Pretty wild to see a drop-in replacement that's actually faster - usually that's marketing speak for "technically compatible but slower in real life." The fact that you can just pop this into existing CM4 hardware and get 2-3x performance is impressive. Finally, my commercial display can run Crysis! (Ok, maybe not, but that USB 3 upgrade is sweet).
Though I have to laugh at the "good news everyone, it's the same price!" followed immediately by "...for the 8GB version only, everything else costs more." Classic Pi Foundation pricing gymnastics. At least we're not dealing with scalpers asking $200 for a CM4 anymore.
The silkscreen specs on top is such a simple but brilliant addition. No more squinting at tiny chips trying to figure out which module is which in your parts drawer.
Thing is once you start looking outside of the base model it becomes a bit less of a good offering when there are others making compatible boards that are arguably better specced.
The only issue of course is some whilst sharing the same formfactor and connector aren't always compatible with the same hardware.
If I recall correctly the Radxa CM3's for example are fully compatible.
My OrangePi 5B runs circles around an Raspberry Pi 5 (cooler, 2x cores, 2x RAM f you want, a decent eMMC, AI acceleration plus better hardware video encoders, etc.).
However, the support is "barely there". It's running standard Debian stable with a pinned Kernel from OrangePi guys, and the Kernel tree is open, but I don't know whether I'll be able to compile a 6.x kernel by lifting their patches and correctly applying them to a mainline kernel.
Then I don't know which drivers are closed source and how they'll play with each other.
On the other hand, Raspian works really well. Plus with a good A2 card, you really don't feel any latency in day to day use of the system anymore.
P.S.: I run both of them as home servers at different locations, so I can "continuously review" them day to day.
Rpis proposition has always been support rather than value per se.
If you want cheaper, yes there are other options. If you want something where a mainline kernel will work, and has a community, Pis are a much better choice.
For some definition of a "mainline kernel." My preferred Linux distro is Debian, I still can't run a Pi 5 on Debian over a year following release. (I don't know if this is for lack of Pi engineers not pushing stuff upstream or Debian/kernel devs not being receptive to PRs, but I suppose the answer is somewhere in between.)
I do run Debian on Pi 4B/CM4 and others. At present I'm working on getting 1-wire with a DS18B20 working on a Pi Zero W and it has not been easy, but that might be me.
I was truly excited to see the CM4 released because it provided a PCIe slot and which can support a proper storage interface on a Pi. I don't see a compelling feature on the CM5 so I'm not going to jump on one just yet. I'll probably get one at some point.
I do agree that the Raspberry Pi community provides reasonable support. I have, however, found the Reddit community occasionally toxic and have had questions on the official Pi forum removed w/out explanation. (I asked why the Imager did not offer the most recent OS releases in the menu, requiring that I manually download them.)
The last time i tried to get pi running with a distro that wasn't raspbian it was huge PITA, with having to use custom kernel patches and weird proprietary firmware and annoying boot process, am I misremebering or did this change ?
On any x86 UEFI system I tried i never had any issues at all, but Pi's I remember as a huge PITA...
I understand that this is the case. IBM did the PC world a solid by not encumbering the original H/W. That provided a basis for a platform that could be copied and provide a "standard" that others could leverage, including copying and reverse engineering the BIOS that launched the OS. That provided the foundation of the PC explosion. They tried to reverse their "mistake" with the Microchannel Architecture and that saw limited success (AKA failed.)
IBM was also part of a consortium that tried to develop a common platform for the Power architecture. I'm not sure if any of that survives in their current offerings but it did not take anything away from the Intel (Windows) market and except for Apple, the Power PC was nowhere to be seen.
ARM processors have proliferated in the hand held (phone, tablet) market where volumes are sufficient for the manufacturers to provide their own solution to bootstrapping the OS. There are some efforts to agree on a common boot architecture (Yocto, UEFI and probably others) and I'm not familiar with them, but I hope one succeeds so we can easily load a variety of OSs on our devices.
Incidentally... The GPU in the Raspberry Pi performs the initial stages of the boot process. Weird. I think that Pi 4/5 generations are moving away from that but I'm not familiar with the detains. As near as I can tell, they've preserved the S/W structure required for the GPU to perform this function.
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.
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.
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:
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.
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.
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
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.
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.)
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
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.
Does anyone have any recommendations for a CM carrier board that has ethernet and ~4 sata ports that I can attach external drives to? I'd quite like to build my own NAS but I don't know what board to use
It's purpose-built for that, and they even have a rackmount chassis for it, though you could design your own too. Power for the drives requires a separate power supply though.
The other option right now is the Radxa Taco (https://radxa.com/products/io-board/taco/), but availability has been scarce. I've ordered one and will hopefully test it again soon, with the CM5!
