Graphics processors

A graphics processor (2) comprises a programmable execution unit (65) (e.g. shader core) executing programs to perform graphics processing operations. The graphics processor further comprises a machine learning (ML) processing circuit (neural processing accelerator, NPA 78) performing processing operations for ML processing tasks. The ML processing circuit (which may include multiply-and-accumulate circuits, MAC) is in communication with the programmable execution unit internally in the graphics processor. The graphics processor is thus configured such that ML processing tasks can be performed by the programmable execution unit, the ML processing circuit, or a combination of both. The processor further comprises storage that is available for storing data for ML processing tasks and that is accessible to both the programmable execution unit and the ML processing circuit. The different units can message each other to control the delegation of processing tasks. The processor may have an external memory interface including a cache system that is shared between the programmable execution unit and the ML processing circuit. The ML processing circuit may have dedicated local storage for storing a neural network (NN) to be executed. The ML circuit may comprise a queue or cache (e.g. translation lookaside buffer) for buffering ML processing task requests.