[product-request] Please design a general-purpose ARM PC

Alexey Eromenko al4321 at gmail.com
Sun Jan 6 10:21:28 EST 2013


Hello,

I would like to request to design something advanced: a
general-purpose ARM PC, to become a
x86 PC desktop & laptop replacement.

Target application:
The system must be powerful enough to run a full operating system,
like Linux-Debian with KDE or Ubuntu with KDE, including the most
popular Linux applications (Google Chromium browser, LibreOffice, GIMP image
editor, KDE Kaffeine media player, etc...), plus video in HD format
(MPEG4-AVC/H.264 and WebM/VP8 codecs).

Today an Intel Atom 1.6 GHz + 1 GB of RAM can do most of this job.
I think that your ARM SoC + 1 GB of RAM may also handle this job.

Why Linux?
==========
The world-wide Linux community is a big market -- there are over
20,000,000+ desktop Linux PCs (currently running on Intel x86).
In addition there are many knowledgeable people, who will help you,
and many will tolerate some of the early product deficiencies, giving
you early buyers/early-adopters.
Android is not good for a desktop or laptop use, because the GUI is
not designed for keyboard + mouse, but only for touch-screen.
Linux with KDE looks like a real Windows system, so having drivers for
Linux (and Linux itself) is very important.

Moreover, the Linux desktop PC can be sold inside China and the
western world for real work for businesses and education schools, as
well as home users, unlike Android O.S, which is only used for content
consumption, not for creation. Android is for
people, who like to play Angry birds. Linux is for people doing the
real work, as well as for home multimedia desktop. A true general-purpose OS.

Linux applications
=============
Linux (Debian and Ubuntu) offer about 40,000 packages in it's
repositories vs 600,000 Android applications in Google Play.
But Linux apps are much heavier and professional, that Android simply lacks.
In short it add-up much more options to the potential buyers alike,
and those do not overlap with Android. Those 2 ecosystems simply
attract different categories of buyers.
You can target both with a single product.

The biggest problem is Linux drivers.

Why hardware video decoder ?
Intel Atom class CPU was tested by me, and found not adequate to run
WebM video in Full HD quality. The video stucks every few seconds.
Your ARM-based CPU is probably not better than the Atom.
This is not a problem at all for serious CPU, such as Intel Core i3,
as it can render WebM @ Full HD 1080p happily without hardware video decoders.

Without those video decoder drivers, the ARM CPU is not powerful
enough to run HD video in software. The lack of Linux video drivers
will limit the use-cases to low-quality video 240p and to Office work
on ARM-based Linux PCs.

If your partners or you could design & manufacture something like the
VIA APC system, but better, it could sell like hot cakes.

===== Competitive Analysis of ARM devices =====

Let's look at several systems: VIA APC (VIA ARM-based PC), Riko MK802
(AllWinner A10-based ARM PC), and traditional Intel x86 PC.

VIA APC $49 system:
http://www.youtube.com/watch?v=2dNA3yHEl2s&feature=related

has ARM11, 800 MHz (VIA wondermedia 8750 SoC)
512 MB of RAM
2 GB of internal flash
ports:
VGA (good) (converters VGA->DVI are much cheaper than HDMI->DVI)
4x USB (very good, no need for USB hub for most people)
neo-ITX form factor (good, as it can be mounted into existing small
mini-ATX & ITX cases with it's holes)
HDMI (for modern TV users)
Power supply (no need for USB powered hub)
Analog audio output + input (great idea for VoIP; no need to buy USB audio card)

Best point:
In short : the PC-like external I/O ports are very strong part in this
design, along with Neo-ITX form factor.

Problems in VIA APC $49 design:
1. Android OS (v2.3, it is touch-oriented, very hard to work with
mouse + keyboard)
Fix: needs Linux: Ubuntu-LTS or Debian-stable.
Video drivers are only for Android, not for Linux.
Without hardware video decoder drivers for Linux, the machine cannot
be used for HD video.
As for Android - there is no Google Play Market.
Without Google Play, it is difficult to get Android applications,
which will scare off users.
Recommended fix: ship with dual-boot OS (Android and Linux,
[Debian-or-Ubuntu]), so users can choose at boot-time. Android image
must have Google Play.
2. RAM 512 MB: not enough to power full Linux/KDE desktop.
It will constantly swap. (for instance a fresh install of Debian 6.0
x86 with KDE takes ~700 MB RAM before starting any apps)
Fix: boards needs at least 1 GB RAM
2.a.  +RAM slot for DDR3 SDRAM module. (optional, for future upgrades, PC-like)
Fix: RAM slot is a very good idea for neo-ITX sized platform. (in
addition to 1GB on-board RAM)
*if RAM slot is impossible, then put 2GB RAM on-board.
3. Storage: internal flash 2 GB is enough for Android, but not enough
for Linux/KDE. 8 GB is minimum.
Fix: put 8 GB flash on-board
4. 2D: 720p is full stop. >80% of PC monitors selling today are 1080p.
Running them in 720p (1280x720) degrades text quality, so the machine cannot be
used for text.
Fix: use display controller and SoC capable of 1080p 2D output.
5. no SATA port: cannot use internal SATA HDD for extra storage and
cannot use DVD drive.
Fix: put 1 or 2 SATA ports on board.
6. (optional) no WiFi: about 50% of users do need WiFi. Surely it can
be added on USB, but if it is inexpensive (under $8 for chip+antenna), it
would be nice to integrate on-board.
Using USB WiFi is inelegant and may create problems with drivers not
matching Linux.
7. overall: the VIA 8750 SoC chip is a crap. (ARMv6, soft-float, and
all of above)

