Oh, we do also heavily parallelize. This blog post is just focusing on the single-core perf.