Why is it that most Single-board computers are based off the ARM chipset instead of x86-x64?

Many embedded processors are ARM processors. They are better suited for embedded stuff. One of the things is that they get more computing work done per unit of energy consumed. Secondly, there are many "system on a chip" (SOC for short) ARM processors. Like the BCM2735. That integrates lots of peripherals into the chip so that you don't need extra chips. So take a look at the raspberry pi zero... You have a single board computer there with HDMI output, a storage interface and half a gigabyte of RAM. All in that single black blob in the middle!

Intel has been trying to get "in on the action" for over a decade now. They haven't been really successful, because their processors require too much support hardware around them, meaning the boards get big and expensive. So for example: https://azerty.nl/0-843-709774/intel-next-unit-of-computing-board-de3815tybe-moederbord-ucff-intel-atom-e3815-usb-3-0-gi… which doesn't even have RAM but is 4x more expensive than a raspberry pi. And it requires a bigger powersupply. (in my quick search for something like this, I found a EUR 73 board, but that was just the breakout for another intel board....)

The cheapest Intel CPUs come in at around EUR 40. extreme low-end ones maybe half that.... An ARM CPU can be bought for 32 cents! Ok, that's a low-end ARM V6...... hmm... just like the raspberry pi (1). OK, that one doesn't have everything the raspberry pi CPU has, but it can be bought for way less than $1 in low quantities. At higher quantities the raspberry pi CPUs apparently also cost only $1-$2. The RAM is probably similar, and they sell at $5 (for the zero).

Hello Matthew,

I will add to to the very clear explanation of rew about the hardware also the software side yo mention/

I think it would be a lot easier if they were x86-x64 based, because it would be easier to use programs

Sorry but I think it is not so true. As embedded devices has a totally different philosophy for both the architecture and usage you should not think to program them in the same way you can program a Window desktop.

Despite this, you should consider that the same programming way available on Intel platforms is almost available on ARM based embedded devices. Think to Python, just to refer to one of the most popular and easy to use languages. What is characterising languages like this and too C and C++ are their high portability as well as development frameworks also when you work on totally different hardware architectures like desktop Intel processors boards and embedded ARM processors.

Enrico

I think this sums up the differences best Intel x86 Microprocessors vs. ARM Microprocessors . This part shows best the reasons:

Licensing makes it possible for a company to develop a custom solution for their hardware, which in turn results in better power efficiency and lower costs. Because Intel does not license x86, anyone who wants a x86 processor must buy a full processor from Intel (or AMD or VIA). ARM processors can be custom-tailored to a specific product.

This is why ARM Holdings, the company that licenses the ARM instruction set, often boasts that ARM processors are more popular than x86. They are, by unit volume, correct. There are far more ARM processors put into products because ARM processors can be custom-tailored to fit so many different roles, embedded into places where we don't even think of them, such as automotive technology.

Clem

My two answers: cost and heat.  ARM processors are a lot cheaper than x86 processors.  ARM's business model is to license the ARM architecture and/or implementations rather than make the chips themselves.  Licencees like TI, NXP/Freescale, Broadcom, Qualcomm, Allwinner, and many others compete with each other to sell the cheapest chips.  ARM's license fees are modest which keeps the chips cheap.

ARM SoCs are designed for very high volume mobile products.  High volume drives the prices down.

From the beginning ARM pushed low power.  Intel has pushed processor performance by fitting as many fast transistors as possible onto a piece of silicon.  Power consumption has always been secondary to speed.  Low power is why ARM is the best choice for mobile devices, where you want the battery to last as long as possible.

Most single-board computers are very small.  If you try to pack high power chips on a small board you end up with heat issues (like poor reliability) and need to add cooling devices.

A lot of people who do embedded programming already use GNU/Linux and other open-source software on their development PCs.  When you have source code, targeting ARM instead of x86 is no big deal.

JMO/YMMV

Hi Matthew,

A little bit of history can answer your question.

When the MCU market exploded, 8 bit computers became the de facto standard for embedded systems.

In the 1970's there were many SBC kits for engineers to learn about embedded systems and to explore potential applications.

Through the 1990's most embedded applications used 4 bit and 8 bit processors.  Intel was the major player in this world as well as the PC world, so while they had some support for the 8086 architecture, they had little reason to add specific embedded capability to the product line.

All of that changed after 2000 with the surge in smart phones, devices and consumer electronics.

Manufacturers needed to pack more capability into their chips than just the processor and some memory.

ARM came into that environment with an easy to use System On A Chip (SOC) architecture that they could easily put standard display and IO components plus enough memory to shrink the chip count dramatically.

Initially, these applications were small runs compared to the PC level devices, so Intel just did not see a good market to invest the 8086 architecture for those types of applications.

Fast forward to the present and Intel now sees that they missed the leading edge of an entire business area and are now playing catch up.

There is no architecture advantages inherent with the 8086 architecture to move it into this product space.

So here we sit today with a few companies advancing along the ARM architecture and some others.

At this point, the level of support dictates the cheapest path, which is the ARM environment, especially with the open source support and freeware available.

For Intel to compete, they need a large investment and time, both of which the competition now has a huge advantage.

So I would not look for an 8086 type MCU in the future.

I would expect a leap in technology out of Intel so that they can run past the competition rather than play catch up.

DAB

Single board computers have been developed ever since it became feasible back in the seventies. The Apple 1 was an SBC based on a 6502, as were many others. Every time  a new cost-effective processor is introduced it spawns a whole raft of SBCs and even new companies that are not married to older technology.

Because it cost less to buy an ARM license than develop your own technology, many companies have taken this path. But it has resulted in an extremely competitive arena where everyone has the same technology and it seems the main differentiator is price. It is tough to differentiate with software since both android and Linux are essentially free.

