Let's say you've got something that needs to be computerised at a slightly higher level than an Arduino, with the computing part costing less than about $100-150, and ideally less than $50 (think Beaglebone, Odroid, PCEngine, Pi and clones, Pine, etc). It looks like the only choice is between ARM at the low end and x86 at the high end. Everything else has fallen by the wayside: The last MIPS-based product was the Ci20/Ci40 from 2015 and neither the hardware nor software have been updated since, PowerPC is out there but only as high-priced SBCs and good luck finding a distro that supports it, Sparc is left with Fujitsu working on it for mainframes, and RISC-V is still a glint in everyone's eye - the few SBCs based on it cost more than a low-end server, and despite various enthusiastic press releases I can't see any timeline where I can get a $50 RISC-V device that performs the same as a $50 ARM-based one. And then there's the software support, once you leave the x86 world you've got, outside of various specialised RTOSes, Linux. A very few systems have one or two of the BSDs, often in a hit-and-miss manner, but that's it.
Has Linux + ARM/x86 killed everything else?
Related Stories
Debian developer Jonathan Carter was recently given a MIPS64-based motherboard which he ran through its paces. The board has a Loongson processor which is intended for both general purpose and embedded processing.
The reason why I wanted this board is that I don't have access to any MIPS64 hardware whatsoever, and it can be really useful for getting Calamares to run properly on MIPS64 on Debian. Calamares itself builds fine on this platform, but calamares-settings-debian will only work on amd64 and i386 right now (where it will either install grub-efi or grub-pc depending in which mode you booted, otherwise it will crash during installation). I already have lots of plans for the Bullseye release cycle (and even for Calamares specifically), so I'm not sure if I'll get there but I'd like to get support for mips64 and arm64 into calamares-settings-debian for the bullseye release. I think it's mostly just a case of detecting the platforms properly and installing/configuring the right bootloaders. Hopefully it's that simple.
In the meantime, I decided to get to know this machine a bit better. I'm curious how it could be useful to me otherwise. All its expansion ports definitely seems interesting. First I plugged it into my power meter to check what power consumption looks like. According to this, it typically uses between 7.5W and 9W and about 8.5W on average.
The Loongson processors are developed at the Institute of Computing Technology (ICT) at the Chinese Academy of Sciences (CAS) in China in conjunction with the BLX IC Design Corporation, also in China.
Earlier on SN:
Is Low-Priced Computing Stuck With an ARM/x86 Duopoly? (2019)
MIPS CPU Architecture to Become Open Source Hardware in 2019 (2018)
Linux-Based, MIPS-Powered Russian All-in-One PC Launched (2016)
(Score: 2) by Gaaark on Tuesday February 19 2019, @12:41PM
The Penguin and Catwoman have killed them ALL!
--- Please remind me if I haven't been civil to you: I'm channeling MDC. ---Gaaark 2.0 ---
(Score: 3, Informative) by Anonymous Coward on Tuesday February 19 2019, @01:27PM (6 children)
The C-SKY architecture was recently added to the mainline kernel:
https://www.phoronix.com/scan.php?page=news_item&px=C-SKY-Approved-Last-Arch [phoronix.com]
Small dev board are very cheap:
https://www.cnx-software.com/2018/12/23/c-sky-linux-board-review-benchmark/ [cnx-software.com]
https://www.ebay.com/itm/PK-OrangePi-NanoPi-Raspberry-Pie-GX6605S-Development-Board-For-C-SKY-Linux-/192755718943 [ebay.com]
I suppose if you are looking for multiple-gigahertz boards, you are probably out of luck. But if simplely supporting the linux kernel is what you are after, there are options. I believe there are MIPS-based chips out of Russia, as well as the Loongson chips best known for the Lemote laptop RMS used to use.
(Score: 4, Interesting) by JoeMerchant on Tuesday February 19 2019, @03:55PM (1 child)
I was going to say, I don't think it's Linux that did the killing - more typical free market forces...
Low end x86 is too costly for some applications, but high end x86 is powerful enough for all but the most esoteric high power applications.
Low end ARM is cheap enough that anything cheaper would need to have production quantities in the millions for the price difference to matter, and high end ARM overlaps with low end x86.
Off the top end of the scale, Linux does have some support for esoteric supercomputing, but unless you need that, you're not going to pay to go there.
