Default timings with an old but stable Intel i5-2300 2.8GHz desktop PC:

Hash function MiB/sec cycl./hash cycl./map size Quality problems
donothing32 15316474.36 6.00 - 13 test NOP
donothing64 15330019.19 6.00 - 13 test NOP
donothing128 15278983.09 6.00 - 13 test NOP
NOP_OAAT_read64 28467.50 18.48 - 47 test NOP
BadHash 524.81 96.20 - 47 test FAIL
sumhash 7169.08 27.12 - 363 test FAIL
sumhash32 22556.18 22.98 - 863 UB, test FAIL
multiply_shift 5418.36 28.69 157.11 (3) 345 fails all tests
pair_multiply_shift 3883.85 36.89 157.83 (4) 609 fails all tests

crc32 392.10 131.62 204.58 (4) 422 insecure, 8590x collisions, distrib
md5_32a 353.64 629.85 799.56 (9) 4419 8590x collisions, distrib
sha1_32a 361.42 953.56 1120.35 (6) 5126 collisions, 36.6% distrib
md5-128 353.76 638.29 803.39 (9) 4419
sha1-160 364.35 948.20 1176.86 (5) 5126
sha2-224 145.46 1349.96 1576.16 (9)
sha2-224_64 147.77 1344.30 1568.74 (9)
sha2-256 148.82 1330.24 1530.96 (15) Moment Chi2 4
sha2-256_64 149.50 1329.86 1561.46 (9) Moment Chi2 7
sha1ni 2019.96 135.84 564.40 (6) 989 insecure,sanity, Permutation, Zeroes, amd epyc only
sha1ni_32 2019.94 136.82 589.46 (1) 989 insecure,sanity, Permutation, Zeroes, TwoBytes, amd epyc only
sha2ni-256 1906.77 145.47 603.08 (22) 4241 insecure,sanity, Permutation, Zeroes, amd epyc only
sha2ni-256_64 1920.36 145.47 603.08 (2) 4241 insecure,sanity, Permutation, Zeroes, TwoBytes, amd epyc only
blake3_c 1285.91 340.01 552.63 (3) no 32bit portability
rmd128 334.36 659.03 838.32 (9)
rmd160 210.94 978.93 1160.56 (10)
rmd256 362.49 617.02 815.44 (8)
blake2s-128 221.08 974.71 1158.36 (8)
blake2s-160 221.06 984.46 1162.95 (9)
blake2s-224 221.07 984.22 1154.56 (7)
blake2s-256 221.07 976.66 1144.27 (9)
blake2s-256_64 221.05 1046.08 1234.16 (7)
blake2b-160 368.85 1170.06 1358.93 (9)
blake2b-224 368.85 1171.48 1357.17 (11)
blake2b-256 369.09 1157.20 1335.65 (8) Sparse high 32-bit
blake2b-256_64 368.84 1198.03 1363.79 (11)
asconhashv12 86.73 1491.01 1413.06 (6) 6490
asconhashv12_64 86.73 684.02 606.93 (6) 4341
sha3-256 93.92 4143.17 4227.90 (10)
sha3-256_64 90.45 4231.05 4223.33 (13)
hasshe2 2372.52 68.64 216.74 (5) 445 Permutation,TwoBytes,Zeroes,Seed
poly_1_mersenne 1431.65 54.49 189.52 (3) 479 fails most tests
poly_2_mersenne 1323.69 66.93 190.88 (2) 479
poly_3_mersenne 1323.59 74.86 206.77 (2) 479
poly_4_mersenne 1323.57 82.67 200.36 (4) 479
tabulation32 4317.34 35.45 197.20 (2) 848 collisions
tabulation 7621.75 42.19 179.93 (2) 554
crc32_hw 6330.42 35.55 170.16 (1) 653 insecure, 100% bias, collisions, distrib, machine-specific (SSE4.2/NEON)
crc32_hw1 23208.73 46.74 179.70 (2) 671 insecure, 100% bias, collisions, distrib, machine-specific (x86 SSE4.2)
crc64_hw 8440.13 34.94 141.15 (2) 652 insecure, 100% bias, collisions, distrib, machine-specific (SSE4.2/NEON)
crc32_pclmul 1972140.38 7.00 - 481 insecure, 100% bias, collisions, distrib, machine-specific (x86 PCLMUL)
o1hash 12439661.09 16.77 166.13 (1) 101 insecure, zeros, fails all tests
