The market for Pi is dead. The article pretends that you can buy a pi for list price $5 or whatever LOL but I just checked Amazon and real humans are paying $170 to $205 for a pi.
The article claim that you can't compare a $150 "mini PC" to a cheap Pi is unrealistic.
I got a lot of flack (elsewhere not here) for setting up a mini-cluster for K8S that used mini PCs that I claimed are cheaper than Pi. Note that I can actually buy a mini-PC for $150 whereas you can make up any imaginary MSRP you'd like for a Pi but you're paying $170 today on Amazon regardless if the unobtainable price is $5 or $99.
There are plenty of low performance PC compatibles from Beelink and similar places that cost less delivered tomorrow than a Pi delivered tomorrow AND have higher performance specs than a legacy Pi.
I have been getting Rpilocator notifications for 6 months and wasn’t able to get a pi. If you start your order 15 seconds later you are basically out of luck. Same story for all my friends. So while I get that some people got lucky, it’s not a matter of “a little bit of patience”.
It's really wild. NUCs and other mini PC's from HP/Dell are _so_ much better at similar if not cheaper prices.
My theory is that, despite the economics for using Pi's completely flipping to making it a ridiculous decision for most applications, the decade+ of tutorials, Youtube videos, forum posts, github READMEs, etc all describe using Raspberry Pi's as the way to deploy XYZ open source project is enough of a "moat" that it just keeps the demand high enough to suck up supply.
Also I don't think that people realize how good 3-5 year old technology is relative to the price. There's definitely a great arb there right now, and I think it's just going to get better.
The Raspberry Pi brand is wildly strong, but it's eroded for me because of scalpers and lack of availability - hopefully they can right the ship before it's tainted for good.
> NUCs and other mini PC's from HP/Dell are _so_ much better at similar if not cheaper prices.
> the decade+ of tutorials, Youtube videos, forum posts, github READMEs, etc
While this doesn't apply so much to mini x86 PCs, most random ARM SBCs have hit or miss software support, as Jeff mentions in the article.
Admittedly, I would prefer a RPI over whatever board this article was about, if I could get one for (near) MSRP. Part of it is the price, a "$35" computer is cheap enough to fry it making a mistake, or permanently lock it in a box as a camera controller in the garage, or whatever project idea requires... then buy more for the next project.
A lot of RPI tutorials treat it more like a software appliance, where you run one thing on the device, and if you want to run a second thing you use a second RPI - more like a Linux-based microcontroller than a server. Thats only possible at the $35 price point. I have a used PC that I use as a server at home for a few things, running a 2015 i5 chip. I would never use it to just run one thing, or lock it up with a project. It basically has to sit in my house, hosting a bunch of things at once. Meanwhile I've been buying a handful of Pis since 2015, and I have them scattered over the house... one controlling smarthome stuff, one controlling the 3D printer... one in the garage... etc.
I'd say its pretending in the sense of comparing a vaporware product you'll never receive or only available for hyperinflated price from scalpers, vs delivered from Amazon Prime on your doorstep tomorrow.
Like comparing a Chevy Volt on the used car lot that I could buy tonight vs a hypothetical Tesla concept car with an anticipated delivery date of 2026. I'm sure the vaporware has better specs but its not "fair" to pretend to compare them.
But, you are in a sense technically correct on a small scale, yes.
The Rock 5 also takes a while to ship right now (I was in an early batch of pre-orders), and the Pi 4 is available at MSRP for either the lucky or the patient. Using rpilocator.com and notifications, most people I know who wanted to buy one have been able to find a 2, 4, or even 8 GB Pi 4.
CM4 are much harder to find though. Hopefully that changes this year, but from what I've heard it will take longer for some reason :(
> The market for Pi is dead. The article pretends that you can buy a pi for list price $5 or whatever LOL but I just checked Amazon and real humans are paying $170 to $205 for a pi.
"Generally available" means that I can order one or a dozen today for MSRP or less and get a shipping date between "today" and "end of the week".
"In demand" means that I can order one today for a little over MSRP or look around for a sale to get one under, but I might not be able to find a single supplier who wants to sell me a dozen at once.
"Speculation bubble" means that I can't find anyone to sell me one at MSRP without going to ridiculous lengths like setting bots out to scoop the competition, and I certainly can't call a distributor to get a dozen at MSRP -- but if I'm willing to pay a high premium, I can buy one today.
Not for new designs, milking .edu and .gov for all they can get, great specs but no delivery date, easier and more convenient to buy weed than their silicon ... it had a good run but it's just done now.
Kind of dead for hobbyists, I'd say. Pi's used to be a super low budget, easily accessible computing platform, which is what made it so great for almost every project that needed more than an arduino. Now it's usually the hardest part to get, and the complete opposite of a "throwaway" computer that you can stick everywhere.