(That is, if you're wanting to go down the Pi NAS route—there are certainly other options out there!)
Those both look good, though I might have to save up a bit to afford them haha. I'd probably end up printing my own case, so that's not really an issue. Looking forward to your testing of the CM5 Taco =)
I have the Taco and was about to ask whether you have tested it with the CM5 yet.
For some reason I thought that Radxa is about to deprecate the Taco but I might be mistaken.
The one issue I found that without a heat-sink, fan and any drives just idling on a 12V supply the board got really hot. So I have my questions about the efficiency of the board. I've been thinking about testing the power usage of it so maybe a good time to follow through on it.
In all honesty, and considering that everything you build with a SBC is not that powerful and a cabling mess, I would recommend to go for a refurbished ThinkCenter Tiny or similar + an external HDD Rack.
Atleast thats the route I went, and I am vastly less annoyed about it.
I got a small thinkcenter a while back that I currently run some docker stuff on (homeassistant, pihole, some custom stuff), and I am pretty happy with it. Do they have a spare pcie or something to attach all the drives to, though? I don't remember seeing anything like that when I was inside mine, though that was a while back, so my memory may be wrong
Some of the models do - I have the M720q and it has a small form factor PCIe 8x slot on the motherboard, but you need a proprietary riser board (cheap on aliexpress) to convert it to a real PCIe slot, and it takes the place of the internal SATA drive cage and rear port adapters (VGA or serial or what have you) so you can't use those.
I'm using mine as an OpenWRT router with a dual 10 Gbe ethernet card in the PCIe slot ever since I got 10 Gbit internet at home and consumer 10 Gbit routers are expensive and leave a bit to be desired
In theory you can use the WiFi-One, though I don't know if that one is BIOS-locked. I just use it with USB3 and a 4 Bay Enclosure, which probably someone is gonna explain to me is the worst idea since leaded gasoline, but I have absolutely no Issues with my ZFS NAS for 5 years that way.
Over on reddit I once floated a similar idea but everyone exclaimed that USB enclosures sucks and you should not use them at all. Now technically you could - with some PCIe expansion cards - connect SAS/SATA HDDs to the ports on your mini/SFF PC, but then you need an enclosure for holding, powering and cooling the disks. Holding and cooling is simple enough and there's enclosures on AliExpress that cost like 10 bucks that do it for you, but you'd still need a PSU outside of the PC to power the disks. At that point the USB enclosure starts looking even more enticing.
Which one are you using right now? There's some brands out there that are considered really bad by the community.
Ah, good point, I hadn't considered a USB solution. I might look into that, though I'll have to see if there's a good way to keep all the cables looking tidy and not have a cable coming out the front and then disappearing back inside again haha
Yeah, I've been considering just going for an x86 thing, though finding a mini pc with 4 satas might be tough, and there's the whole raspi ecosystem that would help with troubleshooting. I'll add that to the potential list though =)
The easiest way might be getting an ITX board with N100/150 from CWWK or Topton. They usually come with 4-6 SATA ports and the CPU soldered on. Then you just buy your favorite case and plug in your drives. Excluding the cost of the drives you might end up spending about $250, depending on your case and PSU selection. The problem with using SBCs or mini PCs that way is that you need an enclosure to hold, power, and cool down your disks, and even that is assuming that you can actually connect them via a SATA/SAS connection which may not be easily available from within your case.
Check out Hardware Haven on Youtube, he tests a bunch of different small form factor NAS computers and similar.
I find the whole "raspi ecosystem" a bit overrated, for the most part it's Linux you're working with regardless of hardware.
It isn't strictly a dropin but it is usually a great upgrade.
If you have the power then video encoding is faster and higher quality on the RPi5.
If you are battery constrained then it can certainly be worse.
The Pi 4 encoder was in hardware but that doesn't imply it was ever actually particularly fast. I'd bet the Pi 5 CPU+RAM improvements are enough to beat the hardware encoder of the Pi 4 and still have enough CPU leftover to beat the Pi 4 CPU unloaded.
I'm guessing that's with whatever stock software/firmware stack Raspberry PI themselves ship, surely can be made a lot faster by not having to probe various hardware ports you know aren't used, for example.
I didn't have it complete in time, but you can use pi-gen-micro from Raspberry Pi (it's a newer project) to build a very lightweight boot image now, on the order of a yocto/buildroot image, but with Pi OS as the base.
Sounds pretty sweet. The Raspberry is now firmly inside a price territory where it has a lot of Intel/AMD rivals.
The Raspberry Pi Zero for $5 was never widely available. I'd love to see something in that price range that is actually available to buy. Hopefully also with low power requirements like the original Pi Zero.
