I need a software to split a problem into some blocks and gives every computer a share to calculate and gets the result back. The software should use computers' graphic card to solve the problem faster than CPU. CUDA programming or similar languages is needed to the one who wants to do this job. The code for the problem is written already. The only challenge is to manage the resources dynamically to keep the balance of calculation between the peers in the pool. The problem is able to be done in parallel format, no worries for the problem.
I'd put it in the most simple way, Imagine you have a problem to solve and it takes 10 days to be done, we need a server to give a proper amount of shares to the computers in the pool and solve it faster. If there are 10 GPUs in the pool then it should be done in 1 day. Computers can go online and offline so it needs to be monitored as well.
MPI and CUDA-aware MPI is the most relevant fields for this project. I'd like to work with someone who has good experience with computer pools, grids or mining pool in any forms.
Detailed proposal will be sent after the initial negotiation.