I agree with Frank, the relevance of energy efficiency cannot be overemphasized.
Transistor technology for ultralow power in the sub-28nm scale will eventually be available to general processes, and ARM will have physical IP characterized at these processes, so that is not a long-term edge (for Intel).
On the other hand, the mobile market is a SoC market, not a CPU chipset market. And ARM has a much more agile business model on this SoC arena, and is building quite a momentum.
Currently, ARM is both process driver for energy efficiency and architectural mainstream for embedded systems.
Hard to beat in the foreseeable future.
I agree, 'Grove's law' would stipulate that 'Intel inside' for smartphones would have to have some compelling 10X improvement to displace the incumbent. Also mobiles are not about lower power only, they're very much about lower margins, an area where historically Intel has declined to play with any gusto.
But what adavantage would an "Intel inside" mobile have over an "ARM inside" mobile if the Intel-based device only lasts at least as long as today's units?
I have to disagree with you on this point -- mobile devices will continue to be pushed to lower and lower power with each generation, and having the longest battery life will still be a competitive advantage years from now.
I don't think that Intel really has to "match" ARM's lower power. They just need to be low enough power so that handhelds with "Intel Inside" will last as least as long as today's units.
Will it make a huge difference if Intel-based handhelds last just a week while ARM lasts 2 weeks ? I don't think so. At that point, it'll come down to cost, performance and features.
When Intel has their 22nm units in production with 20x less standby power, as well as being smaller and cheaper, it'll be a much more interesting matchup. 14nm with even lower power and cost follows 2 years later.
Yep, ARM certainly has the advantage of being the incumbent in the mobile market, and Intel will find it very difficult to make a dent in the firm's market share. I like your tortoise and the hare analogy too. I will mull that over tonight :)
This indeed not at all new, NXP is effectively doing the same thing lower down the food chain with its M4/M0 asymmetric dual core approach. ARM does not need to threaten Intel - they'll prevail because they have the strength of the market behind them. Intel OTOH can't quickly kick the addiction its developed to the high margins on the cash cow it has in the PC space. The tortoise beat the hare remember but the tortoise wasn't perceived as any kind of threat....
Thanks Alex, I appreciate that. Your comment clarified the issue.
You're of course correct that it's not "news" some companies will use their own custom power solutions rather than ARM's, which is why I opted to write this as more of a blog/commentary.
I do agree with you about smaller companies being able to use big.little as an equalizer, though. Qualcomm and Nvidia have long been investing in and playing around with power "fixes", so this gives the other guys with less engineers and smaller R&D budgets a chance to catch up.
Qualcomm have chosen to asynch their cores with independent clocks for each core in cluster. Marvell actually did a Tegra3 in their Armada628 before nvidia white papered their fifth core. big.LITTLE is better than those solutions because it uses different micro architectures; however, nothing precludes partners ALSO using low power processes on one of the clusters AS WELL, gaining further advantage from the scheme. The key here is in the ability to switch clusters transparently - allowing power hating of the one that isn't being used.
The fact that some ARM competitor said some people wouldn't use big.LITTLE is not really news; some companies have their own solutions, spreading innovation further. Now the companies that can't afford to innovate with the likes of Qualcomm can still pull off these tricks too. And who cares what competitors say anyway - they would say those things!
A Book For All Reasons Bernard Cole1 Comment Robert Oshana's recent book "Software Engineering for Embedded Systems (Newnes/Elsevier)," written and edited with Mark Kraeling, is a 'book for all reasons.' At almost 1,200 pages, it ...