An analytical model for a GPU architecture with memory-level and thread-level parallelism awareness