In the fourth comment he explains some of his design choices:
One of those factors I mentioned above was making sure the
board was "replicable". This adds a number of constraints,
such as:
* The price cannot include any subsidies (otherwise people
could not recreate the board at close to the same cost)
* The parts must have a long product life cycle (this
eliminates some of the other chips that we might have
considered) and be readily available (otherwise there would be
shortages in supply and people wouldn't be able to source the
parts to make their own)
* Maker and hacker friendly meant including access to a number
of buses and other interfaces, which also limited our part
selection as well as impacting cost.
There is a fairly lengthy story here, which I'm happy to share
with anyone interested. All in all, I think we managed to come
up with a very capable board that will allow people to tweak
it and hack it and expand it while providing strong
performance (especially in terms of I/O) and a decent price
point. More on that as information becomes available...
Edit:
To the best of my knowledge Intel has never focused on releasing designs that an individual can build/modify on their own. To me this represents a very exciting change in Intel's attitude toward "the little guy." Further, to anyone looking to build their own hardware similar to the Raspbery Pi, good luck trying to get Broadcom to talk to you, let alone give you a BSP or sell you chips. Their NDA process/requirements is/are obscene...
Bunnie's open-source laptop project (Novena) also had similar philosophies with part selection, with the addition of another requirement I think is pretty important to the maker/hacker community:
"You can download, without NDA, the datasheets for all the components, and key peripheral options are available so it’s possible to build a complete firmware from source with no opaque blobs." [1]
I wonder how many of the Minnow's components require NDA/signup/etc for datasheets, or to access (proprietary) firmware.
For a real world story on (not) interacting with Broadcom (and AMD, who don't make a very good impression in that story either), read Jens' second comment on http://review.coreboot.org/#/c/2463/ (Feb 21).
I'm just not sure if Intel's process is really better for small customers (but you probably end up in the "small" tier faster than with Broadcom)
> just not sure if Intel's process is really better for small customers
It depends on what you're buying, but for SoC processors and especially chipsets, it's not so great. And re: "who's small" - yeah, I think ~1M units annually is still "small" to Intel.
However, some of their memory products (NOR and EEPROM come to mind) are very attainable for smaller customers.
This is exactly the type of marketing effort Intel needs in order to steal the embedded systems market back from ARM. As an embedded systems hacker who normally works on projects with budgets well under $10m, Intel has never really been on my radar. They're incredibly unapproachable until you're a bigger OEM, and chances are if you've started from scratch you're already used to (and tooled for) designing for ARM.
Creating an open platform around Atom makes Intel much, much more approachable.
With regard to price, I'm not discouraged at all. $200 is well within my personal budget, and I've got a vision-heavy robotics project in mind that I'd love to use this for.
Why not use an Raspberry Pi or a more powerful ARMv7 SBC? ARM software support is getting better every day, but the vast, vast majority of software is still written for x86. As examples, libBLAS, LAPACK, libav, OpenCV - all of these were written and optimized for x86 MMX, then SSE*, first. Their performance on ARMv7 has come a long way very quickly, but ARM performance is a secondary goal for a lot of these types of projects.
Shit, how well does Matlab or Octave run on ARM? I know more than a few people who'd salivate at the chance to write sensing/control code in Octave and run it on a $200 SBC...
So far their firmware isn't open source. Tianocore isn't useful on real x86 hardware without hardware initialization code (which Tiano doesn't provide).
Let's see if an Intel group manages to coerce the chipset group to release those "precious" (but in reality, very boring) details on memory initialization to the world.
So far I'm doubtful, even if it's "just" for Atom.
[edit: In fact, their website (http://www.minnowboard.org/technical-features/) states "All software will be provided in image and Open Source form (with binary elements per IP restrictions)". That exception _will_ cover Platform Initialization code...]
I think internally this is a huge [1] experiment for Intel. They're saying "is doing the work to support the low-volume 'little guy' going to do anything for our high-volume sales?"