The Korg Kronos (and I suppose the Nautilus as well) are just regular x86 PCs (I believe using Atom CPUs) running a custom Linux kernel and custom software
As far as I can tell, there's no custom "firmware" to speak of, it's a patched Linux kernel running a kernelmode audio synthesis engine and a usermode app for the UI
This CM5 is at least as powerful as laptop CPU from ten years back I'm really surprised nobody one is coming up with upgradable laptop platform or shell powered by RPi compute module. Couple with the available maximum 16 GB with ECC available with CM5 this can be a winner since laptop RAM capacity has been stagnant for around ten years now [1].
There are Pi-based laptops (Pi-Top, CrowPi), but it's very hard to compete with used laptops. And not many people like ARM-based OSes (which is usually Linux or the awful version of Windows).
The Raspberry Pi is one of the only random gadget purchases that I've made over the years that found their way to the back of the drawer, then found their way out, back to the drawer, and out again. It's a really useful general purpose tool.
It's currently running at my parents' place as a VPN server to let them access their smart home and cameras from outside the house (I didn't let them expose these to the internet).
I got a raspberry 4 recently and was tinkering with it. I was appalled at the apparent disk IO performance of the thing, opening a browser takes several seconds.
Is that just me or do I have a particularly crappy SD card? Anyone got any tips on how to improve disk IO? Would a USB3 external drive help with this?
Emphatically yes. It's an entirely different experience using a RPi 4 with a quality SSD. [0] It's also more expensive, so you have to decide whether that's important to you.
In my opinion it's a mistake to split the difference with a cheap SSD or one you need to plug in using a SATA to USB adapter. (troubleshooting first step: if it's not the power supply, it's the SATA to USB adapter...)
USB3 could plausibly be faster than the SD card as the spec allows for 5Gbps+ but I don't know about Pis that much, I'd defer to someone with experience.
As a photographer I can tell you that SD cards, even high class ones have laughably awful speeds - any proper camera that needs to do video will use CFexpress Type B cards which are tremendously faster.
I'm always astounded how little people know about this, but you need to get an Application Class rated card, preferably A2. [1]
The application rating essentially ensures a certain minimum IOPS performance, which is critical because most cards are only intended for sequential reads/writes.
The Pi 4's IO is not great, but the choice of SD card affects it a lot. Brand seems to have more of an impact than speed (or at least the stated speed of the card). Cheap knockoffs will absolutely kill performance. That's why Raspberry Pi started selling their own branded SD cards.
Using an external USB3 drive also helps but it's still pretty slow. I found it too slow to use as a SAMBA file server when connected over WiFi, even though it was hardly pushing the CPU or memory. It would regularly fail to transfer large files.
One of the main selling points of the Pi 5 is faster IO. I'd love to see something tiny and power-sipping like the Zero but with the enhanced IO of the Pi 5.
No it's not just you. I/O on the Pi4 and below is awful, but for most applications it doesn't matter. Not sure on the Pi 5 as I jumped ship after the scalping and overall negativity around the foundation's activities in the last few years but I expect it's a fair bit better.
In general, sdcards are slow... most use kernel cache lazy page flush if the pi4 has 8GB ram, and F2FS to improve flash reliability.
After, "sudo apt-get install preload" it will cause previously used programs from the last session to auto-cache into ram. Combined with the above setting, the usual programs will launch much faster.
There is also the option to use an external SSD drive for better performance.
Part of the fun of pi4 was getting the OS optimized, but unless you have a legitimate "Sandisk Extreme Pro U3" or >30MB/sec rated sdcard it can still be slow (people have seen 4MB/sec on old counterfeit cards.)
sdcard bus overclocking is a thing as the pi tries to maximize slow hardware compatibility by default. Note this feature is limited on the pi4, but can significantly decrease boot times on older models.
Has anyone who sees this attempted to run Home Assistant OS in a Home Assistant Yellow "compute module board" using a Radxa CM5?
It's tempting to me because the Radxa CM5 uses a mainline kernel instead of the Broadcom-modified kernel the RPi CM4 uses, but I understand that there may be some level of work needed to be done to support the compute module + board combo, and I'm curious if that is true for the Radxa CM5 + Yellow board combo.
There is no "Broadcom-modified kernel". There's a fork with changes needed to support all Pi hardware at https://github.com/raspberrypi/linux. Unlike some boards were you're eternally stuck at old versions, the fork is actively maintained (currently mostly at 6.6, soon 6.12) with its changes regularly getting upstreamed. You post make it sound like the Pi kernel is a mess, which it is absolutely not.
I get where you're coming from, didn't mean to make it sound like that but I do see it the same way zokier seems to. End game for me is booting a vanilla kernel compiled entirely from open source code and having access to all the boards hardware.
