A method and an apparatus for providing requested data to a pipeline processor. A pipeline processor in a graphics computer system is provided with a data caching mechanism which supplies requested data to one of the stages in the pipeline processor after a request from a prior stage in the pipeline processor. With the sequential nature of the pipeline processor, a prior stage which knows in advance the data which will be requested by a subsequent stage can make a memory request to the data caching mechanism. When processing reaches the subsequent stage in the pipeline processor, the displayed data caching mechanism provides the requested data to the subsequent processing stage with minimal or no lag time from memory access. In addition, the data caching mechanism includes an adaptive cache memory which is optimized to provide maximum performance based on the particular mode in which the associated pipeline processor is operating. Furthermore, the adaptive cache includes an intelligent replacement policy based on a direction in which data is being read from memory as well as the particular mode in which the associated pipeline processor is operating.