Are there anyone experiencing these two models can give me some remarks on them?
I know PGI Accelerate seem easier to work with, yet I’m not sure about the performance (e.g. pros and cos for each one)
Tuan.
I haven’t tried Accelerate much, but as far as CUDA Fortran concerned, I can state that in my initial testing, the efficiency seems on par with CUDA C, at least for the domain I’m interested in (PDE solvers). Furthermore, CUDA Fortran seems to me more worth investing time in learning in, as dominant type of the accelerator at the moment is GPU, and as dominant API for programming these is CUDA (OpenCL, which is supposed to become standard for this, is very similar to CUDA). On the other side, lots of higher-level accelerator programming paradigms are in the development (take Intel Ct as an example), so Accelerate will probably also face lots of competition, once when this type of programming accelerators gain popularity, in that domain (for example, Intel is basing Ct on the RapidMind work in that area, and these guys were already for several years in doing that kind of development, while PGI is just starting in al lof this…).