- separate i4/r8 code (core functions optimized by v8) - prefer faster methods when convenient - node-based performance tests