Slashdot Mirror
Dual-Core Allwinner A20 Powered EOMA-68 Engineering Card Available
A year after the first schematics were completed and a few months after the first prototype board shipped, Make Play Live has released Improv, the first engineering card for EOMA-68 (EOMA-68 is a specification for modular systems that splits the cpu board from the rest of the system, allowing the end user to use the same core with several devices or upgrade e.g. a tablet without having to pay for a new screen shell). From Aaron Seigo's weblog post: "The hardware of Improv is extremely capable: a dual-core ARM® Cortex-A7 System on Chip (SoC) running at 1Ghz, 1 GB of RAM, 4 GB of on-board NAND flash and a powerful OpenGL ES GPU. To access all of this hardware goodness there are a variety of ports: 2 USB2 ports (one fullsize host, one micro OTG), SD card reader, HDMI, ethernet (10/100, though the feature card has a Gigabit connector; more on that below), SATA, i2c, VGA/TTL and 8 GPIO pins. The entire device weighs less than 100 grams, is passively cooled and fits in your hand. Improv comes pre-installed with Mer OS, sporting a recent Linux kernel, systemd, and a wide variety of software tools. By default it boots into console, so if you are making a headless device you needn't worry about extra overhead running that you don't need. If you are going to hook it up to a screen (or two), then you have an amazing starting point with choices such as X.org, Wayland, Qt4, Qt5 and a full complement of KDE libraries and Plasma Workspaces.
Improv takes advantage of the open EOMA68 standard to deliver a unique design: the SoC, RAM and storage live on one card (the 'CPU card'), the feature ports are on a PCB it docks with (the 'feature board'). The two dock securely together with the CPU card sitting under the feature board nestled in a pair of rails; they are undocked from each other by pushing a mechanical ejector button."
Check out the specs and pictures. The card is available now for $75. Improv is open hardware, with the schematics licensed under the GPL and available soon.
Their blog links to this page of the Open Source Hardware Association providing a definition.
It runs Ubuntu? Crap. I thought it ran Linux.
Isn't that the card that was supposed to cost even less than the RaPi?
In the literal sense (i.e. that some people did suppose this would cost less than the Raspberry Pi), yeah, pretty much; an EOMA-68 CPU card based on an Allwinner SoC was widely reported to have an estimated price of $15.
However, this figure was (1) a BOM cost, not retail price, (2) an estimate before the design was finished (e.g. at that time, I believe the A10 SoC was being considered, whereas the now-available unit has an A20), and (3) only applied to relatively high volume (100,000 units, IIRC). It was never intended to represent a retail price at any volume, but some trigger-happy bloggers repeated the number without describing what cost it represented, some other bloggers assumed it was retail, and ever since there's been a steady stream of people whose only prior knowledge of the EOMA-68 project is that "a CPU card is supposed to cost $15, so it's cheaper than a Raspberry Pi!", and who are consequently disappointed and frustrated to learn that it costs more than that.
The announcement and website clearly state that the feature board which the EOMA68 docks to is open hardware; yes the A20 is not open hardware, and that was never stated otherwise.
> How is this thing compared (hardware wise) to Raspberry Pi ?
RPi is a single core 7o0 MHz ARM11 with 512 MB RAM and no on-board storage; Improv is a dual core 1Ghz Cortex-A7 with 1GB RAM, 4GB NAND flash and a more powerful GPU. Improv is also modular so you can swap out the CPU card as well get feature boards with additional features in future. So Improv is several times more powerful and quite a bit more flexible. You also get things like SATA with the Improv.
As for software, anything that runs on the RPi run on Improv, while the reverse is not true. Some ARM Linux OSes require hard float, such as Ubuntu, which RPi does not provide but Improv does
They are similar in hardware capacity, except that Improv is modular (not everything is hardwired on one board) and is not a sold-and-forgotten piece of hardware but has an active Free software and hardware devel community around it.
This is an engineering board, not a smartphone. If you look around what is available for prototyping and developing projects, you'll find that single core ARM is actually the common case. This is a significant amount of hardware for the market category. This is also considerably more powerful than what smartphones were shipping with 3 years ago, though today's high end phones do come with more cores.
The whole idea of the EOMA concept should (if/when it takes off big) mean that you won't have to "hope the laptop shell's $ATTRIBUTE is $VALUE". There's two reasons for this.
First, you can build your own laptop, because a lot of the complexity that makes designing your own laptop mainboard a ridiculous proposition for almost every hobbyist is now inside the CPU card -- some professionals designed, built, and tested that 6-layer PCB, and then millions (eventually, in the big picture) were run off. For your special laptop, you could if you put your mind to it do most, if not everything, with a 2-side PCB and old-school through-hole components, the main obstacles being not that you can't fit it in a full-size laptop without SMT, but that you can't find some components in through-hole versions. You can pick whatever display you want, slightly tweak the PCB design from some other EOMA-68-based laptop to suit, and have one made. And all this is much more practical than it sounds because you invest the effort once, then keep that laptop for life (ok, realistically for a decade or more) and just swap CPU cards when you need more performance.
