Evaluation of GPU resource sharing solutions for virtualized environment based on evaluation metrics

Kurzfassung

The modern GPUs nowadays are immensely powerful and crucial since they have been managing and improving video, and graphics performance, as well as certain common non-graphics applications like AI. In a virtualized environment where nu-merous instances of operating systems, guests, or applications are running simulta-neously, many embedded systems with a single GPU are being used, and therefore the need to virtualize the GPU arose. The processing power of a physical GPU is abstracted and divided into several vir-tual instances by utilizing specialized software to virtualize the GPU. Multiple vir-tual workloads can run simultaneously on the same physical GPU while still run-ning independently of one another. The GPU can conduct numerous autonomous tasks while being safely isolated from one another thanks to virtualization. There has been a considerable amount of research in this area and many solutions have been provided. This thesis work gives an insightful study of these available virtualization techniques and an experimental overview of some GPU virtualization techniques based on evaluation metrics to understand it more deeply.