I was tempted to do something similar when I got my Yellow because the low availability of the CM4s back then, but it seemed that, at least with older Radxas there were problems with kernels or compatibility. It might work with the newer one if you say it's mainline kernel but haven't found anything in the forum in quick search...
I am keen to try. But I'm even more keen for someone else to tell me they have tried it and the result of said test because I am not keen on my home automation system being down for a longer period of time...
I'm really torn as to whether 'smaller than a credit card' (when something is clearly much, much larger than a credit card), or re-purposing a previously well understood and generally agreed upon definition of 'artificial intelligence' to now mean a frequently wrong or bewildered statistical / language model, annoys me the most.
You're forgetting the third dimension, which comes in at 4.7mm. That indeed makes the CM5 much larger at multiple times the volume - if you want to be pedantic that is.
Looks like the Radxa CM5 is still significantly faster than Pi CM5, while only being slightly more expensive (113USD vs 95USD for the 8GB/64GB model from what I could find). Anyone here has any experience with it?
Technically the BCM2712 doesn't have any hardware encode support, only software/CPU.
It includes an H.265 (up to 4K 60 fps) hardware decode block—I tested that, and it works fine—but no other video hardware encode/decode, unfortunately.
The CPU is good enough for up to H.264 1080p encode at least, and with some tweaking could get a 4K stream or multiple HD streams encoded... but it's not ideal for that use case.
Pretty wild to see a drop-in replacement that's actually faster - usually that's marketing speak for "technically compatible but slower in real life." The fact that you can just pop this into existing CM4 hardware and get 2-3x performance is impressive. Finally, my commercial display can run Crysis! (Ok, maybe not, but that USB 3 upgrade is sweet).
Though I have to laugh at the "good news everyone, it's the same price!" followed immediately by "...for the 8GB version only, everything else costs more." Classic Pi Foundation pricing gymnastics. At least we're not dealing with scalpers asking $200 for a CM4 anymore.
The silkscreen specs on top is such a simple but brilliant addition. No more squinting at tiny chips trying to figure out which module is which in your parts drawer.
> Finally, my commercial display can run Crysis! (Ok, maybe not, but that USB 3 upgrade is sweet).
It actually can! .... if you also plug in a GPU. https://www.jeffgeerling.com/blog/2024/amd-radeon-pro-w7700-...
>Classic Pi Foundation pricing gymnastics
To be fair, theyve dropped the 1GB model, so do you want to compare old and new lowest specs?
I don't know about relative sales figures, I infer that 8GB is expected to be the default going forward.
Thing is once you start looking outside of the base model it becomes a bit less of a good offering when there are others making compatible boards that are arguably better specced.
The only issue of course is some whilst sharing the same formfactor and connector aren't always compatible with the same hardware.
If I recall correctly the Radxa CM3's for example are fully compatible.
My OrangePi 5B runs circles around an Raspberry Pi 5 (cooler, 2x cores, 2x RAM f you want, a decent eMMC, AI acceleration plus better hardware video encoders, etc.).
However, the support is "barely there". It's running standard Debian stable with a pinned Kernel from OrangePi guys, and the Kernel tree is open, but I don't know whether I'll be able to compile a 6.x kernel by lifting their patches and correctly applying them to a mainline kernel.
Then I don't know which drivers are closed source and how they'll play with each other.
On the other hand, Raspian works really well. Plus with a good A2 card, you really don't feel any latency in day to day use of the system anymore.
P.S.: I run both of them as home servers at different locations, so I can "continuously review" them day to day.
Rpis proposition has always been support rather than value per se.
If you want cheaper, yes there are other options. If you want something where a mainline kernel will work, and has a community, Pis are a much better choice.
> where a mainline kernel will work
For some definition of a "mainline kernel." My preferred Linux distro is Debian, I still can't run a Pi 5 on Debian over a year following release. (I don't know if this is for lack of Pi engineers not pushing stuff upstream or Debian/kernel devs not being receptive to PRs, but I suppose the answer is somewhere in between.)
I do run Debian on Pi 4B/CM4 and others. At present I'm working on getting 1-wire with a DS18B20 working on a Pi Zero W and it has not been easy, but that might be me.
I was truly excited to see the CM4 released because it provided a PCIe slot and which can support a proper storage interface on a Pi. I don't see a compelling feature on the CM5 so I'm not going to jump on one just yet. I'll probably get one at some point.
I do agree that the Raspberry Pi community provides reasonable support. I have, however, found the Reddit community occasionally toxic and have had questions on the official Pi forum removed w/out explanation. (I asked why the Imager did not offer the most recent OS releases in the menu, requiring that I manually download them.)
> For some definition of a "mainline kernel."