Off the bottom end of the scale, there's not enough power in the chips to warrant a Linux-like OS, so they're out there on their own with RTOS or less - but, again, production quantities in the tens of millions can support $500K in custom development work to shave $0.50 off the BOM.
🌻🌻 [google.com]
(Score: 0) by Anonymous Coward on Tuesday February 19 2019, @07:49PM
The x86 glue chips make for an expensive design.
You don't need them with ARM or PowerPC.
In many cases, you don't need a CPU/MCU if you use a FPGA, which will often have a hard or soft ARM core.
PowerPC is still dominant for telecom, and single-core performance. It is harder to write good multi-core code (even threaded code can be tough, as it often gets run on a single core because of memory sharing between the threads).
(Score: 2) by driverless on Wednesday February 20 2019, @04:32AM (3 children)
The C-Sky is kinda gutless, about the same as a cheap STM32 board, you're rather limited in terms of what you can do with it, the linked board has 64MB of RAM and 8MB of flash which is going to be painful if not impossible for self-hosted development. Loongson is also pretty hard to get hold of outside of China, and from what I've seen, outside of the Lemote, the ones available to outsiders are typically on multi-hundred-dollar server boards.
(Score: 0) by Anonymous Coward on Wednesday February 20 2019, @06:04AM (2 children)
That is pretty much dead on. The Russian one mentioned is the Balkal T1 and is also several hundred dollars.
https://www.cnx-software.com/2018/09/26/baikal-t1-last-mips-processor/ [cnx-software.com]
There are also the MediaTek MIPS router boards that are sometimes configured like an SBC, but again you are limited to OpenWRT or similar. Not really any consumer level boards outside of x86 and ARM that would be good candidates for self hosted development.
(Score: 2) by driverless on Wednesday February 20 2019, @07:12AM
The Baikal would be kinda cool, merely to say that you're running a Russian-sourced CPU that most people have never heard of. I looked at getting one via some Russian friends just for the coolness factor but they said you pretty much need to be a Russian speaker to work with it, in that all the tech info and support forums are Russian-only. Also, EUR500 is a lot to drop on something like that...
(Score: 2) by driverless on Wednesday February 20 2019, @07:14AM
Also the fact that the RTC is powered by Po210 instead of a battery was a bit of a concern...
(Score: 0) by Anonymous Coward on Tuesday February 19 2019, @02:15PM (2 children)
Pi Zero is €5, not $50. With SD card it's maybe €10.
Why would you want something else? Masochism?
(Score: 2, Insightful) by Anonymous Coward on Tuesday February 19 2019, @03:24PM
Pi Zero and Pi Zero W are unavailable from official distribution at the moment (as per normal), there's some abnormally cheap back stock for about $15-20 (before shipping) from Adafruit or similar.
Shipping to Australia would easily cost about double the board cost on a single unit. So in-country resellers get as many as they can then flog them for slightly under the shipped cost.
Shipped cost on a Zero has fluctuated between $40 and $100 for the entire life of the device in all its iterations. It was far worse in the beginning when there were supply issues, but it has literally never been possible to get a $10/€5 Zero. They're speciality devices that are always out of stock and in high demand. A regular Pi 3B+ costs a fair bit less.
(Score: 0) by Anonymous Coward on Tuesday February 19 2019, @06:12PM
How about binary blobs? That's why.
(Score: 5, Informative) by Rich on Tuesday February 19 2019, @02:21PM (4 children)
If "slightly higher level than an Arduino" is supposed to mean: "Computers for 30 bucks that come loaded to replace my desktop, where I can spend days comparing and deciding why I can't buy them": YES, the world is stuck on ARM. And if it's a duopoly, then only between Raspi and Chromebooks.
If "slightly higher level than an Arduino" means what the words say, there is a shitload of stuff: The ARC line of CPUs is sold by the billions, although there seems to be not a single stand-alone piece of silicon with it. According to one article, ARC is #2 behind ARM (or was at least for a while in the near past). They scale from little systems with Contiki (!) to 2 GHz-class pipelined multicore stuff running Linux. Easy to put into your solution if you make, say, a million of that.
If "over a magnitude more powerful than an Arduino" is enough, then we have the Tensilica Xtensa CPUs, which came to fame by means of the ESP8266 and ESP32. An ESP32 runs at 200 MHz and comes with 520K RAM. I think would be a neat exercise to put a full classic Mac emulation on one (having the service core do the video, ZX81-style).
