While the portable, pure C implementation of BLAKE2 that is used almost exclusively by libraries (e.g. Python, OpenSSL) is already faster than MD5... ...there

While the portable, pure C implementation of BLAKE2 that is used almost exclusively by libraries (e.g. Python, OpenSSL) is already faster than MD5... ...there are even faster implementations of BLAKE2 around. E.g. Samuel Neves' AVX2 implementation is almost 40 % faster on Haswell. libsodium ships a different AVX2 implementation (considered experimental AFAIK by its original author, so I find that a bit strange).