Figure 8 shows write and total DRAM traffic results of SDS for the write-back cache with x8 DRAM chips. We use DRAM chip access per request metric to describe how many chip accesses are removed per memory request. For instance, if four chips out of eight chips are not involved in a write request, we assume that 50% of write traffic is reduced. Since SDS removes only unnecessary write traffic, read traffic is not reduced. With SDS, write traffic is reduced by up to 70% and average 35%. Total traffic including both read and write traffics is reduced by up to 38% and average 16%. Because unmodified data are tracked at finer granularity in the x4 organization than the x8 organization, average 5% more write traffic is reduced. For the write-through cache with the x8 and x4 organizations, average write traffic reductions are 37.5% and 43%, respectively