anyone knows how it compares in costs? when they produce the chips themselves does it make a difference if they put 1,2,4,20 cores there?
Neither Apple nor Nvidia makes the chips. Apple makes practically none of their products.
I think Nvidia is using TSMC as a foundry on Tegra2. They actually make the chips. It is their processes that get leveraged.
The EE Times referenced in the article that kick-stared the thread suggests that Apple is using Samsung as a foundry.
The costs are in the design team that composes the design. What "blocks" you license from other folks. The licensing fees for IP incorporated and the fees to the foundry for "printing" and packaging the chips.
Apple was doing similar process previously by having Samsung do the tweaking and the "printing".
As for how many.... really depends upon how big the cores are.
Nvidia Tegra2 (scroll down to the digram )
http://www.anandtech.com/gadgets/showdoc.aspx?i=3714
On their SoC there is two A9 cores , a image processor , an A7 core , a Video (H.264) core, a audio core , a GPU (with multiple relatively small cores).
All of that is on one piece of silicon. Dump a bunch of the x86 cruft and can fit lost more in the same space. It isn't what A9 class cores that apple may have put on here but perhaps some specialized processors that are just way better at some specific task than a generic computational core. Video and graphics cores respectively are
way more efficient at doing computations they are built for than a general purpose CPU is.
It isn't really a question of how many. It is how many you need and of what type and what is the objective (performance , performance/Watt, etc. ). Also if collapsing cores that used to be in different chip packages (containers with pins on them) then may be saving space in the finished product but taking up more space on an individual wafer. The packages these dies fit into are typically much bigger than the dies themselves.
or is it just getting more expensive when the die is getting bigger and less fit on each waver?
Bigger wafer usually means better costs.
More is fitting on the wafers because the imaging processes are getting smaller.
The other swing in costs is going to be at what width "printing" at. A mature 45nm process per wafer in a plant that has had that equipment for a year or so is going to be more affordable than a plant with a brand new 28nm process. You get a bit of a bang for the buck cause get more out of each wafer. However, with those recent multimillion dollar bills for the new equipment at that level you get charged more too. A brand new, first class 28nm fab facility will put you back around a billion dollars. So fewer and fewer folks build/own these things and more and more folks "share" by hiring folks to do runs of their stuff for them.