Transcode frames with GPUs on host and guest to reduce PCIe and memory bandwidth requirements at the cost of latency.