The other, and even bigger, reason, is because some manufacturer, somewhere, will make a shell with the characteristics you want. Sure, your concern might only occur in a fraction of a percent of consumer (actually, your concern about the backlight is IMO a horrible example, because the whole industry is moving from CCFL to LED for a number of reasons), but when some small Chinese factory is looking for a profitable niche to exploit, that fraction of a percent is a prime target. Because of EOMA, they
(1) have less design work to do to make a new model (just like the hobbyist)
(2) can keep selling that model without investing in a periodic redesign, and without it becoming obsolete and unsellable due to last year's CPU -- just every year buy a load of the hot new CPU cards and receive a magic spec bump, or ship it without a CPU card and let the user slot their new or old card
(3) even if/when they go out of business (or just abandon your market segment) and stop selling new shells, all the used ones keep going (until they break/wear out) without obsolescence.
(1) and (2) mean less cost to pick up tiny market segments, which means niches will be more profitable and thus better served; (3) means that even if you're part of a niche market that looked big enough to make a good profit, but turned out not to be, you get to reap the benefits of some company's "mistake" in pursuing that niche long after the company's learned and moved on.
Regarding the last point particularly, contrast that to the Fujitsu U820 I bought a few years ago, because I really loved the form-factor and the high-PPI screen. At the time, the 1.6GHz Atom processor was slowish and the soldered-on RAM was cramped; it's flat-out obsolete now. The "successor" UH900 is a straight clamshell, lacking the flip-screen which lets the U820 become a paperback-sized tablet, and I'm left casting about amongst gadgets like the Asus Transformer series looking for a near-enough equivalent. If the U820 had been EOMA-based, then Fujitsu could go their way, selling UH900s with better mass-market appeal, but I could keep going mine, swapping up to (say) a quad-core 1.8GHz ARM card in that same delightful chassis.
There is HDMI out, which is digital.
" Of course if you believe this thing will appear on time, work and ever see another module which is compatible with it I have a nice bridge here to sell you"
So, it works. How do we know? We already have finished pieces in hand and use them.
Other modules: are alread add-ons such as VGA connectors and keyboard kits in prototyping; I've already seen two more feature boards; as for other CPU cards, those are further away but on the roadmap.
Who peed in your cereal?
I know it's easier to be cynical than to be helpful, but if you support projects like this they actually do go further.
No, translation "we've been working very hard on this device, and will be releasing them at shipping time". We've put the Open Hardware Logo on the feature board and everyone who has participated in this project has licensed their contributions under the GPL. We're not about to start our first product by violating each other's licenses. Please, give us a bit more credit than that. Most of the people involved have been releasing things far more valuable and work intensive than this as Free software/hardware over the years, after all.
Hey all,
I'm oliver, from http://linux-sunxi.org, the community revolving around the kernel development around this SoC.
First off, the BOARD is OSWH, not the SoC. Now, for those who'd only call it OSHW if the VHDL code would be available, while utopian, that's just plain silly. OpenCores is for that ;) So yeah, this is all OSHW goodness.
Then, documentation wise, yes we lack a lot. Allwinner hasn't released everything to anybody yet, some pieces haven't received any docs at all yet, most likely because it hasn't been written yet, some pieces they can't share the docs as they are under NDA themselves. But for most bits that's not important as we do have code for pretty much everything. The docs we do have, are the 'standard' usermanual, in english, with a lot (but as said before not all) register information. You can download and view them over at http://dl.linux-sunxi.org/ in the various subdirectories. The only closed blobs right now are GPS, GPU and VPU.
Now, the GPS isn't really that important and it hasn't been reverse engineered yet, is because there's no hardware using the GPS. Most platforms use UART or USB for GPS so this hasn't been on anybody's radar. We do have a gps.ko with debugging symbols so once the need arises, it's doable, nobody really just had a need for this.
The GPU, talented Luc Verhagen has been working for the past 1 - 2 years on the LIMA project. This allows a fully opensource stack to be used with the MALI GPU. Luc actually uses the A10/A20 as main development platform (amongst another one). While this is still very much WiP I'm sure we all seen the quake timedemo Luc did last year at fosdem where he actually beat the ARM binary mali blob. Here is his latest mesa work. http://www.youtube.com/watch?v=4WOILEYAxWE but we have to be honest, it's not done yet, so for now we are still stuck with the mali blobs. But yeah, hold your breath for that one.