There are still MIPS offerings left (I think the PIC32MZ was even available in PDIP-28 for the makers), and with MIPS recently panicking about RISC-V, that might get more interesting again. But if you like MIPS and would like it a bit more flexible, there's the Xilinx Microblaze, which is only available for FPGAs. It looks a lot like classic MIPS, and even runs Linux (although the Kernel was not so well maintained when I worked on one; you want to apply my signal handler patch which is not in mainline...).
But at the "slightly higher level than an Arduino", there is also an ARM offer which you can't refuse: the STM32F103C8T6, and that especially packaged on a "bluepill" board. The whole board is cheaper than the chip alone at a western distributor, so I can only assume that it is ARM's way of trying to keep a foot in the door at the low end. Today's spot market price on the 'bay: 1,77€ including shipping. Given the community that has gathered behind it, there's little reason to take anything else for ordinary tasks these days, except for having no experience with it and the cost for a one-off or a few controllers is less than that of your time to get it going.
(Score: 2) by driverless on Wednesday February 20 2019, @04:36AM (3 children)
There's a lot out there if you point at individual SoCs, but a lot less with complete systems (outside of the Arm/x86 duopoly). That's why I mentioned things like the Odroid, Pi, and Pine, you buy the hardware, drop whatever distro you want on it, and it's ready to go. With some of the things you mention, a certain amount of manufacturing is required before you can use them...
(Score: 2) by Rich on Wednesday February 20 2019, @02:57PM (2 children)
That's even the case with the RasPi or a mini-ITX, which usually have to be mounted in some case. Also, those little critters are hardly suitable for many standalone applications, so they will always be "manufactured" into being part of something larger, from heating regulation to home entertainment. How much solder flows depends. You might want to cleanly connect a Pine 64 for audio to a speaker output panel by soldering, whereas a quick and dirty tooling control might just be a PDIP chip or module wired up solderlessly on a breadboard.
What I consider sorely missing, ARM or not, is a cheap widget of any sort with a monitor. Kind of like the 50-bucks China phones. Beefy SoC, couple hundred Megs, or a Gig of RAM, 5-inch IPS screen, but instead of cameras and cell radio they'd have GPIO and mounting holes like the small SBCs and run a proper free Linux Stack. The RasPi screen is still way out of that "no need to think about whether to buy a few" price territory, and especially given that price really badly underperforming. 4-inch-DSI touch screens start at a bit over a tenner in China, that'd need a somewhat coherent offering together with one of the little SBCs.
By the way, I retract my comment that Intel is completely out: I wasn't aware of the prices Asrock sells their all-integrated mini-ITX boards for at the moment. They match or beat the RK3399 offerings. A great deal if you don't need reliable realtime. Loss (?) leader for Intel, I guess. To quote Steve Jobs: "Cannibalize yourself, or others will." :)
Also, I'd like to add that Synopsis have really nasty closed information policies about ARC, despite raving how great open-source is on their gcc pages, but then referring to their NDA-only stuff from otherwise empty gcc documentation. Yeah, right, it's great as long as they get an optimizing compiler for free, without anything in return. They need some serious butt-kicking from the RISC-V scene.
Finally (but ARM again), I saw that Microchip/Atmel has a "SAM4S" series (120 MHz, 64K SRAM, 128K Flash) on offer at about half what the less powerful STM32 i mentioned costs when bought as a chip in the west.
(Score: 2) by driverless on Thursday February 21 2019, @02:22AM (1 child)
In my case I really want something that's self-hosted (which I should have mentioned in the OP). I spend way too much work time interacting with systems via ICEs and similar, so in my free time I want something I can just ssh into and make any needed changes that way. Admittedly the 5V Arm-based systems are nice that way, you drop them in wherever you need them, run 12 or 24V out to them, and locally drop whatever the voltage is by the time it gets there to the required 5V via a UBEC. The WiFi support has also slowly, painfully got better over time, in that you can mostly rely on it remaining network-connected between software updates/upgrades, although running wired ethernet to a WiFi bridge is the way to go for guaranteed reliability.
Still wish there was something other than Arm/Intel to work with though...
