OSDev.org

lot of smaller companies appear to be doing embedded work on ARM architectures rather than x86 due to its low cost, SoC architecture, low power requirement. What is the main barrier if some entity wants to break entry into x86 market with the assumption of small to moderate financial backup?
Here is what I think is the lowdowns regarding the challenges:
- biggest challenge being Intel MRC code. Intel wont talk or give their MRC to small guys (I actually witnessed it b4) unless partnered with and require NDA. Also MRC code is extremely difficult to debug, maintain due to its memory technology consistency and one needs to be in constant touch with Intel engineers.
- Same goes for some of the Intel x86 chipsets and numerous other related technologies, and specification whitepapers etc that only partner has access to. By doing so, I see Intel has a stranglehold on x86 market dominating supposedly 90% of market share.
- There appears to be some x86 dev kit available online but i am sure mostly confined to IoT, client applications.
- Secondly the BIOS needed with all the code modules/drivers that supports all peripheral components on the motherboard. With all the related specs being openly public and accessible (ACPI, SMM, SMBIOS, i2c, plus Intel publishes some of its technology whitepapers publicly, I think this is doable provided MRC mentioned above has been resolved)
- I heard about seabios which is a open-source version of BIOS and downloaded and inspected it, however looking at their source code, it looks a very minimal code. I am not sure how successful any entity has used it and came up with functioning board.

You seem to be very intent on forcing an issue here.

- Look at your own list. Why would you even bother to choose x86 when its inferior in almost all respects? Unless you want to make a PC clone and you have to support the instruction set, there's no reason to even go there.
- Intel has less market share than you claim
- If you have external RAM, you need to configure the CPU and all the peripherals in between to deal with it properly. This can be a bunch of hardcoded values when the RAM is fixed by design, or more complicated if it's user-exchangeable. In addition, the presence of such code has basically been a requirement for all architectures since caches were added.
- So Intel has a reference implementation to do memory controller initialisation for their chips they call "MRC". What is the real difference between theirs and something like that from any other manufacturer that might not even happen to have a marketing name?

Combuster wrote:You seem to be very intent on forcing an issue here.

- Look at your own list. Why would you even bother to choose x86 when its inferior in almost all respects? Unless you want to make a PC clone and you have to support the instruction set, there's no reason to even go there.
- Intel has less market share than you claim
- If you have external RAM, you need to configure the CPU and all the peripherals in between to deal with it properly. This can be a bunch of hardcoded values when the RAM is fixed by design, or more complicated if it's user-exchangeable. In addition, the presence of such code has basically been a requirement for all architectures since caches were added.
- So Intel has a reference implementation to do memory controller initialisation for their chips they call "MRC". What is the real difference between theirs and something like that from any other manufacturer that might not even happen to have a marketing name?

This topic's title itself says to be about a far fetched question, so it all shouldn't matter much.


I really wonder why you say that the x86 is inferior. They are the ones that started standardizing computing at home and in general. The common machine environment would be much more incoherent without the x86 PC. Just look at how fast they introduced advanced peripherals from the first models, more versatile than the rest. It hast most of the good code in existence.



That's why it's so difficult to force migrating to ARM and that's also why everyone ends up wanting x86 emulators in ARM machines, like Android ones. Without those, their environments are much poorer.

ARM CPUs could be the equivalent to BSD. It's good and used, but it's not so popular. Almost always targets are for x86 or x64, and when not, you see how ugly code repositories are, and how fast they disintegrate and either die or are reimplemented in a very confusing way.

The x86 has proven to contain all of the capabilities that are needed by Humanity to solve any existing problem, so chances are that it will never cease existing. It already has its standard base, so it's free to keep defining more advanced standard function. Otherwise, without good standards, computing looks more and more like a joke science if we compare it to Mathematics, Physics, Chemistry... which always choose the best methods instead of reimplementing stuff several times per brand, which is all we would have with an ARM-only world.

ARMs are good, probably excellent at implementing even chipsets or part of them; but for some reason the machines they run in working as the CPU core are so varied that you always need a pre-built software environment, and will almost always. It's like directly touching the nonstandardizable wires of countless microcircuit designs to configure them to make them execute applications.

~ wrote:The x86 has proven to contain all of the capabilities that are needed by Humanity to solve any existing problem

It has disproved it's ability to support mobile devices, for example. And there are more things that it is incapable of running.

"The x86 has proven to contain all of the capabilities that are needed by Humanity to solve any existing problem, so chances are that it will never cease existing"

what about quantum computing?

The x86 has proven to contain all of the capabilities that are needed by Humanity to solve any existing problem

I wish it would solve the Middle East problem.

embryo2 wrote:

~ wrote:The x86 has proven to contain all of the capabilities that are needed by Humanity to solve any existing problem

It has disproved it's ability to support mobile devices, for example. And there are more things that it is incapable of running.