fibonacci 11339.87 26.33 705.64 (2) 1692 UB, zeros, fails all tests
FNV1a 791.84 69.69 177.84 (2) 204 zeros, fails all tests
FNV1A_Totenschiff 6258.23 27.99 198.20 (1) 270 UB, zeros, fails all tests
FNV1A_Pippip_Yurii 6258.46 28.19 184.41 (2) 147 UB, sanity, fails all tests
FNV1a_YT 9643.42 32.06 175.19 (2) 321 UB, fails all tests
FNV2 6258.84 33.25 142.89 (1) 278 fails all tests
FNV64 791.82 70.24 159.29 (1) 79 fails all tests
fletcher2 12011.15 25.29 298.60 (1) 248 UB, fails all tests
fletcher4 11928.55 25.27 293.49 (2) 371 UB, fails all tests
bernstein 791.82 68.63 180.71 (2) 41 fails all tests
sdbm 791.84 67.69 177.06 (2) 41 fails all tests
x17 527.90 98.78 184.09 (2) 79 99.98% bias, fails all tests
JenkinsOOAT 452.48 142.85 213.93 (2) 153 53.5% bias, fails all tests
JenkinsOOAT_perl 452.49 118.78 194.78 (1) 65 1.5-11.5% bias, 7.2x collisions, LongNeighbors
MicroOAAT 977.60 59.61 185.06 (2) 68 100% bias, distrib
beamsplitter 574.48 824.57 904.89 (10) UB, too many bad seeds
BEBB4185 2951.62 222.03 343.63 (4) 1294 UB, too many bad seeds, MSVC specific
pearsonhash64 287.95 174.11 196.50 (1) Avalanche, Seed, SSSE3 only. broken MSVC
pearsonhash128 287.95 171.72 194.61 (1) Avalanche, Seed, SSSE3 only. broken MSVC
pearsonhash256 264.51 184.87 218.79 (0) Avalanche, Seed, SSSE3 only. broken MSVC
VHASH_32 9404.99 77.01 250.57 (2) 1231 sanity, Seed, MomentChi2
VHASH_64 9392.39 74.72 227.92 (2) 1231 sanity, Seed, Sparse
farsh32 14053.09 74.29 245.33 (3) 944 insecure: AppendedZeroes, collisions+bias, MomentChi2, LongNeighbors
farsh64 7216.29 130.30 302.44 (3) 944 insecure: AppendedZeroes, collisions+bias, MomentChi2, LongNeighbors
farsh128 3776.92 232.48 398.67 (6) ??
farsh256 1895.77 459.86 575.95 (3) ??
jodyhash32 1428.37 44.36 185.85 (3) 102 bias, collisions, distr, LongNeighbors
jodyhash64 2848.42 29.99 164.36 (1) 118 bias, collisions, distr, LongNeighbors
lookup3 1658.31 48.84 194.15 (2) 341 UB, 28% bias, collisions, 30% distr
superfast 1956.25 53.61 180.10 (3) 210 UB, bad seed 0, 91% bias, 5273.01x collisions, 37% distr, BIC
MurmurOAAT 452.49 113.07 197.83 (2) 47 collisions, 99.998% distr., BIC, bad seed 0, LongNeighbors
Crap8 3149.63 36.23 195.11 (1) 342 UB, 2.42% bias, collisions, 2% distrib
Murmur1 1804.67 51.51 188.41 (1) UB, fails all tests, 1 bad seed
Murmur2 3146.91 41.87 187.89 (2) 358 UB, 1 bad seed, 1.7% bias, 81x coll, 1.7% distrib, BIC
Murmur2A 3146.79 46.87 191.96 (4) 407 UB, 1 bad seed, 12.7% bias, LongNeighbors
Murmur2B 4882.95 39.72 149.43 (2) 433 UB, 1.8% bias, collisions, 3.4% distrib, BIC
Murmur2C 4092.99 51.84 164.65 (2) 444 UB, 2^32 bad seeds, 91% bias, collisions, distr, BIC, LongNeighbors
Murmur3A 2413.88 53.36 182.37 (3) 351 UB, Moment Chi2 69
PMurHash32 2344.78 58.48 196.43 (4) 1862 Moment Chi2 69
Murmur3C 3197.63 67.90 198.00 (2) 859 UB, LongNeighbors, DiffDist
PMPML_32 6704.53 53.50 197.43 (4) 1084 Avalanche >512, unseeded: Seed, MomentChi2