I'm sure the pricing doesn't bother enterprise or industrial users, and the Pi Foundation is almost entirely devoted to catering to them.
But hobbyists are super sensitive to pricing and ease of access, and the only reason to still use it is that tons of projects are already built to run on Pi's. Even then, Units are almost entirely allocated to enterprise/b2b demand, so the retail customer market is so extremely undersupplied that it's just not worth actively hunting for days just to run an interesting app you wanted to tinker with.
The market for Pi is dead. The article pretends that you can buy a pi for list price $5 or whatever LOL but I just checked Amazon and real humans are paying $170 to $205 for a pi.
That's a temporary situation though. The RPi foundation folks have said the expect supply to be back to normal sometime this year.
Its been the same credibility gap from those people since the unobtainable $5 Pi Zero back in 2015. Eight years and many models later they still don't ship at MSRP despite what they claim. However, their competition can and does ship on time what they advertise at MSRP. Its like two different marketplaces.
They do ship at MSRP, to a large number of corporate partners that embed the Pi in their products. Apparently at the rate of 30M per year. However much like the bitcoin miners who slurped up the GPUs for years, there's precious few Pis available to end users.
So much so that out of 30M a year the Pi foundation decided to allocate an extra 100k for end users, which did improve availability, but still it's clear that most production is going to corporate partners and not end users.
But normal people are getting Raspberry Pi's. Some people by using rpilocator and patiently waiting for units selling at MSRP to come into stock at official retailers, and impatient people by paying inflated prices to scalpers. But anybody who wants a Pi can, to a first approximation, have on today.
And not to beat a dead horse, but there's good reason to believe that this situation is temporary and that supplies (and prices) will stabilize sooner than later.
All of this "the market for Raspberry Pi is dead" stuff seems like pure hyperbole to me.
Pi market ... sure. ARM based SBC not so much. Just get one of the zillion Pi clones that have many advantages like speed, better PCIe, not using USB for storage/net, more ram, better perf/watt, NVMe, etc.
I don't see them as having any major credibility gap personally. Until the crazy chipageddon stuff started (which, BTW, affected the entire industry) I was routinely to buy Raspbery Pi 4 Model B boards at (approximately) MSRP. Same for the Pi 3 before that.
Call me crazy, but I don't have any problem believing that things will get back to normal soon, and that Pi's of various sorts will indeed return to MSRP, or very near MSRP.
Not sure why there's so much focus on the Pi and it's availability, that's not what the article is about.
The article is about the RK3588, it's a substantial upgrade in CPU performance, I/O performance, and network performance over the PI4. It also connects things directly to a much faster PCIe instead of relying on USB connected chips. I think of the Pi4 as popular, but a toy.
I compiled either rust or go (I did both, so I forget) 7 times faster than a 8GB ram Pi4 with my rk3588. It's really in a different class than the Pi4.
The rk3588 is in the order it today, get it this week type availability. I found it for sale in 3 places, 2 with ship times of 2 weeks, 1 with a get it the day after tomorrow.
MiniPCs are in the similar class, but compared to a $120 rk3588 with 8GB ram are often a bit lacking. Often MiniPCs are used, less than 8GB ram, don't have two 2.5G ethernet, require a custom wall wart+barrel connector, often have a fan, are usually in a cheap/plastic case (instead of a CNC machined case with integrated heat sink for $20), and in a much larger case. My case is 3.75" x 2.75" x 1" or so.
So yes, you can get a used MiniPC, maybe you don't mind missing a 2nd ethernet port, and sure you can buy more storage/ram. But for the size and power efficiency it's going to be hard to match. I suspect the x86 in the $120-$140 range (with 8GB of ram) are going to be slower, but I'm hoping someone posts some benchmarks to compare.
> There are plenty of low performance PC compatibles from Beelink and similar places that cost less delivered tomorrow than a Pi delivered tomorrow AND have higher performance specs than a legacy Pi.
This. I replaced my Homebridge/PiHole RasPi 4B 4GB with a J4125 Acer Desktop system at a cost of $88, all in with all accessories needed from ebay. It has 4GB of upgradable ram, a reasonable SSD, 1G ethernet, and a bonus Sata port for a NAS. It even included a power supply! Pihole's web UI is shockingly faster for me too.
Sure, better than a Pi, but better then the rockchip rk3588 system?
From a quick look it looks like the MiniPC has half the cores, half the max ram, half the shipped ram, much less I/O performance (SATA @ 400MB/sec vs NVME at 3GB/sec), much less network (1 x GigE vs 2 x 2.5Gbe), much older/slower USB (2.0 vs 3.1), hugely larger (looks like 20x the volume), and likely has a fan.