I have a feeling were it up to the people leading the project, the init code would be made available. However Intel as a company is only going to take so much risk in support of this experiment. I have a feeling that init code and firmware for softloaded peripherals will be "business as usual" for quite a while.
That said, the squeaky wheel gets the grease. If this is successful and if customers are very vocal about not wanting "half-assed" open systems, it might eventually get there.
1: Actually, in terms of resources I'm sure it's tiny. In terms of culture on the other hand...
I'm involved in the open source firmware struggle with Intel for a number of years now. There are Intel customers with solid business cases, but it actually got worse in the last few years.
Can you say more? What's made it worse recently? I was witness to some struggle between GumStix and Intel/Marvell back around when Marvell bought the XScale processor line, but I've been out of touch with Intel on this issue since then. However I've seen countless peripheral silicon manufacturers maintain strangleholds on their softloaded firmware... Though with things like wireless chipsets and DSP algorithms, it's arguable there's real IP there worth protecting.
I don't think that particular part changed much over the years, and it's also relatively easy to figure out with SerialICE (www.serialice.com), since CAR is one of the first things to happen these days.
Since Intel wants everyone to use their code (preferably in binary form), I'm not sure how useful the BIOS Writer's Guide still is. Probably just a guide on how to integrate their code into whatever you're doing yourself.
Sort of like the Pandaboard. Not as many GPIOs though.
Interesting that Intel is dipping its toe in here, At one level it validates the ARM SoC market that is percolating around the embedded space but its also typical in that this doesn't put any of the 'good stuff' Intel could put out there into the market.
Perhaps most interesting has been the continued fraying around the windows/intel empire, what with Microsoft offering an ARM version of Windows on the RT and Intel building clearly not PC compatible motherboards.
Wow, is it just me or is Intel missing the market on this one? The price is about 4x too high for what you get (compared to other ARM based boards) and the size is a little big as well.
Apparently their market is developers who don't own computers:
> However, the MinnowBoard is still cheaper than most PCs and could appeal to developers looking to write and test commercial applications before they're deployed in servers, embedded devices and other computers.
I'm baffled as to why anyone would purchase this board. Raspberry Pi[1] destroys it on price/value so _will_ Parallella[2]. Why not wait a couple months and get a super computer![3]
The RPi isn't really comparable. The Minnowboard is a much (much!) more capable piece of hardware. Maybe 3-4x as fast at typical CPU tasks, 2-4x the DRAM, gigabit ethernet, PVR SGX vs. that weird Broadcom thing in the Pi.
Is it worth 6x the price? Certainly not for all uses (and I'll admit I'm disappointed to see this thing launch at $200 too). But there are tasks for which this board would work where the RPi wouldn't. And there it competes most directly with things like the Pandaboard, against which it appears to compete pretty well.
> However, the MinnowBoard is still cheaper than most PCs and could appeal to developers looking to write and test commercial applications before they're deployed in servers, embedded devices and other computers.
Relying on having spare computers for testing is also kind of a thing of the past in the age of cheap PC resources and virtual machines. I gave up keeping old computers around for software testing 5-6 years ago when I bought my first 4 core CPU. Also keeps my work area from getting really really hot in the summer.
I'm baffled as to why anyone would purchase this board.
Because it's small enough to be useful but a hell of a lot less work to get started on. I would much rather use this for a personal project (not a product) than spend a year learning a new architecture.
I think the fact that it's an x86 board makes it an attractive first step from Intel into the super-small form factor market -- not that I'd buy an Atom-based snail when I already have a Raspberry-pi turtle at home.
They might not be cost-competitive. Intel's process advantage comes from vertical integration. But it also means that they need to keep pushing out high-margin flagship products to support their scale of operations and technology advantage. A niche thing like this is likely to get thrown under the bus internally, but they can't ignore the pressure from below and the deteriorating PC market anymore.
