Intel Edison

[marshallh]

This is poop. Not even means to talk to an FPGA at speed. All that CPU power is useless with no way to move data in/out.

[miguelvp]

This is poop. Not even means to talk to an FPGA at speed. All that CPU power is useless with no way to move data in/out.

It does have 40 GPIOs that operate at 1.8V, surely you could connect an FPGA to that if you stick with the 1.8V core voltage.

The 70 pin connector qty 1 are under $2.

That said, if you want to talk to anything else you'll need level shifters, also I haven't read on how the GPIOs are accessed so not sure on their performance.

[Rigby]

I just can't help but feel like they have no idea what they're doing in the embedded market.

This seems extremely useful to me.  dual core 500MHz x86 Atom CPU?  That's amazing, to me.

This is not something that is deployed as-is; you build your solution around it.  In that way, this is a very impressive bit of kit, in my eyes.  Built-in Wi-Fi and BT?  I'll take two.

[David Hess]

not to mention third party usb transceiver and wifi chip on a product made by the BIGGEST BADDEST mofo on the block with most advanced chip fabs on the planet? They couldnt even bother to integrate those on silicon?

There may be good reasons for this.  The semiconductor process used for the processor apparently only supports 1.8 volt I/O so including an integrated USB physical interface or WiFi may be impossible or at least uneconomical.

This is poop. Not even means to talk to an FPGA at speed. All that CPU power is useless with no way to move data in/out.

I had much the same thought.  At least the Pi and BeagleBone can natively drive high information content displays.  If I am not driving either a display or high performance interfaces, then that limits the usefulness of high performance processing and I can use a cheaper ARM microcontroller.

Maybe there is a fast bus hidden in those 40 bits of GPIO but if so, I would have expected them to provide details and none of the expansion boards are taking advantage of such a thing.  I wonder how the USB controller interfaces to the processor.

[SirNick]

OK, then in all seriousness, please explain what this thing will be good for?  Here's what I see:

The hobbyists have plenty of options.  AVR, a range of PICs, and entry-level ARM stuff out the wazoo.  There's pretty much nothing you can't build, from blinky to your own tablet.

So what does it have going for it?  It's low power (so is ARM), fast (so is ARM), lots of higher-level I/O potential (like many ARMs), and runs x86 code (OK, unique advantage there -- although rarely all that relevant).

Now, the downsides:  It's expensive (unlike ARM and AVR and PIC), it isn't quite a PC (unlike any of a half-dozen small-enough form factors), and AFAIK not available as bare ICs to integrate into a final design -- so you're essentially stuck with an overpriced dev board forever.

There's such a narrow niche of products that this is well-suited for, I just don't see what market segment it's supposed to be targeting?  What does it do that isn't being done, better and cheaper, by any of the existing players?

It all just seems like Intel felt left out of the embedded market's explosion, wanted to play too, but insists on bringing their flagship IP into the game -- despite the fact that it's not really all that well suited.

[Rigby]

there are lots of competing options for anything this platform could provide, agreed.  There are lots of solutions (usually) for any problem one encounters.

I feel that it's less about Intel joining the fray (that's their prerogative) than it is somehow an offensive move in your eyes.  Am I right?  Probably not.

There are lots of solutions that anyone can choose for whatever project they're working on.  This is just as valid of a choice as any other, considering that individual people have individual preferences and comfort levels. 

In short: I don't see a problem with having MORE ways to solve a problem.

[miguelvp]

You should read this document really:

https://communities.intel.com/servlet/JiveServlet/previewBody/23158-102-3-27299/edison-module_HG_331189-001.pdf

Specially section 4, 5 and 2.

Not sure why they disabled all the graphics processing cores (section 2)

[marshallh]

Maybe the edisons use binned Atoms with bad graphics

[amyk]

I'm not so sure what Intel is trying to do here either, by putting really powerful x86 CPUs in small packages with constrained I/O options. At least give it PCI-E or PCI, even legacy ISA would open up a ton more connectivity possibilities than this...

The other thread here about the Galileo is similar: that one is PC-enough to run a Windows variant, but there's really no point in doing that if adding a video card so you can actually use 'Windows' makes the whole thing look rather silly.

[SirNick]

Competing options are good, so more power to 'em.  But their strategy is just baffling.  For one, there's an Arduino board.  But, the document Miguel linked to claims the SoC is targeted to the smartphone market.  Well, good luck with that.

[miguelvp]

Surely you don't need a full PCIe video card, these for example will do if they have drivers

http://www.nextwarehouse.com/item/?995971_g10e

http://www.nextwarehouse.com/item/?1182700_g10e

There is also mini pci express to mini pci that opens Nvidia boards.

[Rigby]

I'm not so sure what Intel is trying to do here either, by putting really powerful x86 CPUs in small packages with constrained I/O options. At least give it PCI-E or PCI, even legacy ISA would open up a ton more connectivity possibilities than this...

The other thread here about the Galileo is similar: that one is PC-enough to run a Windows variant, but there's really no point in doing that if adding a video card so you can actually use 'Windows' makes the whole thing look rather silly.

There are many production variants of Windows that run headless.  This is just another that could, and the Galileo is another that can.  Headless Windows is just as valid as headless Linux, especially given just how slim that Galileo variant really is.

[langwadt]

I just can't help but feel like they have no idea what they're doing in the embedded market.