So sure the MiniPC is a good for for some uses where power, noise, and size don't matter. But if you want a router, firewall, server, or desktop you might well be much happier with an 8GB ($140 with nice passively cooled metal case) or 16GB rk3588.
On performance it's hard to say, if you want to compare performance post any open source benchmark you want. Although I don't have the 3GB/sec NVME JeffG benchmarked in the original post, but anything opensource that exercises the CPU/ram should work. I do have a kill-a-watt around if there's interest in perf/watt.
For running DHCP, router, dns-masq, firewall, and the mentioned PiHole seems like a much better fit for a Rk3588 @ a few watts and 10 or so cubic inches than a used system @ 600 inch^3. I wouldn't be at all surprised to see the Rk3588 faster than a celeron J4125.
"better" is in the eye of the beholder. You can get very near equivalents to the rk3588 on the x86 side. N5105/N6005 are a pretty good match for CPU benchmarks and better GPU, better I/O (more PCIe), SO-DIMM slots, better software support. All for a similar costs. I got a Intel NUC Essential N5105 for $160 6 months ago (SSD/RAM were separate).
The RK3588 is a 4 big core 4 little core design, so throwing out 8 cores as a point of difference is kind of misleading. Yes, there are 4 extra small cores in there.
If you just want a small cheap system, used ebay systems are pretty hard to beat. J4125 class CPUs are an excellent choice. At this point "faster" is less important than "works" I think. And the RK3588 isn't even touching the price you can get a used J4125/J4105 for. $55 https://www.ebay.com/itm/115671002802
Claiming a J4125 is "big" and "noisy" is similarly misleading. J4125 is a 10W TDP. RK3588 is a 12W TDP. You can passively cool either. N5105 and N6005 are similar power profiles. My J4105 idles at 3-5W, runs proxmox with my firewall and a bunch of VMs for databases and servers. 32GB RAM.
I couldn't find a Rock5 with case and power supply for anywhere near $150, and not even a bare board with 8GB for that. Let alone your claimed $140 for the full package.
Then there's architecture... A lot of packages you can find off the shelf for x86 just aren't built for arm. It bites you in the random places and burns time. Lots of time. I've burned far too much time trying to make ARM servers work.
I was curious about system power (not just TDP), so I hooked up a kill-a-watt, at idle, ubuntu 22.04, normal stuff running (like sshd) and so far it's taken 7 hours to accumulate 0.01 kwh, sadly not very accurate yet, so I'll leave it go till at least 0.02 kwh, looks like an average idle around 1 watt.
I've been pondering some easy benchmarks from the command line, so far I've come up with:
To give some idea my Rk3588 idles somewhere 1.75 to 2.2 watt range, ran it for 17 hours mostly idle except for some apt upgrade/installs on a kill-a-watt to accumulate 0.03 kwh.
Using geekbench 5 your J4105 gets 396 single-core and 1294 multi-core. The rk3588 gets 684 single-core and 2680 multi-core. So rk3588 is 1.7x the single core performance and 2.0x the multi-core performance at around half the power (2w vs 4w).
I've not found make problems with ubuntu 22.04 LTS. Large source code compiles like building rust/go just work, and so far I've found all the ubuntu packages I expected. Sure I'm sure there's some codes that don't like AArch64, but the situation does seem to be improving.
In my personal case, I need to be able to run couchdb. No arm arch package out of the box that I've been able to find. Last time I tried it on arm I gave up after several hours of chasing dependencies while trying to compile from source. Waste of time... Just works on x86. 2 minutes, done. Then I found proxmox, building portalable containers and VMs, and I'm just that much deeper in x86 land. Arm cloud support, while improving, is no where near x86 availability.
Benchmarks are fine and good, but immaterial. I don't really care about 50% more or less speed. It's a server. If I do my job right +/- a few 10s of ms for response time is immaterial. The speed delta between the EMMC and SATA M.2 may well make up for raw processor speed in my use cases which tend to use the I/O.
Your particular board is very niche. I mean, yeah, you have nice ethernet. But that's it. I need my M.2 slots. I need more ram. I could get an x86 board with everything for more $$$, nvme, 2.5-10GBE. It's a compromise.
The J4105 is like 1/3rd the price of the rk3588/rk3588s.
I could go on. To each his own. I'm glad you like your board :) Enjoy it.
If you get an atom line CPU a smaller I/O suite. Say a dell 3040 or 5070 and your at 4W idle - similar to the Rock5 (and cheaper if you're buying used on ebay). When you get to the full fledged core lines you're looking at 8-10W as a baseline. My Pi4, when I was running one, was 2-4 Watt @ idle.
