Detail

Framework for a balanced use of a CPU/GPU server configuration

In the context of a MSc thesis centered on using GPUs to classify the contents of web pages, a concurrent server that resides in the GPU and offers services related with the processing of relevant sequences of words present was developed.
The framework makes an integrated management of the CPU and GPU resources. The software tries to achieve full occupancy of the GPU, but also considers the overheads associated to the transfers between CPU and GPU. In the context of the above mentioned thesis, this framework achieved throughputs that were more than the double of the ones of a CPU only version.