Cache coherent shared memory multiprocessors are an attractive and available target for parallel multi-threaded applications. However, achieving the expected levels of performance has proven difficult. ccNUMA permance depends critically on memory and task allocation, and by the amount and type of the coherency transactions. No performance analysis tool to date has done an adequate job of providing high fidelity information about these application memory performance issues. This paper presents a new memory profiling tool called snperf which does provide such information for SGI Origin systems.