Current GPU cloud instance is usually provided in units of a single physical GPU. However, this coarse-grained approach to GPU resource allocation leads to the problem of “resource underutilization”. We propose gRemote, the first open-source framework for remote-resource sharing and isolation in the GPU cloud, as well as offering an elastic and heterogeneous GPU resource pool.
现如今云服务厂商GPU实例的提供粒度太大，导致GPU的利用率低。本项目旨在提供一种细粒度的异构云端GPU共享服务，在提供良好GPU隔离效果的同时，提高GPU的利用率。
Current GPU cloud instance is usually provided in units of a single physical GPU. However, this coarse-grained approach to GPU resource allocation leads to the problem of “resource underutilization”. We propose gRemote, the first open-source framework for remote-resource sharing and isolation in the GPU cloud, as well as offering an elastic and heterogeneous GPU resource pool.