 4.

Alternatively, most of the production has already been snapped up by large customer(s). NEC is offering them as an option for their commercial displays, for instance.

They do have a history of underestimating demand, though. The original 26-pin Raspberry Pi was unobtanium for a long time after it was introduced. Some things never change. :) The pandemic-related supply chain woes can't help.

Yes, many will remember the crazy waiting lists for the Pi 10 years ago. They did not expect to get a huge amount of demand so early.

At release, we wanted to add support to our product line for the CM4 and needed some to test. However our MoQ was 200. Fast forward to today, we can’t really get CM4s or regular pis fast enough. We recently secured enough for the next 4 months (~800 Pi4 8GB,) but it’s always a battle to find large quantities.

We have a vendor that assembles these for us, so they all come pre-assembled. We’ve swapped parts in our BOM for these at least 13 times in the past year. Change SD cards, change microHDMI adapters, power supply, change case, pi model, etc. Relatively painless, but becomes a problem when 30 days of back order is ~150 raspberry pis.

I still wouldn’t even consider swapping to another device, though.

What's your product line if you don't mind me asking?

Zero-touch deployment kiosks, plug in and go

> However, in these markets, availability on short notice is a fundamental requirement. The CMs appear mostly unobtainable, and it's not clear that the Pi foundation will culturally be able to remedy this in the long term ; their strength is elsewhere.

I think all the other SBC vendors are even worse. E.g. seeedstudio is out of literally every Radxa product right now. 2021 is a tough year for niche products.

Anecdotally, I’ve been ordering a bunch of stuff this year - hardkernel/radxa/friendlyarm/pine64.

Apart from PINE64, I’ve been able to get almost every board I’ve had my eyes on - in due time. They can go in and out of stock in intervals of weeks or months.

This is all ordering directly from vendors. Domestic redistributors seem to be dry all over though. I guess you could say seeeed falls in the middle of those.

Basically if you can afford to wait and subscribe to when things come back in stock (if you’re unlucky), you’ll usually get your stuff in due time.

It takes some discipline to not be part of the problem by ordering extras just in case.

These small Asian vendors are very agile and have swapped out NICs and stuff to make up for missing components.

On the slightly more industrial side it’s ridiculous, though. Due to Intel NIC shortage, PCEngines APU has been out of stock globally for I think 6 months and the manufacturer was transparent from the beginning that it would be a year or two before they are available. I think all of 2022 is already reserved.

> The CMs appear mostly unobtainable

To whom? There might be a world of difference between the experience of buying a CM4 at retail and the experiencing of reaching out to the RPi foundation with a bulk order.

My boss was lamenting how projects would have to be pushed back because they are unavailable

In industrial markets *long-term* availability and robustness are also crucial.

You don't want to develop and optimize your software stack for some hardware that will be out of order in 3 years.

Olimex guarantees long-term availability and industrial standards.

Raspberry Pi does not and is only suitable for the hobby market.

Raspberry Pi does have guarantees, for example if you look at the Compute Module 4 page [1] it reads "Compute Module 4 will remain in production until at least January 2028".

Heck, even the 7-year-old Compute Module 1 will be in production until "at least January 2026". That's a 12-year product support lifecycle!

[1] https://www.raspberrypi.com/products/compute-module-4/

Raspberry Pi drops more SBCs on the floor than Olimex sells to customers.

It'll be awesome now to get a laptop chassis with one (or a couple) connectors for these. If this becomes a long-term standard, we'd be able to upgrade laptops without throwing the old one out (or passing it to someone who needs it).

It seems like the CM4 design was almost made with laptops in mind. The main issue is cooling, though. The case would need to insure that some sort of heat sink would maintain physical contact with the SoC, and some thermal paste would be a good thing as well, so this might be a little more than "drop in a new module." Otherwise, throttling will be an issue with any sustained work. How much of a problem that becomes depends on the intended use.

It wouldn't be an M1 Mac, but an 8 GB CM4 could still be the heart of a nice low-power laptop, with the bonus of GPIO pins, as long as the cooling is sorted out.

Could you just air cool it? I thought pis were supposed to tolerate passive cooling in general, so I would think it would be enough to have a fan to offset the case

Running at 1.5 GHz, air cooling is enough. The Pi 400 has a massive but thin heat sink that keeps it running well at 1.8 GHz (and I overclocked to 2+ GHz with no ill effect or throttling).

If you have ambient temps under 80°F and a light breeze past the bare SoC it doesn't seem to throttle unless overclocked to 2+ GHz.

Laptops don't always get a lot of air. A good thermal connection to something reasonable can go a long way.