Problems in A10-based Riko MK802 design:
1. Android is supported; no drivers for Linux.
2. no I/O ports
HDMI-to-DVI cable here costs >$25
The problem is that >80% of PC monitors are shipping with DVI. (not HDMI)
Fix: needs VGA or DVI port
USB audio card costs ~$100 (driver support for Linux ARM is questionable...)
(alternative is USB speakers, but those also cost a lot, much more
than analog speakers, same for USB microphones)
Fix: needs analog audio output + analog microphone input, like the APC.
USB ethernet card $50
Fix: needs ethernet port, like the VIA APC
USB hub $40
Fix: needs 4x USB ports, like the VIA APC

So in total cost of ownership (TCO) of MK802-based system is not cheap,
after adding all those extra costs:
$75 (Riko MK802) + $25 (DVI cable) + $50 (USB ethernet card) + $40 (USB
hub) + $100 (USB audio card) = $290.
(the prices are based on retail prices in Israel, in USD, except for
MK802 system which is world-market price on Alibaba)

At $290 for the ARM PC, it is will be very difficult sell vs.
traditional x86 Intel PC.
The cost of all those converters and cables eat _ALL_ the price
advantages of the cheap ARM SoC.

Even if I could get all of those for cheap, having lots of converters
is simply not nice user experience.

Surely the ARM and PCs have the bonus of being affordable,
noise-free, small and nice,
but are slower than Intel PCs and lack upgrade-ability (CPU socket,
RAM slots, PCI
slots, SATA ports, ...)

What is needed for the ARM PC market ? A hybrid device between
Rikomagic MK802, the VIA APC neo-ITX board, and the traditional Intel
PC, with good performance, cheap price, plus full set of I/O ports to
avoid buying lots of converters.
Full Linux support with drivers is very important.

It is my prediction, that some type of upgrade-ability in the form of
RAM slot and SATA port will narrow the gap vs. the traditional Intel
PC, and will drove PC users to ARM en-masse.

What product is needed for the ARM PC market ?

1. Neo-ITX form factor (like VIA APC)
2. ARM SoC (Allwinner A10? or Samsung Exynos 5 Dual?) with 1080p 2D
output and hardware video decoders
3. Linux drivers for it
(network/audio/2D/hardware video decoder/WiFi/...)
...your company will have to develop & test them. (or ask the SoC vendor...)
3.a. preferably Open-Source drivers, but closed-source binary-only
drivers are better than no driver (for things like the 3D GPU)
4. external Ports:
4.a. Ethernet (like VIA APC)
4.b. 4x USB (like VIA APC)
4.c. DVI or VGA (like VIA APC)
4.d. HDMI (like MK802 and VIA APC)
4.e. power supply (like VIA APC; so it could power all 4 USB devices
connected to it)
4.f. MicroSD or Full SD slot for storage (like VIA APC)
4.g. Audio ports: analog output (head-phones) and analog input
(micro-phone) (like VIA APC; for VoIP)
5. internal ports: (upgrade-ability)
5.a. SATA 1 or 2 ports (like traditional PC)
5.b. RAM slot: for DDR3 SDRAM up to 4GB module, in addition to
on-board RAM, (like traditional PC), for multi-tasking Linux desktop.
There is enough physical room for it to fit on the Neo-ITX board. (is
RAM slot costly to produce? ) [*]
6. RAM: 1 GB on-board required for basic Linux desktop.
7. Flash: at least 8 GB on-board storage
8. WiFi chip + antenna on-board (like MK802)
9. Slim design (optional; up to 2cm max. height) (unlike VIA APC,
which is too high, so it limits non-PC use-cases, such as mounting on
the back of an LCD monitor. Slim design could result in a mono-block
PC, product like Apple iMac)
10. Dual-boot image: Android (with Google Play) + Linux OS with KDE,
Debian or Ubuntu (so
users could choose which one to boot from). Advanced users could
format the extra space (from 2nd OS) to use with one OS, but new users
should see both operating systems on boot menu. It is a form of
hedging your bets.

[*] Yes I know that the Allwinner A10 does not support RAM slots, but
it can be planned for next-generation ARM PC, if you ask nicely the
SoC vendors. First generation can come without it.

Such a board, basically an improved version of the VIA APC, if priced
under $100 range, could eat the market from Intel very quickly,
and once Linux drivers become available it has the potential to become
a best-seller.
Lots of on-board hardware will avoid most driver problems.

After learning about the Riko MK802 $75 and the VIA APC $49 designs
and Intel-H61-chipset-based $50 motherboards, I believe it is
possible to build such a board.

Thanks a lot and best wishes,
--
-Alexey Eromenko "Technologov"




More information about the discussion mailing list


interactive