Hashing & ids benchmarks

BENCHMARKS · HASHING & IDS

Cryptographic vs non-cryptographic hashing; Guid v4 vs v7. Suite and raw reports: github.com/gdhami-net/dotnet-benchmarks.

Intel Core Ultra 9 285HX · Windows 11 · BenchmarkDotNet 0.15.8 · SDK 10.0.201 · 2026-08-22 · single run — not yet median-of-N · bars/tables: mean time, lower is better

HashPayload

Hash the same payload: cryptographic SHA-256 vs the non-cryptographic XxHash family — pick by threat model, know the cost.

TAKEAWAY Pick by threat model, not speed: content addressing and checksums can take XxHash3's 10x win; anything security-relevant stays SHA-256 - which is already hardware-fast.
In this run: XxHash3_hash is fastest — 13.7× faster than the baseline.

RESULTS — 1024

method	net9.0	net10.0	ratio	allocated	Δ net10.0
Sha256 baseline	325 ns ±6 ns	320 ns ±4 ns	1.00×	56 B	-1%
XxHash64_hash	55 ns ±2 ns	54 ns ±0 ns	0.17×	32 B	-2%
XxHash3_hash	23 ns ±2 ns	23 ns ±0 ns	0.07×	32 B	-0%

RATIO VS BASELINE — 1024 · net10.0

Sha256

1.00×

XxHash64_hash

0.17×

XxHash3_hash

0.07×

RESULTS — 1048576

method	net9.0	net10.0	ratio	allocated	Δ net10.0
Sha256 baseline	213.2 µs ±2.5 µs	209.6 µs ±839 ns	1.00×	56 B	-2%
XxHash64_hash	45.9 µs ±45 ns	47.1 µs ±717 ns	0.22×	32 B	+3%
XxHash3_hash	18.6 µs ±37 ns	17.5 µs ±133 ns	0.08×	32 B	-6%

RATIO VS BASELINE — 1048576 · net10.0

Sha256

1.00×

XxHash64_hash

0.22×

XxHash3_hash

0.08×

SCALING · net10.0 · MEAN TIME BY INPUT SIZE (LOG)

Sha256XxHash64_hashXxHash3_hash

THE LEDGER · SAME WORKLOAD ACROSS RELEASES (LOG)

Sha256XxHash64_hashXxHash3_hash

THE CODE BEING MEASURED

Sha256 — what this measures

SHA-256 — cryptographic, hardware-accelerated on modern CPUs.

[Benchmark (Baseline)]
public byte[] Sha256()
=> SHA256.HashData(_payload);

XxHash64_hash — what this measures

XxHash64 — fast non-cryptographic checksum (System.IO.Hashing).

[Benchmark]
public byte[] XxHash64_hash()
=> XxHash64.Hash(_payload);

XxHash3_hash — what this measures

XxHash3 — the newer, usually faster member of the family.

[Benchmark]
public byte[] XxHash3_hash()
=> XxHash3.Hash(_payload);

MakeIds

Make 1,000 ids: random Guid v4 vs .NET 9's time-ordered v7 — v7 exists so your database index stops fragmenting.

TAKEAWAY Guid.CreateVersion7 costs a whisker more to create and saves you from fragmented clustered indexes forever. For database keys, v7 is the new default.
In this run: Guid_v4 is fastest (it is the baseline).

RESULTS

method	net9.0	net10.0	ratio	allocated	Δ net10.0
Guid_v4 baseline	35.3 µs ±534 ns	34.7 µs ±214 ns	1.00×	—	-2%
Guid_v7	60.2 µs ±752 ns	58.4 µs ±334 ns	1.68×	—	-3%

RATIO VS BASELINE · net10.0

Guid_v4

1.00×

Guid_v7

1.68×

THE LEDGER · SAME WORKLOAD ACROSS RELEASES (LOG)

Guid_v4Guid_v7

THE CODE BEING MEASURED

Guid_v4 — what this measures

Guid.NewGuid — random v4.

[Benchmark (Baseline)]
public Guid Guid_v4()
{
    var g = Guid.Empty;
    for (var i = 0; i < 1_000; i++) g = Guid.NewGuid();
    return g;
}

Guid_v7 — what this measures

Guid.CreateVersion7 (.NET 9+) — time-ordered, index-friendly.

[Benchmark]
public Guid Guid_v7()
{
    var g = Guid.Empty;
    for (var i = 0; i < 1_000; i++) g = Guid.CreateVersion7();
    return g;
}

speed vs allocation

Fast is one axis. What it costs the GC is the other.

Sha256XxHash64_hashXxHash3_hashGuid_v4Guid_v7