The Pinebook pro just uses a thermal pad touching the IHS and the magnesium bottom panel. It works fairly well but will still throttle significantly under a full load for long enough.

I've got a side project where I'm building a tablet based around the CM4, and one big issue is that none of the RPis can suspend to RAM. If I stick with the CM4 (I may switch to one of the two rk3566 boards mentioned in TFA, which hopefully support suspend), I'll probably end up having my tablet suspend to disk.

A replacement board for the Framework laptop would be nifty.

Didn't think of that, but yes - a CM4 carrier with the same connector layout as the Framework would be awesome.

A 100% Apple-free, 100% x86-free, almost 100% Windows-proof laptop would be a great thing.

I've been salivating at the thought of something like this too, but the whole module system being built around USB4 would be kind of a problem.

Really I'm just waiting for host support to trickle out to chipsets other than those by the vendors who wrote the Thunderbolt spec it was built around.

But on a similar note, personally I'd be even more interested in a board that's just a USB4 hub instead of even having its own internals. Given the moddable-on-purpose nature of them, it would enable a whole new range of capabilities if they could also extend other devices, especially if (I hope when?) they branch out to other form factors.

P.S. I'm typing this comment on one right now -- Batch 5 ^.^

> I've been salivating at the thought of something like this too, but the whole module system being built around USB4 would be kind of a problem.

The CM4 provides a pcie bus, so what's the problem?

Whats wrong with laptops now? You can of course 3D print that.

Hopefully these boards have better availability. Normally I would use a Pi Compute Module or just a regular Pi to build a product. But I'm designing things around ESP32s now because I can actually get those.

Isn’t ESP32 generally better for most products? Pi is so heavy, ESP32 S2 is nice too.

Raspberry Pi and ESPs are so different that IMHO you can't meaningfully make such a statement.

IoT I can safely say that running a heavy OS is not the best way to do so. Most IoT are lighter, cheaper, use less power and smaller. Using a computer with an OS for IoT is rarely a good idea from what I seen. He said he used ESP32 instead, so they can do the same thing.

Depends on your product and industry. At low volumes you can get an app running on a Pi quickly. And it already has wifi and bluetooth. Plus lots of available expansion boards. Power is usually not a concern for industrial equipment.

The Pi Zero and Pi Zero W were used in a ton of IoT devices.

Unless you’re comparing it to an arduino, many other boards like ESPs have WiFi or WiFi and BT as well, and aren’t hard to get it running quickly. The zero has no wireless as far as I remember.

What were the zero and pi zero w used in?

I've seen a few product for industrial monitoring use the Pi Zero W.

That’s pretty different from ‘a ton’ and I doubt the zero is used meaningfully in iot. I’m trying to think of a scenario where the pi is better than an embedded system, it seems like overkill or overtly complicated to need Linux to monitor this stuff unless it’s also computing it before sending the data, maybe in thinking about it wrong? IoT and edge computing are what I generally think of, and something like Pi I see as a hub.

What's "an IoT"?

Same as always, it depends on what you are doing. As soon as you have a non-trivial display, data-processing needs, anything involving video or expectations of expanding features later, going to an MCU is a serious compromise and/or additional expense during development at best.

Does it make sense to put a Linux system in a light bulb? very likely not. Does it make sense making your product harder to use and dependent on a Linux server somewhere else just so you can avoid using a system powerful enough to do on-board data processing and get away with an MCU? Also probably not. I see plenty companies learning how to do embedded Linux because their purely microcontroller-based designs don't cut it anymore for what they want to do today.

Usually it’s sensors and it does usually make sense unless it’s also processing the data, IoT is usually simple switches. Aside from graphics such as video streaming off the top of my head, processing the data is not often useful aside from machine vision.

I’m not a big fan of IoT, although Linux based advantages are a good repository of software, familiarity if your a Linux/Unix user, and easy deployment/prototyping.

I'm not sure if I've asked on HN before, but I so wish for something RPI-esque that use usb-c both for power and display output (and actual usb as well).

It would be really cool to have a bus-powered thin client I could just chuck in to the usb-c cables attached to monitors at work. Have it boot really quickly and just open up VNC/RDP/≈Citrix...

But no, the only one I've found is the pretty expensive Intel based Lattepandas, although I see they currently run a Kickstarter for a new one[0]. But sooner or later someone must build a ARM-based one with some usb-c magic chip on them that can concatenate power/video/usb in one port.

[0]: https://www.kickstarter.com/projects/lattepanda3delta/lattep...

You can buy an android phone with usbc vid out to do that.

Thunderbolt is made by Intel and fairly expensive to implement. So no surprises why it shows up on the expensive Intel based board and nothing else.

