Abstract: Speculative multithread architecture （SMA） combines both benefits of speculative execution and multithreading architecture. Performance factors are discussed in this paper. In the SMA model, there are three key factors: contexts load imbalance, inter-thread control speculation failure, and inter-thread data speculation failure. To achieve high instruction level parallelism, some optimization heuristic rules are introduced with emphasis on the design of per-thread dynamic profiling mechanism. An optimization framework prototype is built on the basis of those heuristic rules and mechanisms. Benchmark simulation of SMA optimization framework shows the satisfying results.