I am new to home parallel GPU computing. I want to build a new system with 4 GPUs (asus P6T7 with quad core AMD at 3.2 mhz). I was first unable to believe that parallel processing applications could be done using this, but it has been explained to me. Now, I am looking to see if OpenCL is comparable to the NVIDIA CUDA support. I wish to use 4x Radeon 8570 GPU's (boasting 2 tera flops each. In any case, can the OpenCL compensate (as these are 1/4 the price of Tesla c1070 GPU's with twice the speed). I mainly wish to run software previously run on parallel (8+ CPU's) for molecular dynamics in real space, with associated solute molecules, etc, as well as energy minimization routines. All of this is exhaustivly repetative as far as execution of software goes (the same few sets of constants, 3D space allocations, etc times around 1 million iterations per cycle times around 1 million cycles). I am not the best programmer, but can write small 1 page jimmy riggs for software in C++ and Python, and will most likely be persuing this (ie programming) in the future. Thus, my questions are many fold as well. Pointers to successfull operations in this enviornment for similar data, suggestions, or suggestions in putting the system together are appriciated.
Also, can Python and C++ based programs be integrated and run with the computational power boasted by the GPU comunity? Or, can several different programs (all in C++) be run simulteneously (or a tweak where they are run sequentially a|b|c|d|a....)for x iterations?
I am sure I can tackle this, especially the last questions, but outside feedback is always usefull.
I figure competition in the software and driver (for GPU/CPU integration) are going to be worth alot of money in the near future, especially if somone can just take any type GPU on a motherboard, and use the software/driver.
Stephan Lloyd Riggs