P.S. my table above was produced using the Throughput benchmark mode, so the higher the better (as opposing to the figures posted by Andrej, but both confirming that IHM seems the better one)