In contrast, the ARM ecosystem has flexibility to mix and match the design, manufacturing and marketing amongst a group of companies. This is why a lot of people are confident about ARM's fortunes in coming years; the business model favors them in a "cheaper is better" computing market.
OK, why are they discriminating the rest of the maker, hacker, tinker, population by pricing it so high? They need to take a page of the play book from one of the ARM board folks.
It's interesting, just yesterday I scanned the low cost system boards looking for one with SATA. I was wondering if I should replace the old PCs I use as filestores with something smaller, lower power, that could just run linux and rsync for me.
This does it ... but as others have said, the price is not really attractive compared to a used pc. I don't fault the design team for this. It looks like a great little board ... but probably for strategic reasons Intel would rather have us continue with those old pcs, rather than risk cannibalization of low-end pcs.
Update: I think I was slow to see the embedded focus of this board and it's IO. So of course it is priced against that world, and not the world of low-end or used PCs.
FWIW also, comparing that $200 pricepoint, Newegg recently had a Lenovo server Intel Core i3-3220 3.3GH Dual-Core; 4GB RAM; DVD ROM, no-disk, no-os, for $250. (reputed backdoor at no extra cost!)
Update: I just went to techbargains and searched "server", that one was recently expired. When I'm in shopping mode I just keep an eye on a couple such shopping sites to see what comes by.
We'd prefer if you'd parse your network packets on the CPU. We even built a special part of the chip to do it for you and hardwired your existing com ports to it! I'm sure you developers don't want to touch that stuff anyway.
Would someone be able to confirm this, because if it's the case then there might not be as much Linux support, especially for 3D acceleration, for this board.
The datasheet does not have any power consumption info. How big does the PS have to be? Otherwise it could be a useful board, $200 is a little steep now-a-days. The APC is $100, the RPi is $40 etc.
Project member here. I was disappointed with the way the PCWorld article described it as an "open source PC". MinnowBoard is really an open hardware embedded platform. If you're looking for a small form factor PC, the NUC is definitely the way to go. MinnowBoard is for embedded applications or product development where interfacing with custom hardware (whether I2C sensors, or custom FPGAs through PCIe, etc) is needed.
I can't believe you guys price it at 199$, totally unacceptable as embedded platform. at 1/4 of the price, there are plenty of ARM solutions. Why even bother?
If the board doesn't meet your requirements, then don't use it. The MinnowBoard is part of an ecosystem of boards that offer different feature sets at different prices. One differentiator the MinnowBoard has is that its SATA and Gigabit ethernet are powered by PCIexpress, so you're definitely going to see performance reflect that. Likewise, we offer PCIe lanes through our expansion connector, so the MinnowBoard could be interfaced with high-speed hardware (think FPGAs) for all sorts of interesting applications.
I find the comparisons of MinnowBoard with ultra low-cost embedded boards like RasberryPi to be pretty silly. The MinnowBoard is not Intel's version of a RasberyPi. The media may play this up just to generate conflict and to set up a straw man comparison, but I hope people are capable of seeing through that.
> but I hope people are capable of seeing through that
You and me both, but sadly I don't think it's going to go that way in the general tech-enthusiast media without a pretty significant "nudge" from you folks and maybe a few others. Unless you've got some experience with performance-demanding applications in embedded systems, you have no idea why RPi/Minnow is an Apples/Oranges comparison.
But everyone understands price! All they're seeing is two tiny boards, one produced by David that sells like hotcakes, the other produced by Goliath. Goliath's is late enough to be called a reaction to David's move, and it's nearly six times the price...
It depends on what you're doing. If you're making an internet toaster, go buy an arduino. If you're doing anything CPU intensive, or if you're not familiar with embedded systems tooling, cross-compilation, Yocto or similar build systems, Atom >> ARM.