Another big driver is that ARM chips use RISC architecture, which has fewer transistors and uses less power, which is very important in mobile computers.

So the advantages of ARM in cost and power make it attractive to the mobile market, but for a product development business it is also a low margin proposition due to extreme competition. You can't even stray too far from standard offerings, because as soon as you add something unique, it becomes very expensive to support.

The other factor that must be considered is computing power where CISC has had a traditional advantage, and still does. This is always desirable and until mobile computing became feasible it trumped low-power and low cost, which are also always desirable.

As ARM chips increase in computing power with more sophisticated pipelining, wider data paths and larger instruction sets they will also become bigger, more power hungry and more expensive.

On the CISC front, chip designers are responding to the burgeoning mobile computing market with lower power, lower cost chips.

I hope both trends continue as it is good for all of us to have competition driving things forward. Currently there may be more ARM SBCs appearing on the market, but there are still lots of others showing up as new processors are developed, and of course there are still a large number of mature (non-ARM) SBCs on the market.

The current trend will continue for a while because so many companies are commited to ARM at the moment, but it is hard to predict what might happen, even in the near future. For example maybe augmented reality glasses become a huge advantage for users with many must-have applications, but a new type of processor is needed to eliminated display lag. Such a processor could handle all other mobile computing requirements without raising a sweat, so it could just take over the market - for a while.

Ironically, Intel's acqisition of Altera (who I believe has ARM core capability) might just make Altera ARM implementations the fastest, lowest power ARM chips on the market, because Intel's fine-geometry IC technology is at the leading edge for speed and power consumption. I'm not quite sure ARM technolgy is automatically scalable to Intel's IC process or if ARM themselves need to design for the finer geometry process.

To Robert Wolff & the others:

I have any old RaspPi Mod 1B which I mounted on an Old USB 1.1  powered Hub (b/c this was circa 2013). I use it in conjunction w/ MOOC courses. A Python Course at coursera.edu and a H/W course at udemy.com. A whole different world from my x86 Toys. I bought a Intel NUC5i7RYH for about $500 (barebones) this in contrast to the $35 RaspPi. I tricked-out the NUC w/ 16GB Ram and an M.2 256GB SSD (which died) now running on a 2.5" 480GB SSD. Win10 comes up in 11 seconds. Different Worlds/Functions, RaspPi for dabbing in H/W. x86/windows for High-powered Stuff. Although Intel is trying to get into the SOC market(?). isn't it Microsoft that is making a Push (on the s/w side) with their Win10-IOT (Not full-fledged Win10)?

BTW Thx to all comment-ers on bring me down Memory Lane.

USCDADNYC (NY NY)

Douglas Wong wrote:

 

Ironically, Intel's acquisition of Altera (who I believe has ARM core capability) might just make Altera ARM implementations the fastest, lowest power ARM chips on the market...

Intel has had an ARM architectural license ever since they bought the StrongARM technology from Digital Equipment Corporation in 1998.  Intel tried to sell communications chips based on an ARM version called XScale but I don't think it got anywhere.  Intel sold most of their ARM technology to Marvell in 2006.

For more detail, see XScale at Wikipedia.

Enrico,

What you, "think" is correct, actually isn't. I am hardware and software engineer that works on desktop, servers, super computers and embedded systems. The only thing that is different on the embedded ARM series CPUs versus x86 / x86_64 that you need to be concerned with is the different instruction set and registers. Since a good C/C++/C# cross compiler from GNU or LLVM, there's not too much trouble creating drivers, operating system images, support libaries and ultimately the applications sitting on top of all of this. The only real difference from a programmer's point of view (meaning not doing anything in assembly language) from C/C++/C# is setting the compiler flags to optimize for size vs. speed. The ARM series of CPUs and APUs have two byte ordering modes, big and little endian. Every little SBC I've seen have been hardwired for little endian mode, just like the x86 and x86_64 are (which they don't have the ability to switch back and forth). The advanced CPU and APU architecture of the SBC allows for some really cool capabilities with very little memory. The vast majority of why PCs are slow slow are all the hardware and software interrupts. The embedded ARM solutions only have as many interrupts as they need absolutely which is far less than the PC. Programming these SBCs are very similar to the desktop except you don't have to guess the hardware or try to search for it, you already know what is there, the drivers very rarely need to be changed unless their buggy or seen to be dangerous in regards to security of personal information.

From an operating systems standpoint already running and you have a keyboard with a mouse connnected. Running the Python or Perl scripting languages run just the same as on the desktops or servers, let alone on the embedded system. The reason why the operating system images on the desktop and servers are so big is because you have all these different configurations with different chipsets for LAN, audio cards / chips, graphics cards / chip, different CPUs with slightly different instruction sets. Of course, the applications are built for compatibility or absolute speed (this takes up a lot of space on the disk and in RAM).

Most embedded systems are capable of OpenGL 4.x but many focus on OpenGL ES version 1.0 ~3.2, which is a more stripped down version of OpenGL core profile, as the core profile is a stripped down version of the compatibility profile (full specification).

To be honest with you, the ARM CPU and APUs that have 8 or 10 core in embedded systems are actually a lot more powerful than the 8 and 10 cores on the Intel and AMD side, even at a lower frequency and use a lot less power. The Qualcomm Snapdragon 600 embedded and  800 mobile series are quite remarkable acheivements. My favorite is the Snapdragon 835. I also use the ARM Cortex A-73 in embedded systems and a design as a desktop with slots I can plug accessories I'm developing into the system. The desktop computer I've developed uses standard PCIe 3.0 x16 slots, of course I have other chips handling that I/O and interface logic but the embedded system chips can be used just the same as standard desktop chips are used.