I don't have it on a kill-a-watt even though I should. As the person below you said, it's like 1.5 to 2x power consumption vs a pi4 at idle (Under 10w), but that is still comically small amounts, about the same as a Google Nest Hub.
I wouldn't call it a dead market for Pi. I would say that the demand by some people has push out others (like me). I'm not buying a Pi for $100 or whatever, assuming it's even in stock. I ended up buying a mini PC for about $110 2 years ago, a refurbished desktop with decent specs for $120 last year, and using ESP32s for some of the smaller projects they can handle.
Nothing against Pi. I have some older models and their great. Just the market availability and price gouging makes them impractical right now.
No, the article does not. I just rtfa and even Jeff is stating that once you pass the $100, you're in a different class altogether, and are competing with used mini pcs. He even mentions that trying to get a pi at MSRP isn't likely to happen any time soon, so even if we were to compare a $35 MSRP board to $150 board, the pi still has 2 advantages, cheaper price point (assuming MSRP PIs are available), and working software.
A good take away about modern SBC landscape for hobbyists: you can get much better specced "mini" x86 PC at the cost of slightly more power consumption, but with benefits of much better OS and hardware compatibility at a similar price as an ARM SBC at this point.
Now, like January 2023 numbers, you can get a brand new mini micro PC from a no-name brand with twice the memory and a large SATA drive and vastly faster CPU and probably a pirated win10 or win11 installation for 10% to 20% lower price than a ARM SBC supplied without storage, and usually the slightly higher electrical power is more like double.
Although, if you have twice the ram and a huge SATA and huge amount more CPU power, that just means it spends a higher percentage of time in sleep mode, so its probably about same total energy consumption, just higher peak power consumption.
Its worth pointing out that prices have gone up somewhat but "generally" "away from overinflated areas" you pay a bit more than a buck for a watt-year of wall outlet power, so trying to spend an extra $50 upfront to go ARM and save four watts will take a VERY long time to pay off, perhaps infinity depending on inflation and interest rates, if you're plugged into a wall; battery applications have stranger financials of course.
> Now, like January 2023 numbers, you can get a brand new mini micro PC from a no-name brand with twice the memory and a large SATA drive and vastly faster CPU and probably a pirated win10 or win11 installation for 10% to 20% lower price than a ARM SBC supplied without storage, and usually the slightly higher electrical power is more like double.
NUCs and the like are about twice the cost of Beelink and the like.
You can search on Amazon and pay about 10% to 25% premium to the dropshipper or get a general gist of the market and then order off the website (where it'll cost like 20% less but then you have to pay for shipping, no such thing as a free lunch). Sometimes old models will be cleared off cheaper on Amazon, you really need multiple tabs open while you're shopping. Usually you get a windows 10 pro or win11 installation of questionable genealogy installed on a fraction of a TB SATA for the price.
Its hard to keep track of who's drop shipping from the factory for you on Amazon vs who's just entering orders for you on AliExpress vs who's selling natively from stock (like traditional retail). Unlike the ARM SBC marketplace they actually have stuff in stock for sale when they advertise a MSRP, I haven't run into any issues like that.
I'm pretty happy with the ultra low end of the Beelink products. Their numerous competitors are likely pretty similar.
NUCs and the like are more expensive than buying ARM SBC boards, but they are probably more reliable. Probably. Of course for twice the price you could buy 10% or 20% more devices for a project.
I set up, well, am currently setting up, a medium sized Rancher and Elasticsearch cluster for reasons that are hard to explain, so I needed an odd number around half a dozen. So far, so good.
Ah ok, I thought you were talking about mini PCs in the same form factor as RPis (very small, credit-card sized). Yeah I have a great no-name mini PC as my primary desktop. But it's still 4x bigger than a RPi I think.
Very true, though without the GPIO, so depending on your project(s) you may need to hang an Arduino etc. off of it also. I would also love to see some really cheap and really small x86 machines, like the equivalent of an RPi Zero W, say $20, and small...
"you can get much better specced "mini" x86 PC at the cost of slightly more power consumption" ... if you buy used, which doesn't seem like a fair comparison.
Generally even the mini PCs are much larger than the rk3588.
I've bought recently a Minisforum N40 (Celeron N4020 CPU), and man, it's much faster than the Pi. Installing Manjaro Linux was a breeze, all peripheral supported out of the box. I bought used on ebay, but new is priced like €150, which competes directly with the Raspberry Pi 4, and this one comes with 64GB eMMC and 4GB RAM. No exposed GPIOs though, so it won't be the ideal choice for hardware hacking, but for media boxes or tiny desktops it beats the Pi hands down.
That does look reasonable. Seems like a lower spec though, 1/2 the ram, fairly slow CPU, single ethernet (instead of two 2.5G ethernets ports). I do like usb-c for a power supply over whatever custom wall wart+barrel connector.