How so?

I am seeing an increase of ultra mobile x86 CPUs and ultra mobile PCs.

I have seen machines like the IPC002 which is a Quad Core x86 1.8 GHz, 2GB RAM, 32 GB SSD, with Windows 8 (and thus could run Linux). It's a pocket-sized x86 PC computer.

There's also the older BenQ S6, Intel Atom x86 at 800 MHz (can supposedly be overclocked to around 1 GHz), 512 MB RAM, 2 GB 1.8 disk, can use 16 GB MicroSD cards (doesn't seem to be able to boot from them but probably can use 32, 64 or 128-Gigabyte cards), has 1 mini USB port, has speaker, touchscreen, headphones, WiFi. Can run Windows XP, DOS, Linux (Windows 98 seems to fail, sadly, but could probably be installed somehow), has AMI BIOS. Only uses around 5 Watts constantly. Can run several games from the Windows 98 and Windows XP era.

I'm currently running my web server from one of these.

Maybe phone manufacturers don't want "unlicensed" general developers with powerful and capable radios on their hands that can communicate and automate phone communications at all levels.

I don't support that fact at all but it's how it currently is.

It's too suspicious to see no x86 phones to think that it's because of a technical reason. It's more for reasons to secure a closed environment (phone telecommunications).

It doesn't make x86 inferior. It's because of something that has nothing to do with programming capabilities themselves. They have made sure to keep the PC ideology locked out of the phone world, so I don't think they will allow it.

They must feel like current top GPU brand owners in front of some standards body trying to open their "environment" and their "protocols" to the wild.

~ wrote:How so?

...

There's also the older BenQ S6

Yes, the thing is 15 centimeters long and 22 millimeters thick. It's really mobile. But may be you can point me to a smartwatch with x86 processor?

embryo2 wrote:

~ wrote:How so?

...

There's also the older BenQ S6

Yes, the thing is 15 centimeters long and 22 millimeters thick. It's really mobile. But may be you can point me to a smartwatch with x86 processor?

They could make a better effort.

It's 4.8". It's a bit bigger than a Palm Tungsten T5 or Palm LifeDrive which are ARM. They are big mostly for the screen and user interface.

They could use a Flash as a RAM. It's even cheaper than RAM. They could make it replaceable (using a MicroSD instead of a soldered DIMM module for example). Look at how other devices like the Palm LifeDrive even use a MicroDrive hard disk as part of the RAM.

Instead of using a BIOS they could boot directly to the full fledged device's system code, immediately into 32-bit or 64-bit mode. It's already booted and cached into a Flash RAM module so it only needs to be resumed.

They could effectively turn off completely the x86 CPU part of the device every time and only leave a helper circuitry capable of waking up on user request or on network activity (the network here would mainly be actually the embedded phone radio itself).

~ wrote:There's also the older BenQ S6, Intel Atom x86 at 800 MHz (can supposedly be overclocked to around 1 GHz), 512 MB RAM, 2 GB 1.8 disk, can use 16 GB MicroSD cards (doesn't seem to be able to boot from them but probably can use 32, 64 or 128-Gigabyte cards), has 1 mini USB port, has speaker, touchscreen, headphones, WiFi. Can run Windows XP, DOS, Linux (Windows 98 seems to fail, sadly, but could probably be installed somehow), has AMI BIOS. Only uses around 5 Watts constantly. Can run several games from the Windows 98 and Windows XP era.

This device can only run for 1-2 hours on battery (and it's too bulky, it can hardly be called pocket-size nowadays, I think that embryo2 included some sarcasm when he said "It's really mobile."). It's not even funny, it's just ridiculous. 5W in idle mode can't be called 'only', it is enormous for mobile devices.

~ wrote:They could use a Flash as a RAM. It's even cheaper than RAM.

You should study the subject deeply. Serial NAND flash (that is used in micro-SD and other solid-state storage) isn't random-accessible and in most tasks is awfully slow, as it provides good bulk performance but huge latency for small requests (and very huge latency for write ones). Parallel NOR flash is random-accessible but it's also slow (compared to RAM) and it's not cheap at all.

~ wrote:Look at how other devices like the Palm LifeDrive even use a MicroDrive hard disk as part of the RAM.

It cannot be called "part of the RAM", it's plain old swapping and it's slow and dumb way of doing things.

~ wrote:I'm currently running my web server from one of these.

Oh, I've tried accessing it yesterday and was really wondering why is it so sloooow. Ok, now I got the idea that you are of that kind of people who prefer doing things in a difficult way because "it's very hard but still possible, so let's do it in a usual way instead of a right one".

