A comparison of precharge policies with modern DRAM architectures

The imbalance between the processor and memory speeds makes memory system design an increasingly important task. This paper evaluates one of the basic design issues of a DRAM memory system: selecting the precharge policy. Open and closed page policies are compared using different applications, cache systems, and DRAM architectures. The results show that the open page policy generally leads to 10%-20% better result than the closed page policy. This paper also shows that increasing the cache size reduces the advantage of the open page policy. Finally, one can see that increasing the number of DRAM banks improves a memory system's ability to utilize the open page policy and the locality in the access stream.