I'm curious about the Celeron N4020 vs rk3588. Keep in mind that the rk3588 vs Pi4 isn't really in the same class. In a compile based benchmark my rk3588 was 7 times as fast as the Pi4. JeffG's geekbench multi placed it 3+ times faster than the Pi4. On the same compile benchmark the rk3588 was about half as fast as my quad core xeon system from 2015, not bad for a few watts.
Do you have a transcoding benchmark you prefer to compare? Or something else media related? Or storage? JeffG managed 3GB/sec which is pretty respectable for a small/cheap widget.
I agree it's not a fair comparison. However, that doesn't preclude it from being a better choice. The used SBC market seems to be extremely small compared to other used markets, and with less of a markdown. It's hard to even find new SBCs in some cases, especially not marked up.
Keep in mind that the rk3588 is MUCH more available than the Pi4. The day I looked I found them at 3 stores, but decided to use Amazon with quick shipping instead of the cheaper aliexpress with 2-4 week delay. Keep in mind the rk3588 is smaller and lower power than most (all?) mini PCs. It's takes up less space than my phone on my desk and is about twice as thick. The passive heat sink/case is cool to the touch under normal use, and only gets noticeably warm if I'm running something intensive, like a make -j 12 on a large build.
I've heard Raspbian and OpenWRT have newer kernels working, unsure if all the required bits are upstreamed though. Progress is being made though, and people have newer kernels working.
Dunno, I got a complete working rk3588 with 8GB ram for $139 in a nicely machined metal fanless case, that just needs a modest usb-c power supply. I could have gotten it caseless for $20 less.
So it's silent, reasonably fast, and enough ram to run everything Iv'e tried. Software compatibility has been good, my first choice of OS worked (Ubuntu 22.04 LTS), compiling rust and go "just worked", and it's a decent little server.
So generally smaller, quieter, more efficient, and lower power than whatever plastic encased random mini PC. I think I'd have a problem getting something nicer and cheaper in a PC, even used.
I'm not arguing for the Pi4, I'm arguing for the rk3588. On aliexpress it's $119 with 8GB ram, and $139 with a nice CNC machined metal fanless case.
I've seen decent x86's, but matching the spec (case, 8GB ram, 2x2.5G, usb-c power supply, fanless, etc) leads to some pretty old x86's and higher prices. Not to mention that I'm far from convinced that chips like the celeron N4020 are an upgrade, even ignoring power efficiency.
I’ve been using ARM SBCs for ages now (mostly Pis, but also Hardkernel and a few “alternatives”), and _all_ of them except the Pi have absolutely dismal software support. Armbian does a pretty good job on some alternatives, but until manufacturers ship with _at least as good_ software as the Pi their value proposition is essentially negative unless you have a lot of free time.
And now that the power envelope is closer to a gen 4 Intel (which is trounced by modern Celerons like the n6xxx and n50xx series), the functionality gap is more obvious.
To be fair, though, I think that most people have unrealistic expectations of what low-end SBCs can do as far as “desktop computing” is concerned.
I do use a Pi as a very nice, utterly silent and fast thin client (backed by an i7 in a closet), but if I wanted to run local apps (browser, light office, etc.) I’d just go and get a NUC or an $150 refurbished small form factor desktop (there are great Ryzen boxes with modern mobile APUs from Minisforum and Beelink, but I am extremely wary of their quality control, and the really interesting ones are too expensive to risk it).
I have been monitoring Reddit threads for a few months ever since the 5900x and 6800/6900 APU variants started shipping. Unfortunately there don’t seem to be any mainstream vendors in this space, so you’re essentially dealing with small Chinese companies with no support structure, who have… unusual concepts about board design and cooling, and who manufacture in small batches. The yields on those things are statistically not good in general.
Excuse me for presuming, but that seems to be a 1st world view of ARM SBCs usage. There are plenty of places in the world where this may be the only computer used until traveling aboard or higher education, so the ability to run actual 'desktop' applications, super cheap, plugged into peripherals, is still a worthy usage of a pi and its ilk.
I have never seen anyone in my third world home country just carrying or even owning an SBC. Even the poorest people just buy a phone because it's wildly more useful and does not require a peripheral. Otherwise, they buy an older computer that can be had for cheaper than most SBCs.
Actually, I'd argue that the market segment is squarely a first world one, mainly because it's far from the most convenient computing platform and alternatives are far more useful when you can only afford to own a single computer. Being unable to make calls or use messaging apps (access to whatsapp is almost a requirement in tons of countries), and coming with no built-in battery makes them a complete non-starter as a potential main device.
