Many prior studies have investigated exploring data value characteristics for better performance and power optimizations. The notion of silent stores is proposed in [13]; silent stores write the same values as the old ones already stored. Empirical evidences that silent stores constitute a non-trivial percentage of overall stores are provided in [3, 13]. Since detecting silent stores requires comparison for data differences, detecting and squashing silent stores for free while generating ECC are proposed in [14]. In our proposed scheme, CWV also exploits the same ECC generation logic for detecting actual modification of data. The presence of narrow-width data has also been utilized for performance and power optimizations as well as reliability [5, 24]. SDS benefits from above value characteristics to reduce effective DRAM write traffic and energy consumption since detecting actual modification of data at byte or nibble level granularity naturally exploits such value characteristics. This is another contribution of our work that evaluates such value characteristic and shows experimental evidence at DRAM chip level.