(Score: 2) by Rich on Friday February 22 2019, @12:15PM
What about MicroPython on the ESP32? Checks all the boxes, not? Oddball CPU and REPL shell over WiFi should be doable (funnily enough, ssh shell seemed to be a challenge for them, but web REPL over https worked, as I read one of the threads). If the language has to be a bit oddball, too, there's LUA RTOS which comes with the dropbear ssh server. Board costs a fiver from China, tenner local. A bit more downscale, it can run the Arduino libs (with OTA access from the IDE), and bit more upscale, if it's got to be really UNIX-like, ESP32 will also run NuttX, but they seem to have an issue with getting wifi firmware they can use.
(Score: 4, Insightful) by DannyB on Tuesday February 19 2019, @03:28PM (11 children)
If it is going to be a duopoly, it should be RISC-V and ARM.
And those two should not just be confined to the low-priced computing space.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 2) by JoeMerchant on Tuesday February 19 2019, @03:58PM (10 children)
x86 is job security for too many people to ever be let to die.
I still remember the day that the 80286 addressing schema was first explained to me: such obvious complexity for the sake of complexity (and vendor lock-in...)
🌻🌻 [google.com]
(Score: 5, Interesting) by DannyB on Tuesday February 19 2019, @07:48PM (5 children)
The death of x86 / 64 would have a secondary benefit to the entire world.
The end of Windows.
Bare
bearwith me. The primary value of Windows OS is that it runs legacy apps which are not easily ported to another OS or processor architecture. The ones that can be ported, even if not easily, already have been. That is, they already run on Windows on ARM, or on non-Windows.If x86 goes away, then the entire forty year old blight goes away also. Drive C? Kids today probably don't even know why there is a drive c. Segment registers? The ability to boot DOS 1.0?
Left to compete in software without the Windows monopoly? Microsoft wouldn't know how to compete its way out of a paper bag. Despite pretending to embrace open source.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 3, Interesting) by JoeMerchant on Tuesday February 19 2019, @09:00PM (4 children)
I think you missed it:
is the vendor/customer relationship network. Literally millions of people keep their jobs because they "know Windows" in some capacity, even if it is something as non-value-add as how to deal with their licensing keys. The majority of people continue to train to Windows because the majority of paying jobs available in the market are for people who "know Windows."
You might hope to slowly turn this iceberg to a new course, you might just wait for it to melt, you can even try to blow it up, but reality is: it will take time, lots and lots of time. Putting the squeeze on x86 might be like lobbying for continued use of fossil fuels, to raise the mean global temperature, to try to melt the berg a little faster. Just watch out for the unintended/unwanted side effects.
🌻🌻 [google.com]
(Score: 2) by DannyB on Tuesday February 19 2019, @09:14PM (1 child)
If x86 doesn't go away, it could just shrink in number and get more expensive because it isn't efficient or cost effective. It merely is part of what supports the legacy Windows jobs you speak of. That seems to be a self reinforcing death spiral.
It's like saying the horse and buggy won't go away because it is so entrenched compared to the automobile.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 3, Insightful) by JoeMerchant on Tuesday February 19 2019, @10:50PM
There is definitely a self reinforcing death spiral (and it's not limited to computer OSs...) Unfortunately, the difference in efficiency between ARM and x86 isn't even an order of magnitude, whereas horses and buggies were several orders of magnitude less efficient/capable than automobiles.
🌻🌻 [google.com]
(Score: 3, Touché) by DannyB on Tuesday February 19 2019, @09:17PM (1 child)
How much training do you need to support windows?
1. Did you try restarting?
2. How about power cycling?
3. Re-install the OS.
4. The best advice I can offer is to buy a new PC.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 2) by JoeMerchant on Tuesday February 19 2019, @10:54PM
Exactly! That Linux stuff is HARD by comparison, there are actual answers out there for most things if you dig for them, and if you can't find an answer it's always possible to dive into the source code and figure it out - unlike Windows where: once you've called tech support and gotten an official shrug from the vendor, you're done: case closed, what the boss is asking for just isn't possible.
Are you trying to say that all those people trained to your 4 step solution to all the problems in the computing universe could actually be retrained to support an OS like Ubuntu? Not likely.
