• reduce CPU utllization, especially when using cuvid decode by setting default cuda schedule mode as "sync" (detailed below).
• add option to control CUDA scheduling mode. (--cuda-schedule <string> (default = "sync"))
  • auto : let cuda driver to decide.
  • spin : CPU will spin when waiting GPU tasks, will provide highest performance but with high CPU utilization.
  • yield : CPU will yield when waiting GPU tasks.
  • sync : CPU will sleep when waiting GPU tasks, performance might drop slightly, while CPU utilization will be lower, especially on HW decode mode.
• fix help.
• NVEncC can now show the total GPU / Video Engine load of the device, using NVML.