News Intel officially drops Curie, Arduino 101 embedded ranges

Read more

 

They were proper rubbish. (I blew mine up 'cos when I'd last packed it away I'd somehow put a 12V PSU in with it, and it didn't like that. Imagine my annoyance when one of the improvements of the Galileo Gen. 2 was "doesn't blow up when you feed it 12V!")

Seriously, though, the Quark blew at real-time processing tasks - s'why Intel ended up switching the Edison to an Atom then the Curie to a Quark-plus-shh-don't-mention-the-ARC (with all your actual code running on the ARC). I didn't give it a proper GPIO benchmark (I didn't have a frequency counter when I reviewed it originally, and it's too dead for me to try again now) but I did try running my Sleepduino Shield on it and it sounded like it was running at about 5% of the speed of an Arduino Uno.

 

There's little doubt that if Intel wants to grow it desperately needs to find something other than mainstream and server computing, 'cos it has those markets sewn up. Embedded is a biggie, but Intel has long struggled to unseat ARM there - and its attempts to get in the back door via education by targeting the maker market have, as we've seen, failed miserably. It had a good ol' go at mobile, but that failed too - and, again, to ARM.

As to whether it's now pulling back on its embedded stuff in order to concentrate on fighting AMD, I think that's unlikely simply from the numbers. AMD spends around $279 million a quarter on research and development for CPU and graphics; Intel spends $3,326 million for the same period. In other words, Intel could blow $3 billion on R&D for mobile, embedded, and other toy projects and still be vastly outspending AMD on mainstream CPU R&D.

And I really don't think that Quark took $3 billion to develop, as it's effectively the result of somebody saying "hey, guys, what would happen if we took the original Pentium chip and built it on 32nm instead of 800nm?"

 

You were looking at around 2.2W for the original 400MHz family, which sounds acebest - until you realise that it can be outclassed by an ARM Cortex-A9 chip for most embedded tasks, which gets you two cores to the Quark's single core running at 800MHz to 2GHz to the Quark's 400MHz and at 0.5W to 1.9W power draw to the Quark's 2.2W. (Figures courtesy of this presentation, PDF warning.)

The main failing, though, was that Intel went into it with the mindset of "x86 is king, everyone will overlook crappy performance if they can program in nice friendly familiar x86" - which is great when x86 holds the overwhelming majority of the market as in desktop and server but doesn't really work when you're up against a market where ARM (and, to a lesser extent, MIPS and others) has been holding sway for lo these many years. "Learn x86," all the embedded engineers thought, "for crappier performance and a higher power draw? Nuts to that."

And, thus, here we are.