...but I bet you can port Raspberry's patches to a vanilla kernel tree relatively easily. Good luck with it with other providers.
I used OrangePi and Radxa. Radxa rock was a huge disappointment in that regard.
The last time i tried to get pi running with a distro that wasn't raspbian it was huge PITA, with having to use custom kernel patches and weird proprietary firmware and annoying boot process, am I misremebering or did this change ?
On any x86 UEFI system I tried i never had any issues at all, but Pi's I remember as a huge PITA...
Or is this just a general ARM issue ?
> Or is this just a general ARM issue ?
I understand that this is the case. IBM did the PC world a solid by not encumbering the original H/W. That provided a basis for a platform that could be copied and provide a "standard" that others could leverage, including copying and reverse engineering the BIOS that launched the OS. That provided the foundation of the PC explosion. They tried to reverse their "mistake" with the Microchannel Architecture and that saw limited success (AKA failed.)
IBM was also part of a consortium that tried to develop a common platform for the Power architecture. I'm not sure if any of that survives in their current offerings but it did not take anything away from the Intel (Windows) market and except for Apple, the Power PC was nowhere to be seen.
ARM processors have proliferated in the hand held (phone, tablet) market where volumes are sufficient for the manufacturers to provide their own solution to bootstrapping the OS. There are some efforts to agree on a common boot architecture (Yocto, UEFI and probably others) and I'm not familiar with them, but I hope one succeeds so we can easily load a variety of OSs on our devices.
Incidentally... The GPU in the Raspberry Pi performs the initial stages of the boot process. Weird. I think that Pi 4/5 generations are moving away from that but I'm not familiar with the detains. As near as I can tell, they've preserved the S/W structure required for the GPU to perform this function.
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.
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.
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
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, =)
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.
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.
They are far from the most expensive SOMs on the market.
There are apparently products based on the CM (Ableton Move?), not sure how they sell these at a profit.
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.
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
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
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
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
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.
Does anyone have any recommendations for a CM carrier board that has ethernet and ~4 sata ports that I can attach external drives to? I'd quite like to build my own NAS but I don't know what board to use
The best one for that (IMHO, though I should note this company sent me one to test, and it's been running my NVR for months now) is the Axzez Interceptor: https://www.axzez.com/product-page/interceptor-carrier-board
It's purpose-built for that, and they even have a rackmount chassis for it, though you could design your own too. Power for the drives requires a separate power supply though.
The other option right now is the Radxa Taco (https://radxa.com/products/io-board/taco/), but availability has been scarce. I've ordered one and will hopefully test it again soon, with the CM5!
(That is, if you're wanting to go down the Pi NAS route—there are certainly other options out there!)
I have one of these (the axzez board) with two hard drives in a cheap supermicro 1U case, works fine
Those both look good, though I might have to save up a bit to afford them haha. I'd probably end up printing my own case, so that's not really an issue. Looking forward to your testing of the CM5 Taco =)
I have the Taco and was about to ask whether you have tested it with the CM5 yet.
For some reason I thought that Radxa is about to deprecate the Taco but I might be mistaken.
The one issue I found that without a heat-sink, fan and any drives just idling on a 12V supply the board got really hot. So I have my questions about the efficiency of the board. I've been thinking about testing the power usage of it so maybe a good time to follow through on it.
In all honesty, and considering that everything you build with a SBC is not that powerful and a cabling mess, I would recommend to go for a refurbished ThinkCenter Tiny or similar + an external HDD Rack. Atleast thats the route I went, and I am vastly less annoyed about it.
I got a small thinkcenter a while back that I currently run some docker stuff on (homeassistant, pihole, some custom stuff), and I am pretty happy with it. Do they have a spare pcie or something to attach all the drives to, though? I don't remember seeing anything like that when I was inside mine, though that was a while back, so my memory may be wrong
Some of the models do - I have the M720q and it has a small form factor PCIe 8x slot on the motherboard, but you need a proprietary riser board (cheap on aliexpress) to convert it to a real PCIe slot, and it takes the place of the internal SATA drive cage and rear port adapters (VGA or serial or what have you) so you can't use those.
I'm using mine as an OpenWRT router with a dual 10 Gbe ethernet card in the PCIe slot ever since I got 10 Gbit internet at home and consumer 10 Gbit routers are expensive and leave a bit to be desired
In theory you can use the WiFi-One, though I don't know if that one is BIOS-locked. I just use it with USB3 and a 4 Bay Enclosure, which probably someone is gonna explain to me is the worst idea since leaded gasoline, but I have absolutely no Issues with my ZFS NAS for 5 years that way.
