Skip to content

benchmarks.md

Latest commit

 

History

History
235 lines (208 loc) · 22.1 KB

benchmarks.md

File metadata and controls

235 lines (208 loc) · 22.1 KB
Raw

Benchmarks

Back to the main README

These benchmarks use BenchmarkDotNet on .NET 10 with fixed, randomly generated input data. They cover three focused comparisons:

  • Serial one-shot hashing with this repository's fully managed Blake3 package and the Blake3.Native wrapper around the SIMD-optimized official Rust implementation, measured on x64 and ARM64.
  • One-shot hashing with Blake3 and other .NET hash implementations, measured on x64: Blake3.Managed 1.2.1 (shown as Blake3 managed (ext)), Blake2Fast, and System.Security.Cryptography.SHA256.
  • Explicit parallel hashing with Hasher.UpdateWithJoin from Blake3 and Blake3.Native, measured on x64 for inputs from 512 KiB to 10 MB.

Blake3.Managed is an independently maintained package, not a previous or alternate package name for this repository's managed implementation. Its one-shot Hasher.Hash API may use internal parallelism for large inputs, so those large-input results are not a serial-to-serial comparison.

Highlights

  • The fully managed Blake3 package delivers near-native serial performance without shipping or loading a native library. Across the x64 and ARM64 measurements, it ranges from about 11% faster to 22% slower than Blake3.Native.
  • On x64, Blake3 is about 12% to 22% slower than native from 1 KB through 10 MB, while it is 11% faster for the 100-byte input. On ARM64, it is about 13% to 20% slower from 1 KB through 10 MB and 5% faster for the 4-byte input.
  • The x64 serial results show the benefits of BLAKE3 on larger inputs. At 64 KiB, Blake3 is about 2.4x faster than the external Blake3.Managed package, 3.9x faster than SHA256, and 9.9x faster than Blake2Fast. For large inputs, the external Blake3.Managed one-shot API uses its own parallelism and can outperform a serial call.
  • The serial Blake3 and Blake3.Native measurements complete without managed allocations. Explicit parallel hashing adds scheduling and allocation overhead, but substantially improves throughput for multi-megabyte inputs: at 10 MB on x64, UpdateWithJoin is about 4.4x faster than serial Blake3 and 6.5x faster than serial Blake3.Native in these runs.

Benchmark results depend on the CPU, available cores, runtime, and input shape. In particular, parallel crossover points can vary and should be measured with representative application data.

The tables are BenchmarkDotNet ShortRun results (three warmup and three measurement iterations), so small differences should be treated as indicative rather than absolute.

Serial: Blake3 vs Blake3.Native (Hasher.Hash)

AMD Ryzen 9 9950X (x64)

BenchmarkDotNet v0.15.8, Windows 11 (10.0.26200.8655/25H2/2025Update/HudsonValley2)
AMD Ryzen 9 9950X 4.30GHz, 1 CPU, 32 logical and 16 physical cores
.NET SDK 10.0.302
  [Host]   : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v4
  ShortRun : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v4

Job=ShortRun  IterationCount=3  LaunchCount=1 WarmupCount=3
Method N Mean Error StdDev Ratio Code Size Allocated Alloc Ratio
'Blake3 default' 4 82.12 ns 2.671 ns 0.146 ns 1.00 3,218 B - NA
'Blake3 native' 4 75.27 ns 0.196 ns 0.011 ns 0.92 355 B - NA
'Blake3 default' 100 123.12 ns 0.984 ns 0.054 ns 1.00 3,211 B - NA
'Blake3 native' 100 138.03 ns 2.187 ns 0.120 ns 1.12 355 B - NA
'Blake3 default' 1000 1,114.55 ns 3.604 ns 0.198 ns 1.00 3,211 B - NA
'Blake3 native' 1000 992.50 ns 9.308 ns 0.510 ns 0.89 355 B - NA
'Blake3 default' 10000 3,612.77 ns 115.985 ns 6.358 ns 1.00 7,350 B - NA
'Blake3 native' 10000 3,098.95 ns 20.440 ns 1.120 ns 0.86 360 B - NA
'Blake3 default' 65536 6,030.67 ns 17.941 ns 0.983 ns 1.00 8,169 B - NA
'Blake3 native' 65536 4,987.99 ns 86.206 ns 4.725 ns 0.83 360 B - NA
'Blake3 default' 100000 10,905.14 ns 320.218 ns 17.552 ns 1.00 8,494 B - NA
'Blake3 native' 100000 9,225.64 ns 171.406 ns 9.395 ns 0.85 360 B - NA
'Blake3 default' 131072 11,754.00 ns 25.263 ns 1.385 ns 1.00 8,155 B - NA
'Blake3 native' 131072 9,745.71 ns 164.301 ns 9.006 ns 0.83 360 B - NA
'Blake3 default' 262144 23,587.73 ns 852.524 ns 46.730 ns 1.00 8,155 B - NA
'Blake3 native' 262144 19,280.80 ns 237.858 ns 13.038 ns 0.82 360 B - NA
'Blake3 default' 524288 46,399.06 ns 2,031.634 ns 111.361 ns 1.00 8,155 B - NA
'Blake3 native' 524288 38,418.34 ns 183.719 ns 10.070 ns 0.83 360 B - NA
'Blake3 default' 1000000 88,234.66 ns 2,110.065 ns 115.660 ns 1.00 8,485 B - NA
'Blake3 native' 1000000 75,019.60 ns 429.557 ns 23.545 ns 0.85 360 B - NA
'Blake3 default' 10000000 907,587.73 ns 45,363.456 ns 2,486.524 ns 1.00 8,441 B - NA
'Blake3 native' 10000000 756,075.81 ns 41,412.564 ns 2,269.962 ns 0.83 360 B - NA