🌻🌻 [google.com]
(Score: 2) by driverless on Wednesday February 20 2019, @04:44AM (3 children)
x86 also just works. It's not surprising that it's the platform that OSS OSes support first and foremost, it's got a (mostly) standard init and boot process and, barring oddball hardware settings, you can generally just drop a generic distro on an x86 system and it'll work, even if it doesn't necessarily use every exotic hardware feature out of the box. OTOH even with something as widespread as Arm you need custom builds for almost every single device and hardware config. An extreme example of this is the PI 3B+, which despite it's name was incompatible with the Pi 3B until they patched the kernel to handle it.
So x86 is never going to die, because "it just works" gets you a lot of support in the industry. I'm no great fan of x86 either, but it is annoying that if you need to run X, where X is some arbitrary OS or application, it's got a far greater chance of working on x86 than on anything else. Which was part of the motivation for my original post, why am I always pushed back onto x86 unless I want to spend a week with a kernel debugger trying to figure out why FooOS 12.7 hangs on boot on aarch64 system XYZ?
(Score: 2) by JoeMerchant on Wednesday February 20 2019, @02:58PM (2 children)
Perhaps not until the rise of the machines - when AI takes control they'll have no use for backwards compatibility with all the legacy crud, and they'll probably also want to "confuse the enemy (us)" with something like a simplified RISC architecture that just looks like patternless binary noise to meatbags...
🌻🌻 [google.com]
(Score: 2) by driverless on Thursday February 21 2019, @02:10AM (1 child)
Then we'll be back to 6502's [apl2bits.net].
(Score: 2) by JoeMerchant on Thursday February 21 2019, @12:45PM
Fine by me... around about 1983 I was of the opinion that you could do just about anything that needed doing on a computer with a 6502 based system. They needed help with the graphics, and sound, and would also need a network co-processor to handle TCP/IP - ethernet today, but with those I could easily enough be posting this message to Soylent News from a 6502 based PC.
🌻🌻 [google.com]
(Score: 1, Interesting) by Anonymous Coward on Tuesday February 19 2019, @03:47PM
Renesas has a bunch of CPU architectures which aren't x86 or ARM. And there are some dev/eval boards available with those chips. Can't say how good they are, but if you're just looking for something different then check it out.
(Score: 2) by bobthecimmerian on Tuesday February 19 2019, @04:08PM (1 child)
PowerPC and POWER are the same thing, right? Or do I have my terminology wrong? There are videos and demos out there of people running Debian, Ubuntu, and OpenSUSE on the Raptor Talos 2 PowerPC workstation with full desktop, web browser, office suite, etc... etc... so it's not cheap but it works fine. If I was childless I'd have one, but kids are expensive.
(Score: 3, Informative) by DannyB on Tuesday February 19 2019, @07:51PM
I'm getting old, but if memory serves from my 1990s days as a classic Mac developer, Power PC is mostly a subset of POWER. But I would be happy to be corrected if this is wrong.
This [wikipedia.org] might actually explain it better.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 4, Interesting) by mr_mischief on Tuesday February 19 2019, @05:47PM (2 children)
https://www.sifive.com/boards/hifive1 [sifive.com] is $59. If that's more than your low-end servers then please point me to the truck from your servers fall off.
(Score: 0) by Anonymous Coward on Tuesday February 19 2019, @06:33PM
I was pretty exited about this find (mentioned on HaD):
https://www.seeedstudio.com/sipeed [seeedstudio.com]
(Score: 2) by driverless on Wednesday February 20 2019, @05:06AM
That's an expensive Arduino-equivalent. I was looking for something "at a slightly higher level than an Arduino" as per the OP, think a self-hosting dev system for Linux/*BSD/whatever.
(Score: 1, Insightful) by Anonymous Coward on Tuesday February 19 2019, @07:08PM
"Has Linux + ARM/x86 killed everything else?"
like linux + x86 killed risc servers in the 90s/early 00s.
computing has reached a local maxima: x86/arm - c/c++ - linux/windows
celebrate or cry?
(Score: 0) by Anonymous Coward on Tuesday February 19 2019, @08:24PM
ARM is not a chip, it is an instruction set. ARM the company controls licensing of it, but you can get microcontrollers with an ARM chipset for literally pennies in volume from numerous sources.
If you're looking for a SoC, ARM is the best choice going right now because of its low transistor count and high compute power to watt ratio and that isn't likely to change because there's only so many ways to skin that cat.
There are other instruction sets too, and if you want MIPs, there are cores you can license and have put into anything you like
https://www.mips.com/products/ [mips.com]
Right now MIPs is trying to target IoT applications though.