GPUs are more known for parallel computing in contrast to CPU (sequential processing where parallelization is possible using cores). GPUs are mostly used in machine learning (Especially in Deep learning) where the extensive amount of high-dimensional matrix (tensors) computation is needed. So GPUs are used where parallel computation is needed. Like while playing games, GPU renders graphics processing pixels parallelly. So yes GPUs are used in Big Data processing. But that is not a restriction. MapReduce parallelizes the computation using CPUs by distributing jobs to CPUs (client nodes). 

see the followiung URLs:

http://www.nvidia.com/object/caffe-deep-learning-framework.html

or

http://caffe.berkeleyvision.org/

GPUs are generally used for parallel computations; therefore, they could be used to process large amount of data. However, big data platforms, such as MapReduce, Spark, etc., splits and distributes the big data across parallel computing nodes.

Mostly distributed computing based platforms like Spark or Flink are considered as a big data platform. GPU's came under parallel processing and use high-cost infrastructure. Big data platforms are generally considered to use cheap commodity hardware. Therefore, I don't think GPU's are big data platforms.