Over on reddit I once floated a similar idea but everyone exclaimed that USB enclosures sucks and you should not use them at all. Now technically you could - with some PCIe expansion cards - connect SAS/SATA HDDs to the ports on your mini/SFF PC, but then you need an enclosure for holding, powering and cooling the disks. Holding and cooling is simple enough and there's enclosures on AliExpress that cost like 10 bucks that do it for you, but you'd still need a PSU outside of the PC to power the disks. At that point the USB enclosure starts looking even more enticing.
Which one are you using right now? There's some brands out there that are considered really bad by the community.
A 4-Bay one from Amazon, Icy-Box. Pretty Cheap. I decided on external Backup over Enterprise-Hardware. YMMV though.
https://www.amazon.de/ICY-BOX-Externes-Festplatten-Aluminium...
Ah, good point, I hadn't considered a USB solution. I might look into that, though I'll have to see if there's a good way to keep all the cables looking tidy and not have a cable coming out the front and then disappearing back inside again haha
I cut some kind of DIY-Rack on my Laser Cutter from Plywood, because I couldn't get it stored better in any way.
https://i.ibb.co/XWKDmwS/IMG-20241127-134555-527.jpg
Which enclosure? I've got an upgrade on the horizon and like this solution.
A 4-Bay one from Amazon, Icy-Box. Pretty Cheap. I decided on external Backup over Enterprise-Hardware. YMMV though.
https://www.amazon.de/ICY-BOX-Externes-Festplatten-Aluminium...
These are raspberry hats that allow that, like the radxa penta sata hat or the geekworm x1009. I don't know how good they are though
Odroid H4+. Four SATA ports, 2x 2.5gig Ethernet, IBECC. Not a compute module/carrier board combination though. Also x86_64.
Yeah, I've been considering just going for an x86 thing, though finding a mini pc with 4 satas might be tough, and there's the whole raspi ecosystem that would help with troubleshooting. I'll add that to the potential list though =)
The easiest way might be getting an ITX board with N100/150 from CWWK or Topton. They usually come with 4-6 SATA ports and the CPU soldered on. Then you just buy your favorite case and plug in your drives. Excluding the cost of the drives you might end up spending about $250, depending on your case and PSU selection. The problem with using SBCs or mini PCs that way is that you need an enclosure to hold, power, and cool down your disks, and even that is assuming that you can actually connect them via a SATA/SAS connection which may not be easily available from within your case.
Check out Hardware Haven on Youtube, he tests a bunch of different small form factor NAS computers and similar. I find the whole "raspi ecosystem" a bit overrated, for the most part it's Linux you're working with regardless of hardware.
If you go with x86 you've got a bunch of people doing it under the "homelab" banner, so you'll hardly be alone.
I am a bit confused because I had read that the RPi5 had dropped hardware acceleration for video encoding. Is that wrong?
Because that wouldn't be a faster, drop-in upgrade, right?
It isn't strictly a dropin but it is usually a great upgrade. If you have the power then video encoding is faster and higher quality on the RPi5. If you are battery constrained then it can certainly be worse.
Apparently the new ARM cores can do decoding faster than the Pi 4's hardware decoder. But no encoding, in fact.
The Pi 4 encoder was in hardware but that doesn't imply it was ever actually particularly fast. I'd bet the Pi 5 CPU+RAM improvements are enough to beat the hardware encoder of the Pi 4 and still have enough CPU leftover to beat the Pi 4 CPU unloaded.
Depends if you’re encoding video or not, surely?
> It boots up in seconds,
> 23 seconds until ssh login
Well ok, technically correct, but I expected faster when I read that.
I'm guessing that's with whatever stock software/firmware stack Raspberry PI themselves ship, surely can be made a lot faster by not having to probe various hardware ports you know aren't used, for example.
I didn't have it complete in time, but you can use pi-gen-micro from Raspberry Pi (it's a newer project) to build a very lightweight boot image now, on the order of a yocto/buildroot image, but with Pi OS as the base.
Just pruning a bunch of systemd units will get you to half that time.
Yes, but you could do that with most devices running linux.
Milliseconds, even!
Sounds pretty sweet. The Raspberry is now firmly inside a price territory where it has a lot of Intel/AMD rivals.
The Raspberry Pi Zero for $5 was never widely available. I'd love to see something in that price range that is actually available to buy. Hopefully also with low power requirements like the original Pi Zero.
The Zero 2W has much better availability and at $15 it should still be cheap enough to essentially be a no-brainer if specs fit your needs.
THe Raspi CM is also used in some synthesizers like the Korg Wavestate or the Ableton Move.
Are they implemented on bare metal, or is modern synthesiser just a regular computer with a keyboard attached?