To what extent are these used for this use case vs. very inexpensive laptop computers or Chromebooks? In most of my travels I've seen a lot more extremely cheap laptops than ARM SBCs.
I agree that there's a stronger justification for the desktop (and TV-dongle / emulation station, for which there's just a lack of better-suited competition) use case than the home lab one.
Additionally, many places have flaky or nonexistent power infrastructure. Having a computer that runs at 2-3W is a lot easier on small solar + battery than a computer running at 8-10W idle.
This is my only requirement for a Raspberry Pi killer - to work out of the box with stock kernels and distributions rather than needing some hacky device-tailored distribution that will eventually go unsupported, or worse some hacked up kernel that will stagnate from the day it's released.
I'm not running a server farm with 1000's of nodes and full time staff to make Linux work on them. Rather I've got a bunch of homogeneous devices for personal use and the more similar I can make them all the better. It's why I moved away from the embedded ARM dumpster fire to low power amd64 motherboards for my Kodi/WifiAP machines. They generally just work, and if they do break there is still no bespoke tinkering.
The Raspberry Pi only mitigates this by being popular enough that there is interest to keep it going and develop around its warts and keep the software going. But it's not without its pain when you step away from Raspian.
What kind of AMD motherboards are you using? I need to have ARM hardware (porting software, etc.) but am interested in getting something x64-based I can stack (3-5 units) and that doesn’t break the bank (also, something that doesn’t require individual PSUs would be great).
I'm sorry that I don't have a good answer for your question. I've got two such machines, both regular ITX motherboards. The main is an Asrock N3700-ITX (only driving 1080p), the secondary is an ECS KBN-I/2100. The main AP is the "router", which is a Ryzen 5700G on a full ATX board.
If I needed another AP, I would get another decently powerful but lower idle current machine to replace the main entertainment center one, and then trickle the N3700 elsewhere. Or if I didn't want to do that, I believe I have seen one or two amd64 boards targeted at being RPi replacements. The biggest thing is I want expansion PCIe so I've got flexibility later.
For power supplies, I'm using PicoPSUs on both. But I've also seen motherboards that take 18VDC. Or if you are willing to do some wiring you could power multiple motherboards from one supply.
If a bunch of wall warts is your only problem, you can solve that by using one larger power supply (same voltage, higher current) and wiring all the units in parallel. Although even though it would be in my wheelhouse, I've never really done this for production - it makes it harder to change the setup and more prone to error (reverse polarity and the like).
This. If it's not ARM System Ready, you will be stuck with some outdated kernels provided by the vendor. Unfortunately, it seems that most SBC are no certified.
I find it curious that Jeff doesn't even mention Armbian here and it'd be interesting to see if results differ. Apart from necessary board-specific patches, dev trees, and some defaults that make sense for typical SBC workloads, it's close to vanilla Debian and can run better than vendor dists.
Arch Linux on ARM has also come a long way in the past few years.
Orange Pi is a bit cheaper (starts under $100 for 4GB ram) but less slots. The hundred dollar barrier is where I start to balk at ARM sbc.
Fun seeing Jeff try out the expansion slots. The m.2 NIC is cute. I hope eventually CXL starts giving us some new smaller more modular add-ons. But it's not that small so I imagine cards being cable-attached in many situations.
It's wild how long the PC market has been able to go without facing intercompatibility issues. x86 everywhere is a slow thing to change course on.
I wish Intel's Lakefield had a cost-optimized sbc for it. That chip was so awesome, so small, so low power. With 1 very very nice core, solid gpu, and all ram on package. It would have made a very interesting sbc part. Trying to ship them in enough volume to get somewhat modestly priced chips from Intel would have been an absolute nightmare though, a huge gamble. That seems to be something brands like Chuwii have gotten really good at- finding what chips they can buy for below retail, and selling really nice machines around that at screaming value to consumers.
rant: I've learned, excluding microcontrollers such as the wonderful rp2040, every ARM & RISC-V SBC I've been interested in purchasing relies on some extremely proprietary and paranoid bits. If the device has a massive, undying following, like the Raspberry Pi, people will find a way to bring it into the future and keep it useful. If not, well, in as little as a year or so the manufacturer will likely simply tell you to buy the new one.
Truly useful, truly open hardware, cannot exist under current market conditions, for one reason or another. For now, we're all stuck with defaults; despite all of our progress, every computer you can imagine primarily using relies on someone's monopoly, and I imagine there is a nonzero amount of wealthy, powerful people who do not want this arrangement to change.
Rockchip devices have stellar mainline (that chip rk3588 doesn't have it yet), for business reasons (radxa relies on it for industrial support, Chromebooks push mainline very heavily). I can't speak for this new one, but previous rockchip flagship (rk3399) have exactly one binary blob, which is used for display port over usb-c. Everything else is supported as open source, and mostly supported mainline (Linux and uboot and atf)
Yes, this is a problem. ARM SystemReady certification program is aiming to solve it. Unfortunately the only boards that meet this certification are significantly more expensive.