PMPML_64 8161.19 53.20 179.16 (2) 1305 unseeded: Seed, MomentChi2
xxHash32 6040.87 51.77 177.91 (4) 738 LongNeighbors, collisions with 4bit diff, MomentChi2 220
metrohash64_1 9664.57 45.37 152.31 (2) 624 UB, LongNeighbors, BIC, MomentChi2
metrohash64_2 9668.51 44.45 164.30 (2) 627 UB, LongNeighbors
metrohash64crc_1 14000.50 49.08 150.54 (2) 632 UB, cyclic collisions 8 byte, BIC, MomentChi2, machine-specific (SSE4.2/NEON)
metrohash64crc_2 14034.84 48.94 162.54 (2) 632 UB, cyclic collisions 8 byte, BIC, machine-specific (SSE4.2/NEON)
metrohash128_1 9558.17 59.04 175.94 (2) 773 UB, LongNeighbors
metrohash128_2 9584.94 59.10 167.43 (2) 773 UB, LongNeighbors
cmetrohash64_1o 9658.31 42.84 163.45 (1) 3506 LongNeighbors, MomentChi2
cmetrohash64_1 9683.33 45.20 161.01 (2) 652 LongNeighbors, BIC, MomentChi2
cmetrohash64_2 9669.95 44.75 149.67 (2) 655 LongNeighbors
City64noSeed 9090.42 32.23 171.53 (3) 1038 Avalanche, Sparse, TwoBytes, MomentChi2, Seed
City64 9066.90 47.81 197.78 (1) 1120 Sparse, TwoBytes
t1ha1_64be 7481.37 38.16 193.22 (2) 555 Avalanche
t1ha0_32le 5132.18 54.83 193.53 (2) 509 Sparse, LongNeighbors
t1ha0_32be 4585.59 55.98 183.45 (2) 533 Sparse, LongNeighbors
t1ha2_stream 9068.55 74.56 219.85 (6) 1665 Sparse, Permutation, LongNeighbors
t1ha2_stream128 9065.50 93.19 236.50 (3) 1665 Sparse, Permutation, LongNeighbors
aesnihash 2963.39 71.24 217.73 (3) 1209 fails many tests, machine-specific (x64 AES-NI)
falkhash 20202.42 173.63 321.52 (2) 264 Sparse, LongNeighbors, machine-specific (x64 AES-NI)
MeowHash 17371.91 85.48 247.96 (2) 1764 Sparse, machine-specific (x64 AES-NI)
MeowHash64low 17378.06 85.48 237.60 (2) 1764 Sparse, machine-specific (x64 AES-NI)
MeowHash32low 17374.64 85.48 258.53 (2) 1764 Sparse, machine-specific (x64 AES-NI)
t1ha1_64le 9723.86 34.39 176.91 (1) 517 Avalanche
tifuhash_64 35.84 1553.56 1081.57 (10) 276
floppsyhash 35.72 1870.83 1399.26 (10) 623
chaskey 753.23 153.42 288.26 (2) 1609 PerlinNoise
SipHash 980.88 127.77 246.19 (4) 1071
HalfSipHash 755.78 114.47 243.72 (4) 700 zeroes
GoodOAAT 1052.00 71.62 192.19 (1) 237
pearsonbhash64 1486.34 104.32 185.03 (2) 683
pearsonbhash128 1347.03 121.75 214.84 (2) 1134
pearsonbhash256 998.90 167.05 261.29 (3) 844
prvhash64_64m 2308.85 53.03 241.00 (5) 349
prvhash64_64 2302.42 53.31 253.73 (5) 384
prvhash64_128 2296.75 106.79 323.24 (5) 718
prvhash64s_64 2455.28 369.19 624.65 (18) 2640
prvhash64s_128 2462.49 471.74 697.91 (19) 2799
SipHash13 1889.10 89.00 199.95 (4) 778 0.9% bias
TSip 3228.14 57.96 211.71 (3) 519 !msvc
seahash 4796.97 58.55 201.58 (2) 871 PerlinNoise, !msvc
seahash32low 4801.33 58.54 227.31 (4) 871 PerlinNoise, !msvc
clhash 4472.31 82.72 229.73 (3) 1809 PerlinNoise, machine-specific (SSE4.2/NEON)
HighwayHash64 6242.58 99.55 248.41 (3) 2546
Murmur3F 5226.40 52.18 175.85 (1) 699 UB
fasthash32 4737.61 45.32 181.86 (2) 566 UB
fasthash64 4737.21 42.79 164.87 (2) 509 UB, Moment Chi2 5159 !
MUM 7134.56 37.85 172.34 (1) 1912 UB, too many bad seeds, machine-specific (32/64 differs)