The Korg Kronos (and I suppose the Nautilus as well) are just regular x86 PCs (I believe using Atom CPUs) running a custom Linux kernel and custom software
That's super interesting! Has anyone managed to dump the firmware? :-)
You can probably ask the one who did the most reverse-engineering :) https://marcan.st/2016/06/hacking-and-upgrading-the-korg-kro...
As far as I can tell, there's no custom "firmware" to speak of, it's a patched Linux kernel running a kernelmode audio synthesis engine and a usermode app for the UI
The Ableton Move is just a CM4 with linux (abletonos) running and using a SD-Card in the box.
https://www.reddit.com/r/ableton/comments/1fzd9ln/move_is_a_...
This CM5 is at least as powerful as laptop CPU from ten years back I'm really surprised nobody one is coming up with upgradable laptop platform or shell powered by RPi compute module. Couple with the available maximum 16 GB with ECC available with CM5 this can be a winner since laptop RAM capacity has been stagnant for around ten years now [1].
[1] Why do laptops have so little RAM? (2015):
https://news.ycombinator.com/item?id=42234501
There are Pi-based laptops (Pi-Top, CrowPi), but it's very hard to compete with used laptops. And not many people like ARM-based OSes (which is usually Linux or the awful version of Windows).
The Raspberry Pi is one of the only random gadget purchases that I've made over the years that found their way to the back of the drawer, then found their way out, back to the drawer, and out again. It's a really useful general purpose tool.
It's currently running at my parents' place as a VPN server to let them access their smart home and cameras from outside the house (I didn't let them expose these to the internet).
I got a raspberry 4 recently and was tinkering with it. I was appalled at the apparent disk IO performance of the thing, opening a browser takes several seconds.
Is that just me or do I have a particularly crappy SD card? Anyone got any tips on how to improve disk IO? Would a USB3 external drive help with this?
> Would a USB3 external drive help with this?
Emphatically yes. It's an entirely different experience using a RPi 4 with a quality SSD. [0] It's also more expensive, so you have to decide whether that's important to you.
[0] https://www.samsung.com/us/computing/memory-storage/portable...
In my opinion it's a mistake to split the difference with a cheap SSD or one you need to plug in using a SATA to USB adapter. (troubleshooting first step: if it's not the power supply, it's the SATA to USB adapter...)
Perhaps the M.2 HAT+ and a decent SSD?
USB3 could plausibly be faster than the SD card as the spec allows for 5Gbps+ but I don't know about Pis that much, I'd defer to someone with experience.
As a photographer I can tell you that SD cards, even high class ones have laughably awful speeds - any proper camera that needs to do video will use CFexpress Type B cards which are tremendously faster.
> Perhaps the M.2 HAT+ and a decent SSD?
That won't help OPs Pi4, the PCIe interface is new on the Pi5. USB is the best you can do on the older ones.
Good to know!
Proper cameras use film ;)
Seriously though, there are plenty of top-end cameras that use SD cards (all Nikons, for example).
EDIT: completely missed that you said video, ignore me!
I'm always astounded how little people know about this, but you need to get an Application Class rated card, preferably A2. [1] The application rating essentially ensures a certain minimum IOPS performance, which is critical because most cards are only intended for sequential reads/writes.
[1] https://en.wikipedia.org/wiki/SD_card#Application_Performanc...
The Pi 4's IO is not great, but the choice of SD card affects it a lot. Brand seems to have more of an impact than speed (or at least the stated speed of the card). Cheap knockoffs will absolutely kill performance. That's why Raspberry Pi started selling their own branded SD cards.
Using an external USB3 drive also helps but it's still pretty slow. I found it too slow to use as a SAMBA file server when connected over WiFi, even though it was hardly pushing the CPU or memory. It would regularly fail to transfer large files.
One of the main selling points of the Pi 5 is faster IO. I'd love to see something tiny and power-sipping like the Zero but with the enhanced IO of the Pi 5.
I am using a USB3 drive as a samba drive and it works fine, although I use it mostly for streaming video not to transfer large files.
My OS however is in the SD card which is Sandisk which I assumed was the good fair price brand.
There is quite a bit of variation between SD cards, browser taking several seconds is surprisingly bad.
On Pi 4 I use a 2.5" SATA SSD (WD blue) on a startech USB3 adaptor and the performance seems to me much better than even good quality microsd cards.
No it's not just you. I/O on the Pi4 and below is awful, but for most applications it doesn't matter. Not sure on the Pi 5 as I jumped ship after the scalping and overall negativity around the foundation's activities in the last few years but I expect it's a fair bit better.
In general, sdcards are slow... most use kernel cache lazy page flush if the pi4 has 8GB ram, and F2FS to improve flash reliability.
After, "sudo apt-get install preload" it will cause previously used programs from the last session to auto-cache into ram. Combined with the above setting, the usual programs will launch much faster.
