Intel's Xeon Phi

[Valtiel]

Tom's Hardware just posted a fairly thorough article on Intel's Xeon Phi, a derivation of their Larabee research from the past few years.

 

http://www.tomshardware.com/reviews/xeon-phi-larrabee-stampede-hpc,3342.html

 

Obviously it wouldn't be as simple as dropping one of these into your PCIex16 slot and hitting render, as it still requires particular code optimization, but it's nice to daydream of what 60 quad-threaded 1Ghz cores can do for your render times . Being an x86 base, it may not be so far fetched to see production renderers take advantage of this in the coming years (For those that remember, raytracing was one of Larabee's flaunted strengths back when Intel seemed to be directing it toward the GPU market).

[Dimitris Tolios]

It is pretty powerful, and actually cheaper than top of the line K20 Teslas (despite having 8GB of RAM instead of 6).

 

It is highly specialized, and priced according to its capabilities - aka a far cry of what it actually costs to manufacture. They will milk the hi-end super-computer market for a few years, before we get a financially viable solution for consumers, and proper code to take advantage of it.

It is x86 based, but deviates from many of the standard instruction sets - it is not exactly a plug-n-play solution: not a problem for the custom coded models super-computer administrators create, but...

[Slinger]

The #7 fastest machine in the world, Stampede(#1 in open scientific research use) has these in them. Wait till June to see the new numbers when Intel will be upgrading the Phis again.

 

 

http://www.top500.org