Honest question, no snark - Why not gut out an elcheapo laptop's mainboard (Atom or AMD Brazos) and dedicate a usb port to an Arduino or something. You get the power of x86 along with the expandability (i.e. GPIO, SPI, I2C, etc) of a microcontroller chip.
I've done sort of what you suggest professionally on projects where design decisions are driven largely by per-unit cost. If the SoCs that do everything you need aren't cheap/available, go with the best-fit main microprocessor to handle the heavy lifting, and one or two cheap micros to fill in the missing pieces. It's also a good approach for when a part of your solution needs good strict real-time software.
However for low-volume and/or personal projects, dev cost/time often trumps hardware cost and heterogeneous systems have a whole host of secondary challenges. Specific to your recommendation, it's a more complicated power architecture, more components to enclose, more tooling to worry about (software and hardware), and I have to worry about how to synchronize and communicate between the SBC and the micro/arduino.
It's worth an extra $100+ to be able to focus my limited free time on solving the problem I want to solve rather than on "shaving yaks."
It all depends. Are you running off batteries, or wall power? Does your system have idle time that it can take advantage of for power savings? Were the boards in question designed with features to allow for low-power sleep, etc? Are you working with analog? Do your boards have low noise supplies, or do they expect a low noise input source? Are you doing any high current or high voltage switching? Is power-coupled noise an issue? Are we doing any switching of mains power? Is safety an issue? Should we use isolated supplies? If so, how much isolation do we need?
Software folks, myself included, tend to drastically underestimate the complexity of power design...
it's a 1GHz single core CPU from 2010. While Atom might have some IPC advantage over ARM, I doubt that as a whole it would be competitive against high-perf ARM boards, like the ODROIDs (up to 1.7GHz quad-core, beginning at 89 dollars).
[Edit: In the embedded world "as a whole" rarely makes sense. Design takes a very top-down approach where only specific features of a hardware platform (those necessary for the OEM product/solution) are considered for value comparison. This is why most SoC vendors have so damn many chips in each of their SoC product families.]
It depends on the problem you're solving, how much time/effort you're willing to dedicate to your solution, and what type of solution you choose.
If you have an embarrassingly parallel problem that optimizes well for ARMv7, you're 100% correct. If you're working on a more serial solution utilizing libraries optimized for SSE, Intel cache heuristics, Intel pipelining techniques, or similar (many of which don't exist or have equivalent siblings on ARM), I'm betting on the single-core 1GHz Atom pony.
While I have great respect for Arduino, if you are looking to do something with networking, go buy a BeagleBone Black (or other cheap ARM linux board).
I was using a bit of hyperbole to illustrate the spectrum of choices. I think if you're building an internet toaster, you might want to reevaluate your priorities.
My only disappointment is that there's no option for a 64-bit Atom. Yes I know that would mean a 2GB option instead of 1GB would likely be required, but doing OS development is way easier on a 64-bit system.
The price depends on the specific model (last I checked there where three). The one you linked is the dual HDMI, i3 one. I purchased a 1xHDMI Celeron one for $150. There are also bare motherboard versions.
Oh ok, I did a quick search and found this blog article:
http://intelnuc.tumblr.com/
and it looked like the cheapest model was the dual port HDMI one.. but you're right, checking on Intel site[1] there's also a Celeron based model, DCCP847DYE, that costs 165$ on Amazon[2]
I think those all come with fans, they have two PCIe connectors, but they are the fragile edge connector style, so those two things rule them out for a lot of embedded applications.
https://plus.google.com/101378268367730737515/posts/DGQ9Nkr5...
In the fourth comment he explains some of his design choices:
Edit:To the best of my knowledge Intel has never focused on releasing designs that an individual can build/modify on their own. To me this represents a very exciting change in Intel's attitude toward "the little guy." Further, to anyone looking to build their own hardware similar to the Raspbery Pi, good luck trying to get Broadcom to talk to you, let alone give you a BSP or sell you chips. Their NDA process/requirements is/are obscene...