the aparapi project describes as:
aparapi allows java developers take advantage of compute power of gpu , apu devices executing data parallel code fragments on gpu rather being confined local cpu. converting java bytecode opencl @ runtime , executing on gpu...
i'm wondering benefit offers on traditional concurrency frameworks such gpars or akka.
under circumstances converting jvm bytecode opencl faster or more preferable these alternatives offer? why opencl/gpu paradigm "faster" (at least under circumstances) java/cpu? circumstances warrant doing this?
i this, akka high level abstraction concurrency, whilst stuff cuda, openmp, mpi , on computationally expensive stuff. akka great implementing task scheduling system , handling , throttling requests no great task execution if computationally expensive , parallelized. have high abstraction , can use futures them make whole system reactive not locking anywhere. gpu libraries great implementing massively parallel tasks have heaps of drones doing same thing. that's stuff blas. they've become more popular in recent times think because of larger advancements of graphics cards compared cpus, still specialized area though large learning curve imho oposed openmp can stick few preprocessor derivatives on loops, or java8 new parallel collections. anyway that's 2 cents.
Comments
Post a Comment