Hacker News Reader

Comment by micw | Hacker News Reader

Does anyone know if it's a good idea to base a product on Pi CM4/5 (or has experience with this)? I wonder how the availability any price in large quantities is (like 10k per year) and how reliable the connector is in comparison to a soldered SOM.

View on HN

In general, the pi SoM usually offer 8 to 12 year availability maps, and haven't pulled price ballooning after launch.

If you are producing less than 1k pcs a month, than it certainly makes sense for smaller design runs. That being said, it is probably safer to eat the cost on the standard 0.1" ribbon cable header form-factor if you plan to run the line for a few years. The compute-modules have a tendency to change, and the micro-contact headers could be a failure point in some situations. Also, hardening the Pi design to be more reliable takes extensive testing, and experience.

Compared to other SoM manufacturers the Raspberry Pi foundation has a good reputation in both the open source community, and commercial roles. The pi4/5 FCC modular pre-compliance also saves around 11k USD when you go for lab testing. Also, pushing pi SoM production volumes higher means lower unit costs for everyone, and a double win is always nice.

If you don't need the gpio, than a mini PC form factor may offer more value.

There is also the Kria KR260 kits around, but it will not offer anywhere near the software/kernel ecosystem support of the pi community.

Best of luck, =3

As someone who spent many months designing a product around CM4 and then waiting for more than 2 years for modules to arrive over the parts shortage era (eta kept being pushed out by the distributor) I will never put myself in that position ever again. The solution I have found was to skill up and learn how to do hardware design myself. At that point there are many more options. Understandably that's not ideal for a lot of people.

Unfortunately, chips EOL all the time, charge pin-outs, and the temptation to in-house an integrated SoM solution may add hundreds of thousands to the cost. If you are not moving >3m chips a quarter, than you are still vulnerable.

We also had to violate the Design for Manufacturability guidelines to adapt to the shortages and part skelpers hitting JIT lines. Even today, we incur a questionable 12 USD labor cost on every product to ensure a generic carrier PCB drop-in population option is always available (0.1" pins 1980's style).

Training slide deck: "Rule #21: No unicorn parts, and no excuses"

We dodged the CM4 choice luckily due to my concerns, but still were tagged by a proprietary missing RF module needed for legacy system interoperability. The vendors lied about inventory levels, and kept the order tied up for years before the spools arrived.

Best of luck, =)

I see your point, and in most cases I would agree with you; this was an unusual time; you would have had the same problem using STM32s, etc.

Can you highlight the resources that you used to level up this way? I can design my own microcontroller boards, but the complexity of SoCs and required peripherals seems too much for this part-timer.

Admittedly I ended up building a board with a microcontroller rather than a microprocessor because I could interface with the modem over spi too. However, this YouTube channel was great to work through:

http://www.youtube.com/@PhilsLab

It has some high speed stuff with ddr memory, fpga’s gigabit Ethernet and usb3 too so I’m pretty sure you could use the same concepts to build a board with a microprocessor on it. Proper grounding, emi, impedance control and length matching are well explained. I think he has some more advanced courses via his website too.

Additionally, this was an ah-ha moment video for me: https://www.youtube.com/live/ySuUZEjARPY?feature=shared

While we have never used the compute module in large quantities, we have used it numerous times for small batches with 1-2k/month and had no issue there

If you plan on using it for small series of <10k/a I wouldn't worry too much

Even if it were reliably available in such quantities, I'm not sure Pi's pricing would make sense in the context of any mass produced product.

There are apparently products based on the CM (Ableton Move?), not sure how they sell these at a profit.

Design for Manufacturability paradigms sometimes generate odd choices that only make sense in certain use-cases. When scaling up to over 50k pcs/month the incremental costs do get ugly fast.

However, for many of the same reasons a 3k USD PLC makes sense in many low volume long-tale market applications (i.e. pre-certified and stamped hardware.)

Best of luck, =3

The Formlabs Form 4 uses a CM AFAIK and at a 4-5k price point it's not hard to see how they sell them at a profit.

They are available in that quantity. They do make sense up to around that number, although the around 1K is the sweet spot. Laying out and certifying a high speed design (not just a noddy A9 with DDR2) is expensive and it's all NRE cost. The cost of the components for the CM is driven down by the combined volume significantly - if you want to buy 1-10K of those by yourself, you will pay a premium over what RasPi gets them for. You might not need a full 6/8 layer board for the whole product as well. As important, the software support is great as it's just a standard RasPi - you don't have to support your own custom image.

Volume pricing on custom commercial Pi SoM are not common, but do happen at the Pi foundation.

I have seen 3 or 4 products with these pi+Debian SoMs running network services etc. Getting them to be reliable is a different set of issues with dozens of edge cases, but its the same with every other vendor. Best regards =3

Mass producers do not buy at retail prices.

Sometimes we do, but it hits both us and the end customer.

Recall the chip shortages... =3

They are far from the most expensive SOMs on the market.

i have. avoid pis in retail products period unless you work at broadcom or want to sell a subpar device

Probably a better idea to go for the ROCK-chip TFA is using in the benchmarks or something similar.

Reply on news.ycombinator.com