I am scientist who develop High performance computing code. I developed applications using
a classical mix mode API : MPI + OpenMP + SIMD. I am going to work on algebra application, since
I discovered the speed up of a matrix product on GPU (X140 speed up), I decided to jump on
the GPU and OpenCL.
I started to program a few basic example, it runs quiet well. I started to associate a mpi/opencl code
on my macbook (Intel core 2 duo + GeForce 9400M) in a "Hello word" mpi/OpenCL, great success.
But I don't know if I could "divide" my gpu by two. If I understand well I have 256 cores in my gpu, so :
1) Could I use 128 cores for one MPI task and 128 other for the second MPI task ?
2) The first MPI task will complete its calculation on the GPU, and the second task will start just after ?
3) Run a MPI/OpenCL code necessitates one CPU core for one GPU ?
Thank you !