Nable wrote:

~ wrote:There's also the older BenQ S6, Intel Atom x86 at 800 MHz (can supposedly be overclocked to around 1 GHz), 512 MB RAM, 2 GB 1.8 disk, can use 16 GB MicroSD cards (doesn't seem to be able to boot from them but probably can use 32, 64 or 128-Gigabyte cards), has 1 mini USB port, has speaker, touchscreen, headphones, WiFi. Can run Windows XP, DOS, Linux (Windows 98 seems to fail, sadly, but could probably be installed somehow), has AMI BIOS. Only uses around 5 Watts constantly. Can run several games from the Windows 98 and Windows XP era.

This device can only run for 1-2 hours on battery (and it's too bulky, it can hardly be called pocket-size nowadays, I think that embryo2 included some sarcasm when he said "It's really mobile."). It's not even funny, it's just ridiculous. 5W in idle mode can't be called 'only', it is enormous for mobile devices.

~ wrote:They could use a Flash as a RAM. It's even cheaper than RAM.

You should study the subject deeply. Serial NAND flash (that is used in micro-SD and other solid-state storage) isn't random-accessible and in most tasks is awfully slow, as it provides good bulk performance but huge latency for small requests (and very huge latency for write ones). Parallel NOR flash is random-accessible but it's also slow (compared to RAM) and it's not cheap at all.

~ wrote:Look at how other devices like the Palm LifeDrive even use a MicroDrive hard disk as part of the RAM.

It cannot be called "part of the RAM", it's plain old swapping and it's slow and dumb way of doing things.

~ wrote:I'm currently running my web server from one of these.

Oh, I've tried accessing it yesterday and was really wondering why is it so sloooow. Ok, now I got the idea that you are of that kind of people who prefer doing things in a difficult way because "it's very hard but still possible, so let's do it in a usual way instead of a right one".

My website is slow because of my forum, which contains HTML templates with highly unoptimized PHP code. If I remove it, the website is very smooth, but I want to fix my forum because it's a good automated, indexed content aggregator to point to other resources.

About the RAM, the Palm LifeDrive works well for a 500 MHz machine and for using the disk as RAM to arrange the whole system as a preloaded ROM image.

An x86 phone could also use a MicroSD as a ROM to read the system into normal RAM, which could be restored easily even if the CPU is shut down and reset every time the user turns off the screen to wait for calls.

It could be used for many tasks that aren't critical or for running programs in emergency situations where the main RAM is full.

I think that it would run greatly for an x86 phone, specially if they use quad core CPUs of around 2 GHz. Then it could easily run the things people always want like PC and DOS emulators with increasing ease through iterations of phones, memory controllers/modules and CPUs.

embryo2 wrote:

~ wrote:The x86 has proven to contain all of the capabilities that are needed by Humanity to solve any existing problem

It has disproved it's ability to support mobile devices, for example. And there are more things that it is incapable of running.

ARM is low power and not as powerful so I think it has got upper hand in mobile arena. Intel is pushing some (giving away free) low-power CPUs but from what i heard, they are not succeeding well. I do remember several quarters back Intel committed to ~4? billion $ loss for giving out their low-power mobile CPU free or at a loss to seize certain portion of mobile market (obviously other profitable segments made up the loss anyways).

x86 instead is more dominant in areas where powerfull processing is needed. The higher Intel CPUs draw about 150W power at peak usage whereas mobile ARM is much lower than it. There are some initiative by several big names to make ARM entry into data server market with modular servers, but so far no significant presence is felt.

ARM has already caught up quite well in performance, mind you.

I started this thread as I was vaguely interested in hearing about opinions regarding hypothetical scenario where someone seemingly enough moderate financing would enter x86 market by offering and see what the entity will encounter however the topic appears to take entirely different direction.

ggodw000 wrote:Intel is pushing some (giving away free) low-power CPUs but from what i heard, they are not succeeding well. I do remember several quarters back Intel committed to ~4? billion $ loss for giving out their low-power mobile CPU free or at a loss to seize certain portion of mobile market (obviously other profitable segments made up the loss anyways).

Intel now phases out Atom design and concentrates it's efforts on clouds. No Intel based smartwatch for at least nearest 10 years.

ggodw000 wrote:x86 instead is more dominant in areas where powerfull processing is needed. The higher Intel CPUs draw about 150W power at peak usage whereas mobile ARM is much lower than it. There are some initiative by several big names to make ARM entry into data server market with modular servers, but so far no significant presence is felt.

I suppose it's the monopoly which has played the bad trick on Intel. Low power processing market is highly competitive and Intel is just unable to catch the required speed. So, it concentrates on the areas where it still has monopolistic position (or something close to it, because AMD looks pale as Intel's only competitor).