The VPU is also being reverse engineerd. This is much further behind of LIMA so I shouldn't talk too much about it and get people excited yet, but here's a decoding demo: http://linux-sunxi.org/Reverse_Engineering/Cedar_Status where you can see we can decode h264 video without using any proprietary blobs (mali isn't needed for this).
Then finally, compared to all other SoC's out there that do have some form of Linux support, the Allwinner chip is one of the limited ones, that actually have u-boot support. I'd almost say full u-boot, but MTD support is still WiP.
So to compare this to the Raspberry Pi, It's much faster (armv7 vs armv6, hard-float available, dual core CPU and dual core GPU, up to 2 GiB ram possible to name just a few).
Finally, is everything open? No, the BROM isn't open source, the BOOT-ROM, a 32k block embedded (unchangable) in the chip that performs initial boot. What it does is check the supported media (SPI, NAND, SD) for a valid signature and boots it. I'm quite sure the same blob is in any CPU on the market right now. Your AMD or Intel CPU also has a bootrom, that tells it to load the bios from SPI into ram and start executing it. So this is moot, but I do think it's fair mentioning it.
So hopefully I've put some things to rest here, if not I'll try to check back at a later date and reply appropriately.
If you want more info, I'm planning to hold a talk at FOSDEM 2014 so stay tuned over at http://fosdem.org
As others have said, the Pi has an FPU and supports hard float. The issue with running Ubuntu on the Pi is that they only support ARMv7 while the Pi is ARMv6. I also don't think the Mali 400 MP2 in this thing is more powerful than the Videocore IV in the Pi.
Mada mada dane.
You own a 40nm process fab? Are you a multi-billionaire?
Face the facts, you'll either be stuck running your hardware on very very expensive $1000+ FPGAs in order to get 1/10th of the performance of a $10 Allwinner SoC, or you'll be getting 1/100th the performance on a mere $200 FPGA dev board.
In the mean time, most people are happy to just use free software on the hardware. The software enables the hardware, but it's important that it is free so that problems can be fixed, code can be made more efficient, etc, by the people who run the hardware. This board is "open" in the sense that the aim to have no binary blobs for the functional units.
As a comment above says, they still have work to do on the GPS unit, the GPU (open source LIMA driver) and the VPU (can currently play H.264 only). In addition there is a boot ROM that is very much a ROM and which scans the various buses for a boot image.
Faster CPU, and two of them. Note that the A7 is not amazingly faster (maybe 2x) than the ARM11 in the RPi, but it's more up to date (ARMv7 instead of the very old ARMv6) - and two of them does help a lot.
Slower GPU. The RPi uses a very advanced SoC in terms of GPU. The ARM11 is actually just a microcontroller for the GPU. The SoC was aimed at video applications, and is pretty darned amazing, for the price.
1GB RAM instead of 512MB.
4GB of flash storage, instead of none. Not to be sniffed at, but most people would just stick a fast SD card in anyway.
SATA support, versus no SATA support.
Open form factor, versus custom. Two boards (SoC board, EOMA68 I/O board) versus one board. Higher cost.
No software ecosystem versus an ecosystem of 2 million sold devices.
Allwinner SoCs are also absolutely unbrickable. This with or without an SD card. You can always get it to show up over USB by holding a device specific button at boot, and then you can get it to boot whatever you want.
The big advantage of the Allwinner chips (especially the mali based ones) is their very high degree of freedom. The GPU and VPU are the two bits which require work still, but progress is good. There is full u-boot source, there is full linux kernel source, and parts are making it upstream. All there is that is not free and that cannot be made free is the tiny bit of code in some microcontroller to make it act like a USB device during its special unbrickable boot mode.
Given that the RPi has a massively powerful DSP running the show on a 2MB large RTOS that is absolutely closed, the allwinner devices are _unbelievably_ free.
as for other CPU cards, those are further away but on the roadmap.
they are indeed. tracking down a cost-effective desirable SoC from - and this is also a really important bit - a fabless semiconductor company that respects the GPL - is very very hard. let's go through the list so far of CPU Cards that i've 30-98% made the PCB CAD/CAM drawings for (the A20 one is the only one that's reached 100% completion so far)
* AM3389 CPU Card. GPL-compliant: yes. cost-effective: most definitely not. desirable: well, it turned out that there was a proprietary blob for HDMI, and it was to be an FSF-Endorseable CPU Card, so no.
* iMX6 CPU Card. GPL-compliant: yes. cost-effective: at $35 for a quad-core SoC in 1k volumes when the competition is $USD 12: mmmm.... no. desirable: yes.
* Ingenic jz4760 CPU Card. GPL-compliant: yes. cost-effective: yes (around $7). desirable: as it's only a 1ghz single-core MIPS with no HDMI output... mmm... no not really.