Apple M4 Pro (ARM64)

BenchmarkDotNet v0.15.8, macOS Tahoe 26.5.1 (25F80) [Darwin 25.5.0]
Apple M4 Pro, 1 CPU, 14 logical and 14 physical cores
.NET SDK 10.0.201
  [Host]   : .NET 10.0.5 (10.0.5, 10.0.526.15411), Arm64 RyuJIT armv8.0-a
  ShortRun : .NET 10.0.5 (10.0.5, 10.0.526.15411), Arm64 RyuJIT armv8.0-a

Job=ShortRun  IterationCount=3  LaunchCount=1 WarmupCount=3
Method N Mean Error StdDev Ratio Allocated Alloc Ratio
'Blake3 default' 4 49.87 ns 1.231 ns 0.067 ns 1.00 - NA
'Blake3 native' 4 52.63 ns 1.330 ns 0.073 ns 1.06 - NA
'Blake3 default' 100 102.62 ns 3.250 ns 0.178 ns 1.00 - NA
'Blake3 native' 100 91.25 ns 2.549 ns 0.140 ns 0.89 - NA
'Blake3 default' 1000 836.55 ns 20.618 ns 1.130 ns 1.00 - NA
'Blake3 native' 1000 700.29 ns 22.154 ns 1.214 ns 0.84 - NA
'Blake3 default' 10000 5,218.23 ns 110.609 ns 6.063 ns 1.00 - NA
'Blake3 native' 10000 4,514.60 ns 38.689 ns 2.121 ns 0.87 - NA
'Blake3 default' 65536 28,487.52 ns 821.890 ns 45.051 ns 1.00 - NA
'Blake3 native' 65536 25,077.70 ns 136.511 ns 7.483 ns 0.88 - NA
'Blake3 default' 100000 44,171.43 ns 1,142.003 ns 62.597 ns 1.00 - NA
'Blake3 native' 100000 38,953.93 ns 189.971 ns 10.413 ns 0.88 - NA
'Blake3 default' 131072 57,069.08 ns 2,428.221 ns 133.099 ns 1.00 - NA
'Blake3 native' 131072 50,187.54 ns 441.632 ns 24.207 ns 0.88 - NA
'Blake3 default' 262144 114,543.10 ns 7,969.881 ns 436.856 ns 1.00 - NA
'Blake3 native' 262144 100,528.36 ns 1,200.951 ns 65.828 ns 0.88 - NA
'Blake3 default' 524288 228,890.88 ns 6,392.852 ns 350.414 ns 1.00 - NA
'Blake3 native' 524288 201,074.93 ns 1,575.382 ns 86.352 ns 0.88 - NA
'Blake3 default' 1000000 440,723.65 ns 9,942.807 ns 544.999 ns 1.00 - NA
'Blake3 native' 1000000 384,006.50 ns 12,111.836 ns 663.891 ns 0.87 - NA
'Blake3 default' 10000000 4,428,962.46 ns 300,739.435 ns 16,484.543 ns 1.00 - NA
'Blake3 native' 10000000 3,846,564.83 ns 25,030.885 ns 1,372.027 ns 0.87 - NA

One-shot: other hash implementations (AMD Ryzen 9 9950X, x64)

BenchmarkDotNet v0.15.8, Windows 11 (10.0.26200.8655/25H2/2025Update/HudsonValley2)
AMD Ryzen 9 9950X 4.30GHz, 1 CPU, 32 logical and 16 physical cores
.NET SDK 10.0.302
  [Host]   : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v4
  ShortRun : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v4

