Or none of those.
You see , back in 2004 , Intel had a revelation so to speak.After having problems with Prescott and mounting AMD pressure they had to act.
In typical Andy Grove fashion ( who said paranoia is dead within Intel ? :P ) the company decided to change the rules of the game :
-the future is multicore ( as in dozens/hundreds of simpler cores tightly interconnected )
-asymmetric and heterogeneous cores on the same die
-GPU and CPUs will converge
-x86 everywhere
The results are here :
-Intel Terascale chip ( 80 cores on a single die , TBs of BW)
-die stacking
-Silverthorne/Moorestown ( x86 into the smallest form factors )
-The minicore series ( Intel had 3 ongoing minicore projects Keivet, Keifer and Larrabee - the last was the most promising , the rest got the axe )
Larrabee is very versatile ; has a lot of x86 minicores with a vector core , all coupled to a lot of cache and huge BW.You can use it as a CPU , GPU , accelerator or whatever.
Nehalem has the entire NB on die ( unlike AMD which only included the IMC ) and will also incorporate graphics ( MCM 1st , single die later ).
AMD realized the threat and did the obvious : bought a GPU company in order to get know-how and try to cut Intel's lead by taking the short route. ( Intel builds a graphics team )
If you look at AMD slideware over the past 2 years you'll see where they are going ( same as Intel ) , they need to announce something comparable to Silverthorne and Larrabee.
And now NVIDIA : well , it lacks the most important thing , a CPU division.In other words , NVIDIA is with a gun at its head and has to act quickly and decisively otherwise they're toast.So , I do think NVIDIA will buy AMD.They have to.
By playing the antitrust game with Intel they will get an x86 license.Similarly , Intel will probably want graphic patents.
There is an interview with Fred Webber , former AMD CTO , guy who developed Opteron about all this.
http://www.theregister.co.uk/2008/02...d_weber_scc13/
Bookmarks