You don't need Thunderbolt for Display + power + USB over USB-C, just DisplayPort alternate mode. You only need Thunderbolt if you also want PCIe

The PinePhone and PinePhone Pro both do this. They have a cheap “convergence” dongle breaking out USB-A, Ethernet, USB-C PD and HDMI from the USB-C port.

I had such a silly smile on my face the other day when I got Arch up and running on the PPP with Sway and external keyboard. With a bit more tweaking I’ll feel comfortable not bringing the laptop for weekend trips “just in case”.

If you really want another form factor, I don’t find it unreasonable to disassemble it and fit it in a custom case sans touchscreen and battery.

If one is making a board for CM4 module: how does one solder those two connectors to be perfectly aligned? Are manufacturers using some kind of jig for each board going through reflow?

The pins are sufficiently small and tightly-aligned that the connectors seat themselves nicely centered during reflow. Though a few of the boards I've tested do seem to be very slightly off, but you can still pop the Pi into the connectors.

Lots of flux and a way to inspect it. There's tons of cheap USB microscopes out there now that you can use for soldering. If you put a tiny bit of flux underneath the connector, heat it up, and let it sit you'll make it tacky and can lay the connector on it. Tape is also usable. Just get the ends and then remove it to do the full thing.

I wish there was a board that focused on providing a VS Code server to iPads over USB-C, with all the cost allocated to processor and memory, and very basic (or no) display hardware. It feels strange that such a powerful tablet would need such a thing but, if the board was powerful enough and got out of the way, it would be a nice option to overcome the restrictions that prevent iPad from being a more general mobile software development machine.

Jailbreak it, my iPhone is a mobile Unix OS. Relevant CLI has been ported, APT since iPhone 2G. I have neovim for instance. You can use VNC. Here is some Unix tools https://github.com/ProcursusTeam/Procursus

Or consider this: https://en.jingos.com/jingpad-a1/

Encountered this too with Asus Tinkerboards. The ecosystem is so comparatively weak that the (at time) better hardware isn't enough to tip the balance

Any crappy IoT tutorial will work on a rasp...other boards you better have troubleshooting skills

I don't like Pi for IoT, its too heavy, a fucking OS for IoT is heavy, no way am I using that for controlling light bulbs.

Well I’ve got esp too but pi is much easier to troubleshoot and program

What are you running on your ESP? Are you using https://github.com/FreeRTOS or arduino? Low level embedded is harder for sure.

I’m using ESPHome extension for home assistant which has been fantastic because you can flash code on over the air and the initial flash can be done from a browser

(Which still blows my mind)

Yeah the ESP is like what I wanted in the arduino. A WiFi module would have cost so much more for that hardware.

How is it heavy? You can have it turn off / on a light bulb in less that a millisecond. Why would you need it to be faster than that?

I don’t need Linux to turn off a lightning bulb the same way I don’t need a supercomputer to run Microsoft word.

Similarly you could run Word on a computer from the 90s, but it's easier and more continent to use a modern PC.

A Raspberry Pi will absolutely not be faster at turning on a lightbulb versus embedded solutions, running it through the linux stack will make it much slower.

Nobody cares if your light bulb turns on in 1 microsecond vs 1 millisecond (these are example figures). Sure it may be slower, but both fall within the nonfunctional requirements of the under.

I do, having an OS will have more errors so it won't be as reliable. Want to use a Rube Goldberg machine to turn on your lightbulb? Not me.

Breaking out PCIe in ways you can add lots of m.2 SSD is interesting for people who feel throttled behind USB based disk io.

There are 4x carrier cards. Each m.2 gets a lane. You get closer to HBA class saturated device speeds.

ZFS home NAS

Now, make a notebook capable of accepting this form factor and we'll have something more upgradeable than the framework notebook.

https://www.pine64.org/wp-content/uploads/2021/12/MNT-Reform...

Here's an adapter to use a SOQuartz in the MNT Reform laptop, mentioned in today's Pine64 blog post.

ARM SBC's will always lag Intel in the single most important factor - software.

Unless some large corporation(s) get serious about spending on money on first class Linux implementations - and there's no sign of that.

The best ARM software is the community developed ARMbian, but that's not enough to tempt me to buy anything except Raspberry Pi.

Raspberry Pi's greatest strength is its first class software.

The weird thing is that none of the ARM device vendors seem to realise that "it's all about the software".

> The weird thing is that none of the ARM device vendors seem to realise that "it's all about the software".

Pine64 basically exists to ship (mostly ARM) hardware with good open source software support.

Reply on news.ycombinator.com