AMD heeft een nieuw patent geregistreerd met de titel "Prefetching data using profile of cache misses from earlier code executions. Afgaande op onderstaande omschrijving lijkt het erop dat AMD een methode heeft uitgedokterd waarmee cache-misses verlaagd kunnen worden:
During execution of a code sequence, a profile is generated containing addresses of the data cache misses experienced during the execution. The profile is associated with the code sequence such that, during a future execution of the code sequence, the profile is available. Prefetching may be performed, based on the profile. Since the profile records a sequence of miss addresses, an arbitrarily complex miss pattern can be prefetched. In one embodiment, multiple profiles may be associated with a code sequence having multiple entry points (e.g. multiple instructions at which execution may begin within the code sequence). When the code sequence is executed, the profile associated with the entry point of the current execution may be selected. Additionally, a new profile may be generated. If the entry point of the current execution is not associated with a profile, the new profile is saved. If the entry point of the current execution is associated with a profile, the new profile is saved if the current profile is found to be ineffective.