Parent is meditating over the questionable choice of words in combination. The board could be half as fast or twice as slow when compared to the M1 Mini, but if it is half as _slow_, how fast is it actually?
Half as fast would make much more sense. I've seen things like "I'm 40 years young" or a laptop is 10mm thin instead of thick. Half as fast takes it so far though and doesn't make sense.
I'm surprised he had problems powering it up. USB PD will start at 5V as a safe default, and remain there if it cannot negotiate a higher voltage. My Rock5B runs fine at 5 volts from "regular" USB-C. When I fully load all 8 CPU cores the board wants about 1.6A (~8 watts). Perhaps Geerling's power adapter refused to provide anything without bilateral PD presence.
The 5W space is perhaps the space I understand the least.
Things get weird at 5W. There's phones and tablets, but these aren't really general compute devices. But then comes these SBCs like Rasp. Pi, Rock 5B, Le Potato, Beaglebone Black, etc. etc. These use phone/tablet CPUs but in a form factor conductive to USB ports and HDMI ports... like a SFF Desktop.
I've seen people say "Its a great small desktop!!", and hook up the Rasp. Pi (or whatever) to a $200 27" monitor drawing 80W and I'm just... confused by that. You're not making a good power-consumption argument, or a cost-benefit argument either, most of your costs are on the damn monitor (both in power and in $$$).
I think there's value in the $60 7" screens though, which return the form factor closer to its Phone/Tablet roots. With such a small screen, the price is dramatically lower, while the power-consumption plummets to 5W SBC + 5W Screen (10W total).
-----------
In theory, 10W total power consumption fits inside of 15.4W requirements on PoE. So there are real benefits to cutting the power consumption to an appropriate size.
I guess there's also a myriad of "headless" applications where a 5W Rasp. Pi (or competing SBC) is useful. But in my experience, most of those headless applications should be a VM and/or Docker image instead.
----------
With that being said: this Rock 5B seems to draw 15W worst case, which leaves the available area for PoE. That could be fixed with a local battery pack (1270 Lead Acid packs are only $35) + some kind of power-controller to even out the power-draw. I admit that's a bit of custom electronics, but those kinds of custom-electronic problems are why we have ATMega / Arduino Unos laying around, right? :-)
Haven't really heard the power consumption argument, perhaps it has increased in the last 12 months. In general the $35-70 SBCs (when they could be bought in retail) reveal an ugly truth: circa 99% of users don't ever need more than that in processing power/memory, even disk space with all the clouds for everything, hence all the $1,000+ machines (laptops, desktops, tablets, smartphones) are just massive overkill and a terrible waste.
> In general the $35-70 SBCs (when they could be bought in retail) reveal an ugly truth: circa 99% of users don't ever need more than that in processing power/memory, even disk space with all the clouds for everything, hence all the $1,000+ machines (laptops, desktops, tablets, smartphones) are just massive overkill and a terrible waste.
I suspect that quite a bit more than 1% of users play graphics heavy games or otherwise tax their machine beyond the practical limit of a cheap ARM SBC. If you'd said 2/3, I'd be much more inclined to agree.
True, however it is going to be harder and harder to pay $2,000+ for a top GPU to play the latest AAA game, and it is going to easier and easier to pay ~$10/month for a cloud-rendered gaming subscription. Stadia was just cursed for being a non-ads-based Google product, GeForce Now seems to be thriving [1].
Somewhat funny is that the top 2 and 3 PC games most popular in 2022 have 1993 graphics, Minecraft and Roblox [2]. Another list puts Minecraft on 4 and Roblox on 3 [3].
802.3af was the old turn of the century standard limit of 15 watts. You can probably buy new stuff used but everything is probably 802.3at now from around a decade ago, which is twice the power 30 watts. There is a new standard came out about covid era that AFAIK is totally vaporware at this time and that goes up to 70 watts for charging laptops or whatever nonsense. WRT the newest plus products I'm a little puzzled how they put 77 volts on something rated by the fire dept and NEC as "low voltage" but that's not my paygrade LOL.
The $10 to $20 price point for "PoE Splitters" runs around 15 watts at 5v fixed and you're going to have to pay the $20 to $30 price point if you want selectable 5 / 9 / 12 volt output and more than 25 watts or so.
The place to buy these things is the security camera field. They're all about wanting a couple amps at 12 volts for IR illuminators off a PoE cable and applications similar to that.
