To achieve this goal, we focus on compare instructions, mainly targeting table/index scan in in-memory databases [5]. For example, table scan depicted in Fig. 1 searches for a specific data in the given table. With a conventional method, the data are read from the memory to the host processor, and compares are performed. On the contrary, by executing compare operations of table scan at the memory side, most of the data can be consumed inside the memory, only requiring the internal bandwidth and reducing the off-chip channel access. Similar patterns are found in many big data applications which can be characterized by large footprint, almost no locality, and high memory-boundedness. In those applications, we identify a compare-n-op pattern, which performs a compare and an additional operation over a bounded set of data.