What is CUDA ? And why we need it to perform TensorFlow’s computation with GPU?

CUDA (Compute Unified Device Architecture) is a parallel computing platform and programming model developed by NVIDIA on its own GPUs (graphics processing units). GPU is designed exclusively for number crunching and almost all of real world TensorFlow training data will require the full power of your GPU. CUDA enables developers to speed up compute-intensive applications by harnessing the power of GPUs for the parallelizable part of the computation.

Previously this cores only served the purpose of rendering the Graphics but since High computing requirement has arisen, Nvidia made a way to use this cores in a way GPGPU(General Purpose GPU) Computing.

Now on the question of “Why do we actually need it?”- We need it because it’s an interface of communication between your application/program and GPUs. Nvidia has already implemented library called cuDNN which is only for these kind of applications. So using CUDA will give better performance.

#gpu #ubuntu #machine-learning #data-science #tensorflow