Like a lot of no-name power equipment don't expect it to last in hot summers in an attic and don't expect to run at 99% of listed current for very long without overheating. Its better than the totally fake budget USB charger market or mostly fake ratings PC power supply market, but its not like buying engineering stuff from big names where you can trust the ratings (like a Mean Well 15 watt industrial supply will actually output 15 watts at industrial temps for a decade or so, at least so far, but you'll pay $$$ for real true ratings as opposed to consumer imaginary ratings)
If you actually need 15 watts I'd buy a "25 probably fake watts" splitter and expect a long life from it for less than $30 delivered.
The market price was about "fifty bucks" for similar just a couple years ago.
> I'm a little puzzled how they put 77 volts on something rated by the fire dept and NEC as "low voltage" but that's not my paygrade LOL.
The newer standard uses all 4 twisted pairs, so its 71W power / 48V / 4 wires == 370mA on the average on each pair of wires. Though "worst case" 960mA due to unbalanced power draw it seems.
The 4PPoE standards also run with 1Gbps protocol, meaning all 4 twisted pairs are data. So all 4 are data + power. I'm not quite sure how it all works electrically... though I loathe to be the guy to design the splitters to pull power evenly from 4 different pairs.
VLM assumed one wire pair, 71W specification, 960mA worst case current draw == 70V+.
Which was incorrect, because there's 4 wires. In PoE classic, 2-pairs were for data + 2-pairs for power. In 4PPoE, there's 4-pairs for data and the _SAME_ 4-pairs for power.
I am talking about U6-Pro, specs here [pdf]: https://dl.ubnt.com/ds/u6-pro_ds which is asking for a 48V, 0.5A PoE adapter, so should supports 24 W max. the consumption is listed to be 13W though. U6-LR is 17W.
> The original IEEE 802.3af-2003[1] PoE standard provides up to 15.4 W of DC power (minimum 44 V DC and 350 mA)[2][3] on each port.[4] Only 12.95 W is assured to be available at the powered device as some power dissipates in the cable.[5]
Talking about original PoE standard here. Looks like 13W is closer to the practical maximum.
--------
It looks like there are newer PoE+ and 4PPoE standards that increase the power.
> U6-LR
U6-LR requires the updated PoE+ standard which gives more power.
I also don't get people who use these for desktops or especially homelab servers. It's never made sense to me. But, I really like them for "computer that's a computer rather than a microcontroller, but closer to the Real World than a Real Computer."
These little boards usually offer GPIO, SPI, I2C, and sometimes even CAN natively, which makes them awesome for plugging into real world devices without needing another microcontroller hop in the middle. I've used boards in this 5-10W class for automotive reverse engineering, higher-order drone control like computer vision / flight planning (paired with a microcontroller for the hard-realtime control loop), and robotics. I think they're awesome for that use case.
Just a week ago, when I was editing the video, the Rock 5 wasn't yet listed on Armbian's website (now it is: https://www.armbian.com/rock-5b/).
Until very recently, the Rock 5 was only supported by more bleeding edge builds and I tend to stick to more stable images when I'm reviewing a board—especially the builds that come recommended by the vendor themselves.
Yeah, I was wondering about that. I haven’t checked the Armbian reference kernels, but they might fix some of your build issues.
But it was good to see you confirm one of my red flags about the board:
I have been looking for a good board to run https://github.com/pimox/pimox7 (I currently use my Pi 4 8GB for that, hosting a bunch of test LXC containers), and something faster that could boot off an NVME without hassle would be great. Looks like the Rock 5B isn’t it.
There's a whole section about it in the linked post…
> This Orange Pi 5 uses the same Rockchip RK3588—well, almost the same. The Orange Pi has the RK3588S, which has a little less bandwidth than the chip in the Rock 5. It has a lone M.2 slot on the bottom running at PCIe Gen 2 speeds, which means this SSD only gets about 400 MB/sec—just as slow as on a Pi. The Orange Pi 5 also only has 1 Gbps Ethernet, and not even a full 40-pin GPIO header, but it does have a full size microSD card slot, so that's an upgrade! But the Orange Pi 5 is a lot less expensive. You can get the same CPU and GPU performance, at least—and in my case it actually benchmarked a tiny bit faster—for half the price!
The article claim that you can't compare a $150 "mini PC" to a cheap Pi is unrealistic.
I got a lot of flack (elsewhere not here) for setting up a mini-cluster for K8S that used mini PCs that I claimed are cheaper than Pi. Note that I can actually buy a mini-PC for $150 whereas you can make up any imaginary MSRP you'd like for a Pi but you're paying $170 today on Amazon regardless if the unobtainable price is $5 or $99.
There are plenty of low performance PC compatibles from Beelink and similar places that cost less delivered tomorrow than a Pi delivered tomorrow AND have higher performance specs than a legacy Pi.