Acceleration of CUDA programs for non-GPU users using cloud

Abstract

The use of Graphics processing unit (GPU) and cloud computing has increased at a higher rate. GPU provides high speed computational power for various applications and accelerates there executional speed by the parallel processing units. Maximum utilization of GPU is enabled by CUDA which is one of the parallel processing model. The power of GPU is effectively utilized by it. Cloud computing on the other hand provides remote nature to access the pool of various computational services on the network. If a network connection is in existence then cloud computing model can allow you to access computer resources and information from anywhere. Cloud computing provides a wide range of shared resources such as data storage space, network, processing power and specialized and specific corporate and user services. Cloud services allows an individual to access hardware and software from a remote location which is managed by a third party. The paper proposes a model which combines these two technologies: Processing CUDA programs on GPU and cloud computing. Due to which non-GPU user can access GPU services and resources remotely on cloud. In the model we combine the processing power of GPU and capabilities of cloud computing. It will also enhance the overall computing speed. The issues of cost, flexibility, scalability will be conquered. The system could also accelerate the overall execution speed of a single application by assigning multiple GPU's to it at a time.