MUMlow 7225.18 37.85 197.92 (3) 1912 UB, 5 bad seeds
mirhash 5413.73 39.68 154.47 (3) 1112 2^36 bad seeds, UB, LongNeighbors, machine-specific (32/64 differs)
mirhash32low 5412.76 39.79 182.13 (3) 1112 4 bad seeds, UB, Cyclic, LongNeighbors, machine-specific (32/64 differs)
mirhashstrict 2217.32 65.53 182.07 (2) 1112
mirhashstrict32low 2218.87 65.48 190.59 (4) 1112 1 bad seed, MomentChi2 9
mx3 6146.02 52.48 173.09 (3) 734 UB
pengyhash 8744.48 85.31 222.45 (4) 421
City32 3675.04 57.73 212.04 (3) 1319
City64low 9089.45 47.75 201.73 (1) 1120
City128 9640.19 88.45 225.38 (3) 1841
CityCrc128 12343.43 74.50 209.75 (2) 295
FarmHash32 17112.05 47.70 214.71 (2) 11489 machine-specific (x64 SSE4/AVX/NEON)
FarmHash64 8684.76 48.13 200.51 (4) 3758
FarmHash128 9409.63 74.52 210.25 (3) 163
farmhash32_c 16299.81 47.79 219.19 (4) 762 machine-specific (x64 SSE4/AVX/NEON)
farmhash64_c 8713.16 47.96 201.00 (2) 3688
farmhash128_c 9244.13 79.08 209.44 (2) 1890
metrohash64 9664.61 44.59 150.74 (2) 624 LongNeighbors
metrohash128 9569.16 58.68 167.53 (2) 624 UB
metrohash128crc_1 13948.67 65.20 168.08 (2) 723 UB, machine-specific (SSE4.2/NEON)
metrohash128crc_2 13920.19 65.12 176.70 (1) 723 UB, machine-specific (SSE4.2/NEON)
xxHash64 8936.63 51.31 174.34 (3) 1999
Spooky32 9747.13 62.24 196.96 (4) 2221 UB
Spooky64 9747.47 62.20 191.71 (2) 2221 UB
Spooky128 9751.14 63.84 192.47 (2) 2221 UB
SpookyV2_32 9529.38 64.36 194.47 (1) 2069
SpookyV2_64 9505.69 64.44 192.19 (1) 2069
SpookyV2_128 9524.31 65.81 187.24 (1) 2069
t1ha2_atonce 9237.12 38.94 194.32 (2) 541
t1ha2_atonce128 8350.99 55.65 203.53 (4) 613 LongNeighbors
t1ha0_aes_noavx 22785.26 38.71 180.61 (3) 925 LongNeighbors, machine-specific (x86 AES-NI)
t1ha0_aes_avx1 22598.11 38.71 177.23 (1) 843 LongNeighbors, machine-specific (x64 AVX)
t1ha0_aes_avx2 22345.33 44.38 556.47 (89) 792 LongNeighbors, machine-specific (x64 AVX2)
ahash64 10923.49 31.33 528.54 (14) 412 rust
xxh3 16538.52 32.81 184.86 (2) 744 Moment Chi2 14974, BIC
xxh3low 16462.36 32.77 199.79 (2) 756 Moment Chi2 1.8e+9 !
xxh128 15174.14 40.46 195.65 (2) 1012 Moment Chi2 14974
xxh128low 15174.85 33.79 187.05 (2) 1012 Moment Chi2 14974, BIC
MeowHash 17371.91 85.48 247.96 (2) 1764 Sparse low32, machine-specific (x64 AES-NI)
MeowHash32low 17374.64 85.48 258.53 (2) 1764 Sparse, machine-specific (x64 AES-NI)
wyhash 12879.00 30.35 196.77 (2) 474 2^33 bad seeds
wyhash32 2532.89 37.95 222.17 (4) 426 2 bad seeds, 32-bit
wyhash32low 12911.09 29.59 205.43 (2) 474 2 bad seeds
umash32 4633.19 53.42 216.33 (3) 1530
umash32_hi 4662.92 54.22 214.20 (2) 1530
umash64 4662.09 53.42 188.09 (1) 1530
umash128 2427.46 70.60 197.29 (2) 1530
halftime_hash64 4990.72 120.55 281.64 (3) 1530
halftime_hash128 13478.23 97.79 252.14 (2) 1530
halftime_hash256 11620.28 98.44 252.60 (2) 1530
halftime_hash512 7681.62 125.81 274.01 (3) 1530
nmhash32 7850.01 56.74 207.59 (1) 2445
nmhash32x 7850.92 46.94 193.94 (1) 1494
Other timings:

