Doug S (foo.delete@this.bar.bar) on September 26, 2013 11:51 am wrote:
> Michael S (already5chosen.delete@this.yahoo.com) on September 26, 2013 10:52 am wrote:
> > For the tasks that people today find CPU limited, high L1I cache hit rate is rather typical.
> >
>
>
> Can you provide some examples of tasks for which people are CPU limited on phones today?

Two obvious examples are heavy web pages and PDF decode of sophisticated pages. These are two problems that
- affect a lot of people every day
- appear to be amenable to only the most minimal threading
- are still slow enough to be irritating

(Admittedly I run these on an iPad3 --- A5 and iPhone 5 --- A6 so it's possible that on an A7 system they've crossed to the point where they're no longer worth worrying about.)

In the future we'll want our devices to do things like recognize images for us, and while most of the heavy lifting will be done on servers, there's probably some degree of pre-processing we will want done on the phone, like running SIFT over each video frame. On the other hand, that's the sort of task that may be delegated first to the GPU then to dedicated HW.

One use case that I personally think has a lot of potential once someone figures out how to do it properly is (someone like the above case) live modification of video. Remember the old Apple gimmick when they introduced iChat of sticking your face in front of an aquarium or the Eiffel Tower or whatever. It never worked very well. But a version of that which DOES work well has potential --- in multiple directions.
(a) Map the face motion onto a skeleton (like the MPEG-4 facial animation skeleton) and send that over the network. Lower bandwidth PLUS
(b) At the remote end you can reskin the facial animation as sender or receiver like. Transmit yourself as Britney Spears. Map your manager onto Michael Scott. Basically a much fancier version of smilies, stickers, and the whole Line economy.
(And I'd like to point out that I thought of, and sent to Apple, a recommendation that they should sell packs of cute animated emoticons in the App store, for use in Messages.app, long before Line came long. Obviously I was ignored... :-( )
(c) Beyond the fun you can have with the face, you can likewise have fun with the non-face (ie background), as before projecting yourself into space, the ocean, another city, a video game, etc.

To get something like this to work as many moving pieces. Each part of the tech has to be very good (good enough that glitches which destroy the illusion are rare, not visibly obvious on every frame). Also the set of "scenarios" which are provided has to be attractive, artistic, and immediately appealing to many users --- simply providing raw tech and assuming users will by themselves figure out interesting faces and backgrounds to fill in is not going to work, you have to provide some hand-holding at first to show everyone what's possible.

I agree that it's an interesting problem, and I'm not much interested in the standard answer of "games". I'd add to it the second problem of tasks on phones which make good use of multiple cores.