This seems extremely useful to me.  dual core 500MHz x86 Atom CPU?  That's amazing, to me.

This is not something that is deployed as-is; you build your solution around it.  In that way, this is a very impressive bit of kit, in my eyes.  Built-in Wi-Fi and BT?  I'll take two.

But why would you pick a dual core 500MHz x86 over a dual core 1GHz ARM ? Sure in theory you might be able to run windows but who would want to?

I can only see that Intel refusing to face the fact that they lost as the only reason that they don't make an ARM, it is definitely not technology or production capability that is holding them back
afaik they are already making FPGAs with ARM cores for Altera

[ludzinc]

There is a lot of legacy code for embedded x86 machines out there.

That's one reason!

[Rigby]

I just can't help but feel like they have no idea what they're doing in the embedded market.

This seems extremely useful to me.  dual core 500MHz x86 Atom CPU?  That's amazing, to me.

This is not something that is deployed as-is; you build your solution around it.  In that way, this is a very impressive bit of kit, in my eyes.  Built-in Wi-Fi and BT?  I'll take two.

But why would you pick a dual core 500MHz x86 over a dual core 1GHz ARM ? Sure in theory you might be able to run windows but who would want to?

I can only see that Intel refusing to face the fact that they lost as the only reason that they don't make an ARM, it is definitely not technology or production capability that is holding them back
afaik they are already making FPGAs with ARM cores for Altera

By that logic the only valid platform for embedded development is ARM. 

Why use ATMega chips when there are dual core ARM chips?

If that is not what you're saying please restate it.

[legacy]

Exactly, why everything MUST be ARM ?

btw, i like the sparkfun reaction the Intel's new product 

[miguelvp]

The edison has out of order execution for what is worth.

[Stonent]

Exactly, why everything MUST be ARM ?

btw, i like the sparkfun reaction the Intel's new product 

Make a zillion adapter boards?

[langwadt]

I just can't help but feel like they have no idea what they're doing in the embedded market.

This seems extremely useful to me.  dual core 500MHz x86 Atom CPU?  That's amazing, to me.

This is not something that is deployed as-is; you build your solution around it.  In that way, this is a very impressive bit of kit, in my eyes.  Built-in Wi-Fi and BT?  I'll take two.

But why would you pick a dual core 500MHz x86 over a dual core 1GHz ARM ? Sure in theory you might be able to run windows but who would want to?

I can only see that Intel refusing to face the fact that they lost as the only reason that they don't make an ARM, it is definitely not technology or production capability that is holding them back
afaik they are already making FPGAs with ARM cores for Altera

By that logic the only valid platform for embedded development is ARM. 

Why use ATMega chips when there are dual core ARM chips?

If that is not what you're saying please restate it.

there's embedded and then there's embedded

ATMega would be more in line with something like ARM cortex M0, neither have mmu and can't run regular linux. Apart from being capable of 5v IO it is hard to see why you would
choose an ATMega given a clean sheet

dual core ARM and now Edison is some where between that and a full blown PC, just like the obvious choice for a desktop is an x86 the obvious choice for some thing just below
that is an ARM

[timb]

If you've got all these GPIOs, there's no reason you can't just bit bang an LVDS interface to an FPGA. The thing has two 500MHz cores plus a 100MHz MCU, it'll be plenty fast.

Also, an I right in understanding it has a Mini PCI-e interface?


Sent from my Smartphone

[miguelvp]

If you've got all these GPIOs, there's no reason you can't just bit bang an LVDS interface to an FPGA. The thing has two 500MHz cores plus a 100MHz MCU, it'll be plenty fast.

Also, an I right in understanding it has a Mini PCI-e interface?


Sent from my Smartphone

The Galileo does have a Mini PCI-e not the Edison. The Edison is limited to 50MHz I/O speed as far as I can tell from the doc unless I'm missing something.

[timb]

Ugh, fucking Intel... It's like that first Galileo board they released with I2C based GPIO. What were they thinking???


Sent from my Smartphone

[marshallh]

No PCIe.
Also, GPIO on embedded CPUs is usually a losing scenario. Unless there's DMA provisions you could be stuck with 10mhz toggle rates.

[miguelvp]

No PCIe.
Also, GPIO on embedded CPUs is usually a losing scenario. Unless there's DMA provisions you could be stuck with 10mhz toggle rates.

And you are spot on (from their GPIO section):

To ensure that a pulse is detected by the edge detection register, the pulse should be five clock cycles long.
• 100 ns for a 50 MHz clock when SoC is in S0 state.

[David Hess]

If you've got all these GPIOs, there's no reason you can't just bit bang an LVDS interface to an FPGA. The thing has two 500MHz cores plus a 100MHz MCU, it'll be plenty fast.

No PCIe.
Also, GPIO on embedded CPUs is usually a losing scenario. Unless there's DMA provisions you could be stuck with 10mhz toggle rates.

And you are spot on (from their GPIO section):

To ensure that a pulse is detected by the edge detection register, the pulse should be five clock cycles long.
• 100 ns for a 50 MHz clock when SoC is in S0 state.

This is quite common with microprocessors and microcontrollers that include integrated I/O.  Many current ARM implementations running in the 50 to 200 MHz range have bit-bang I/O rates slower than a 20 MHz PIC.  Only recently has this improved but their I/O still does not operate at core speed.

If you want high throughput I/O, a dedicated function block like a bus interface is needed.