Summary

I added some SSE assisted hashes and fast intel/arm CRC32-C and AES HW variants, but not the fastest crcutil yet. See our crcutil results. See also the old https://code.google.com/p/smhasher/w/list.

So the fastest hash functions on x86_64 without quality problems are:

Hash functions for symbol tables or hash tables typically use 32 bit hashes, for databases, file systems and file checksums typically 64 or 128bit, for crypto now starting with 256 bit.

Typical median key size in perl5 is 20, the most common 4. Similar for all other dynamic languages. See github.com/rurban/perl-hash-stats

When used in a hash table the instruction cache will usually beat the CPU and throughput measured here. In my tests the smallest FNV1A beats the fastest crc32_hw1 with Perl 5 hash tables. Even if those worse hash functions will lead to more collisions, the overall speed advantage and inline-ability beats the slightly worse quality. See e.g. A Seven-Dimensional Analysis of Hashing Methods and its Implications on Query Processing for a concise overview of the best hash table strategies, confirming that the simplest Mult hashing (bernstein, FNV*, x17, sdbm) always beat "better" hash functions (Tabulation, Murmur, Farm, ...) when used in a hash table.

The fast hash functions tested here are recommendable as fast for file digests and maybe bigger databases, but not for 32bit hash tables. The "Quality problems" lead to less uniform distribution, i.e. more collisions and worse performance, but are rarely related to real security attacks, just the 2nd sanity zeroes test against \0 invariance is security relevant.

