A mechanism is described for facilitating intelligent thread scheduling at autonomous machines. A method of embodiments, as described herein, includes detecting dependency information relating to a plurality of threads corresponding to a plurality of workloads associated with tasks relating to a processor including a graphics processor. The method may further include generating a tree of thread groups based on the dependency information, where each thread group includes multiple threads, and scheduling one or more of the thread groups associated a similar dependency to avoid dependency conflicts.