There is also the option to use an external SSD drive for better performance.
Part of the fun of pi4 was getting the OS optimized, but unless you have a legitimate "Sandisk Extreme Pro U3" or >30MB/sec rated sdcard it can still be slow (people have seen 4MB/sec on old counterfeit cards.)
sdcard bus overclocking is a thing as the pi tries to maximize slow hardware compatibility by default. Note this feature is limited on the pi4, but can significantly decrease boot times on older models.
YMMV =)
This feels like a good place to ask this.
Has anyone who sees this attempted to run Home Assistant OS in a Home Assistant Yellow "compute module board" using a Radxa CM5?
It's tempting to me because the Radxa CM5 uses a mainline kernel instead of the Broadcom-modified kernel the RPi CM4 uses, but I understand that there may be some level of work needed to be done to support the compute module + board combo, and I'm curious if that is true for the Radxa CM5 + Yellow board combo.
Or have I misunderstood anything?
There is no "Broadcom-modified kernel". There's a fork with changes needed to support all Pi hardware at https://github.com/raspberrypi/linux. Unlike some boards were you're eternally stuck at old versions, the fork is actively maintained (currently mostly at 6.6, soon 6.12) with its changes regularly getting upstreamed. You post make it sound like the Pi kernel is a mess, which it is absolutely not.
> There is no "Broadcom-modified kernel". There's a fork with changes needed to support all Pi hardware
That's literally the same thing
Right. I read it as "modified by Broadcom". Just wanted to point out, that this isn't inherently a bad thing if done right.
I get where you're coming from, didn't mean to make it sound like that but I do see it the same way zokier seems to. End game for me is booting a vanilla kernel compiled entirely from open source code and having access to all the boards hardware.
I was tempted to do something similar when I got my Yellow because the low availability of the CM4s back then, but it seemed that, at least with older Radxas there were problems with kernels or compatibility. It might work with the newer one if you say it's mainline kernel but haven't found anything in the forum in quick search...
I am keen to try. But I'm even more keen for someone else to tell me they have tried it and the result of said test because I am not keen on my home automation system being down for a longer period of time...
I'm really torn as to whether 'smaller than a credit card' (when something is clearly much, much larger than a credit card), or re-purposing a previously well understood and generally agreed upon definition of 'artificial intelligence' to now mean a frequently wrong or bewildered statistical / language model, annoys me the most.
Credit cards are bigger than you think. The CM5 is smaller (55mm x 40mm) than credit card (85mm x 54mm). You could almost get two on one credit card.
I always thought it was deceptive to compare regular Pi to credit card. But I just checked and credit card is almost the same size as regular Pi.
You're forgetting the third dimension, which comes in at 4.7mm. That indeed makes the CM5 much larger at multiple times the volume - if you want to be pedantic that is.
> when something is clearly much, much larger than a credit card
Is it? Too me it seems smaller both in width and depth, but the height seems some mm or more than a typical credit card.
Only two total MIPI ports (DSI/CSI, for displays and cameras respectively) compared to 4 ports on the CM4.
I wonder how the USB 3 works, considering it is drop in compatible with the CM4 which only had USB 2.
I guess "mostly" refers to the "drop-in upgrade", not to the 2-3x faster. The pinout has been modified slightly, so it's not 100% compatible.
Yeah I wasn't too sure about the compatibility, but JG mentioned he was able to just pop it in his NEC Display and it was shown in a blade carrier.
Looks like the Radxa CM5 is still significantly faster than Pi CM5, while only being slightly more expensive (113USD vs 95USD for the 8GB/64GB model from what I could find). Anyone here has any experience with it?
As the article says, you pay for that with a lack of support compared to the Raspberry Pi.
> That's another CM5, this one being made by Radxa. It uses a Rockchip RK3588S2
Would rather go for this. The NPU on the 3588 are now starting to be usable for stuff like basic LLMs
Unsure about the S2 variant though
Does it still only encode to H.264, no HEVC/AV1?
Technically the BCM2712 doesn't have any hardware encode support, only software/CPU.
It includes an H.265 (up to 4K 60 fps) hardware decode block—I tested that, and it works fine—but no other video hardware encode/decode, unfortunately.
The CPU is good enough for up to H.264 1080p encode at least, and with some tweaking could get a 4K stream or multiple HD streams encoded... but it's not ideal for that use case.
I know it should theoretically work, but has anyone already tried to replace the CM4 of a HomeAssistant Yellow by the RPi CM4?
https://www.home-assistant.io/blog/2024/11/27/home-assistant...
Thanks to Raspberry Pi providing us with an early sample, we have been able to add Compute Module 5 (CM5) compatibility to the Home Assistant Yellow
Wow!! Those are really great news, thank you!