* Rockchip RK3188 Quad-core CPU Card. GPL-compliant: no. only "leaked" source code is available. cost-effective: yes (around $12. for quad-core! amazing). desirable: yes (good features). but, the GPL-compliance nixes it. that and the huge NREs demanded by rockchip for their development board details.
the list keeps going on and on like this. much of these issues go away once we have some sales. so if you'd like to see this project succeed, help out by buying one of these engineering boards. in the future you'll be able to re-purpose the old CPU Card by getting an alternative chassis (just the chassis), or you'd be able to sell the old CPU Card on ebay.
The whole idea of the EOMA concept should (if/when it takes off big) mean that you won't have to "hope the laptop shell's $ATTRIBUTE is $VALUE".
you know what? whoever you are, foobar bazbot, i'm amazed and delighted to see that you clearly Get this concept. there are a couple of things that you left out:
1) from a CPU Card manufacturer's perspective, they love the fact that a short-lived SoC in a ready-to-go pre-packaged product can be sold in much bigger volume because it's shared - for the relatively short duration that the SoC has its day - across potentially dozens of mass-volume products.
2) from your perspective (1) translates into cost savings due to the CPU Card manufacturer being able to take advantage of stable huge volume pricing, as well as the Foundries, having larger orders, being able to dedicate and optimise a fab to get the yields up. both the volumes and the better yields automatically (one might hope!) translate into lower pricing
3) from a cost perspective, the fact that there is about an extra $6 on the BOM when compared to a monolithic product... this is *completely* dwarfed by the immense cost saving when you buy one or more EOMA68-compliant "chassis" and share a single CPU Card between them. laptop and tablet are the two obvious examples, with the clear additional benefit that applications and data transfer conveniently *between* the products.
4) from an environmental waste perspective, EOMA68 significantly reduces e-waste by making it possible to re-purpose older CPU Cards down a chain. today's latest-and-greatest laptop/tablet CPU Card becomes tomorrow's router/NAS/SoHo server CPU Card.
so there is an enormous amount going on here in what appears to be an otherwise unobtrusive "wtf??" moment. i haven't begun to describe the benefits to the linux kernel developers yet (but have posted a number of times on LKML explaining the N CPU Cards plus M products instead of N*M monolithic designs.)
You don't have to be a multi-billionaire to own a fab. You also don't need millions to use a fab, either.
And 1/10th speed on a $1000 FPGA? Forget it. You're looking at 1/300th speed on a $250K FPGA (seriously - the ASIC I used ran at 1GHz on silicon, and 3MHz on the FPGAs). Granted, if you spend $1M or so, you can get maybe a 1/50-1/100th speed FPGA system (the FPGAs themselves cost $35K in 1000 quantity, and the system had 4-8 or more of those - yes, easily over a quarter mill in FPGAs alone).
In fact, university students routinely crank out ASICs relatively cheaply - so cheap that they're basically "free" so they run experimentation on the fab process including transistor matching, transistor performance qualifications, etc.
Granted, we're not talking 40nm here, as the student budget is closer to 1um - no deep sub-micron here. And the largest cost after that is well, the package. But such ancient fab equipment is basically free at this point in time, and you can still use basic photolithography systems so your masks don't have to cost $100K each (yes, masks are around $100K, and you need 20-30 depending on how many processing steps and layers you choose - that means taping out costs anywhere from $2M-3M.).
If you ever wonder why you have A0/A1/A2/B0 style steppings - it tells you what masks changed - you can do a full set from transistors to metal layers (A->B->C), or just modifications to the metal layers only (x0->x1->x2). The latter is often done because there are always spare transistors that are not connected to anything, so fixing bugs by wiring up those unused transistors saves a number of masks. Do it carefully enough and you can really minimize the number of new masks you need to wire up additional logic. Transistor density is often so low as wire density is the issue (which is why we have 10+ metal layers), so making a whole pile of spare gates in the unused areas means you can easily edit significant pieces of logic without needing a transistor level mask change. Heck, you can make them of different sizes in case the old logic lacks sufficient drive capability.
Being able to replace the core of your tablet doesn't fix sctrached screens, aged batteries, and general wear...
you need to remember to view this from both sides. it's possible to replace *either* the CPU Card *or* the chassis, and in each case you have significant advantages and lower costs.
when did you ever buy a hermetically-sealed product that you could upgrade? the clue is in the word "hermetically-sealed".... :)
and any tablet that you can replace something on is going to be thicker
true. i have a design which uses PCCARD 3.3mm. if you have around $250,000 for the tooling costs for all the parts (assemblies, housings, sockets, casework) i can get it done... maybe in about 6 months time. or... we could use off-the-shelf parts and get immediately into production.
which would you prefer? perfect waffle-ware - more expensive due to the investment and NREs - or actual product that's reasonably-priced because there's no investor overheads?
and less "tablet like" than a 'nice' current tablet.
simply not true.