Columns

MiB/sec: The average of the Bulk key speed test for alignments 0-7 with 262144-byte keys. The higher the better.

cycl./hash: The average of the Small key speed test for 1-31 byte keys. The smaller the better.

cycl./map: The result of the Hashmap test for /usr/dict/words with std::unordered_map get queries, with the standard deviation in brackets. This tests the inlinability of the hash function (see size). The smaller the better.

size: The object size in byte on AMD64. This affects the inlinability in e.g. hash tables. The smaller the better.

Quality problems: See the failures in the linked doc. The less the better.

Other

SECURITY

The hash table attacks described in SipHash against City, Murmur or Perl JenkinsOAAT or at Hash Function Lounge are not included here.

Such an attack avoidance cannot be the problem of the hash function, but only the hash table collision resolution scheme. You can attack every single hash function, even the best and most secure if you detect the seed, e.g. from language (mis-)features, side-channel attacks, collision timings and independly the sort-order, so you need to protect your collision handling scheme from the worst-case O(n), i.e. separate chaining with linked lists. Linked lists chaining allows high load factors, but is very cache-unfriendly. The only recommendable linked list scheme is inlining the key or hash into the array. Nowadays everybody uses fast open addressing, even if the load factor needs to be ~50%, unless you use Cuckoo Hashing.

I.e. the usage of SipHash for their hash table in Python 3.4, ruby, rust, systemd, OpenDNS, Haskell and OpenBSD is pure security theatre. SipHash is not secure enough for security purposes and not fast enough for general usage. Brute-force generation of ~32k collisions need 2-4m for all these hashes. siphash being the slowest needs max 4m, other typically max 2m30s, with <10s for practical 16k collision attacks with all hash functions. Using Murmur is usually slower than a simple Mult, even in the worst case. Provable secure is only uniform hashing, i.e. 2-5 independent Mult or Tabulation, or using a guaranteed logarithmic collision scheme (a tree) or a linear collision scheme, such as Robin Hood or Cockoo hashing with collision counting.

One more note regarding security: Nowadays even SHA1 can be solved in a solver, like Z3 (or faster ones) for practical hash table collision attacks (i.e. 14-20 bits). All hash functions with less than 160 bits tested here cannot be considered "secure" at all.

The '' vulnerability attack with binary keys is tested in the 2nd Sanity Zero test.

CRYPTO

The official NIST hash function testsuite does not do such extensive statistical tests, to search for weak ranges in the bits. Also crypto does not change the initial state, which we do here for our random 32bit seed. Crypto mostly cares about unreversable key -> hash functions without changing the initial fixed state and timings/sidechannel attacks.

The NIST "Cryptographic Algorithm Validation Program" (CAVP) involves the testing of the implementations of FIPS-approved and NIST-recommended cryptographic algorithms. During the NIST SHA-3 competition, the testing methodology was borrowed from the "CAVP", as the KATs and MCTs of the SHA-3 Competition Test Suite were based on the CAVP tests for SHA-2. In addition to this, the “Extremely Long Message Test,” not present in the CAVP for SHA-2, required the submitters to generate the hash value corresponding to a message with a length of 1 GiB. “NIST - Cryptographic Algorithm Validation Program (CAVP),” June 2017. Available: http://csrc.nist.gov/groups/STM/cavp (No testing source code provided, just high-level descriptions)

Two other independent third party testsuites found an extensive number of bugs and weaknesses in the SHA3 candidates. "Finding Bugs in Cryptographic Hash Function Implementations", Nicky Mouha, Mohammad S Raunak, D. Richard Kuhn, and Raghu Kacker, 2017. https://eprint.iacr.org/2017/891.pdf

Maybe independent researchers should come together to do a better public SHA-4 round, based on better and more testing methods, open source code for the tests, and using standard industry practices, such as valgrind, address-sanitizer and ubsan to detect obvious bugs.

PROBLEMS

Typical undefined behaviour (UB) problems: