Frames of a virtual desktop are encoded using a hybrid approach that combines the strength of software encoding by a central processing unit (CPU) and hardware encoding by a graphics processing unit (GPU). A method of encoding frame data of one or more virtual desktops in hardware and in software and transmitting the encoded frame data to one or more client devices, includes the steps of encoding a first portion of the frame data in the GPU to generate a first encoded frame data, encoding a second portion of the frame data in software, i.e., programmed CPU, during encoding of the first portion, to generate a second encoded frame data, and transmitting the first encoded frame data and the second encoded frame data from a host computer of the one or more virtual desktops to the one or more client devices as separate video streams.