Le 30/04/2013 20:11, jhu nous fit lire :
> for $2700! So who's willing to plunk down the cash to test Povray on this thing?
> 
> BTW, it's an addon card with 64 x86 processors on it. Apparently programming
> this thing is the same as programming any other x86 computer.
> 
> Here's some benchmarks comparisons between Xeon Phi and NVidia Tesla:
> 
>  http://goparallel.sourceforge.net/independent-test-xeon-phi-shocks-tesla-gpu/
> 
> 
The parallel test Phi vs Cuda is unrealistic when expanding to povray:
Phi test (and Cuda test) all use many core, local data (small size).

Povray would use many core, global data (huge, the scene): the
bottleneck is going to be the memory bus.

Post a reply to this message