Job=ShortRun  IterationCount=3  LaunchCount=1 WarmupCount=3
Method N Mean Error StdDev Ratio RatioSD Gen0 Code Size Allocated Alloc Ratio
'Blake3 default' 4 82.42 ns 11.776 ns 0.645 ns 1.00 0.01 - 3,218 B - NA
'Blake3 managed (ext)' 4 86.94 ns 0.212 ns 0.012 ns 1.05 0.01 - 3,745 B - NA
Blake2Fast 4 133.42 ns 0.486 ns 0.027 ns 1.62 0.01 0.0052 5,453 B 88 B NA
SHA256 4 103.44 ns 4.587 ns 0.251 ns 1.25 0.01 - 342 B - NA
'Blake3 default' 100 123.47 ns 1.800 ns 0.099 ns 1.00 0.00 - 3,211 B - NA
'Blake3 managed (ext)' 100 150.45 ns 1.767 ns 0.097 ns 1.22 0.00 - 3,992 B - NA
Blake2Fast 100 134.52 ns 10.890 ns 0.597 ns 1.09 0.00 0.0052 5,450 B 88 B NA
SHA256 100 126.18 ns 5.356 ns 0.294 ns 1.02 0.00 - 342 B - NA
'Blake3 default' 1000 1,115.35 ns 3.225 ns 0.177 ns 1.00 0.00 - 3,211 B - NA
'Blake3 managed (ext)' 1000 1,053.13 ns 15.556 ns 0.853 ns 0.94 0.00 - 3,992 B - NA
Blake2Fast 1000 949.12 ns 4.116 ns 0.226 ns 0.85 0.00 0.0048 5,546 B 88 B NA
SHA256 1000 444.94 ns 8.808 ns 0.483 ns 0.40 0.00 - 342 B - NA
'Blake3 default' 10000 3,606.93 ns 20.725 ns 1.136 ns 1.00 0.00 - 7,354 B - NA
'Blake3 managed (ext)' 10000 3,834.32 ns 66.911 ns 3.668 ns 1.06 0.00 - 18,213 B 56 B NA
Blake2Fast 10000 9,242.27 ns 247.965 ns 13.592 ns 2.56 0.00 - 5,549 B 88 B NA
SHA256 10000 3,673.16 ns 141.781 ns 7.771 ns 1.02 0.00 - 342 B - NA
'Blake3 default' 65536 6,025.40 ns 267.785 ns 14.678 ns 1.00 0.00 - 8,169 B - NA
'Blake3 managed (ext)' 65536 14,552.86 ns 262.955 ns 14.413 ns 2.42 0.01 - 18,114 B 56 B NA
Blake2Fast 65536 59,733.70 ns 80.460 ns 4.410 ns 9.91 0.02 - 5,827 B 88 B NA
SHA256 65536 23,581.89 ns 748.634 ns 41.035 ns 3.91 0.01 - 342 B - NA
'Blake3 default' 100000 10,907.97 ns 582.644 ns 31.937 ns 1.00 0.00 - 8,456 B - NA
'Blake3 managed (ext)' 100000 14,958.62 ns 235.925 ns 12.932 ns 1.37 0.00 0.1221 30,276 B 2231 B NA
Blake2Fast 100000 91,314.10 ns 3,061.176 ns 167.793 ns 8.37 0.03 - 5,549 B 88 B NA
SHA256 100000 35,867.08 ns 1,181.185 ns 64.745 ns 3.29 0.01 - 342 B - NA
'Blake3 default' 131072 11,538.65 ns 34.346 ns 1.883 ns 1.00 0.00 - 8,155 B - NA
'Blake3 managed (ext)' 131072 17,473.52 ns 625.486 ns 34.285 ns 1.51 0.00 0.1221 30,044 B 2508 B NA
Blake2Fast 131072 119,413.87 ns 413.862 ns 22.685 ns 10.35 0.00 - 5,827 B 88 B NA
SHA256 131072 46,945.24 ns 559.570 ns 30.672 ns 4.07 0.00 - 342 B - NA
'Blake3 default' 262144 23,386.06 ns 544.747 ns 29.859 ns 1.00 0.00 - 8,155 B - NA
'Blake3 managed (ext)' 262144 18,825.93 ns 745.672 ns 40.873 ns 0.81 0.00 0.1526 30,020 B 2858 B NA
Blake2Fast 262144 239,922.34 ns 2,700.268 ns 148.011 ns 10.26 0.01 - 5,826 B 88 B NA
SHA256 262144 93,802.67 ns 257.119 ns 14.094 ns 4.01 0.00 - 342 B - NA
'Blake3 default' 524288 46,882.88 ns 3,506.974 ns 192.229 ns 1.00 0.01 - 8,155 B - NA
'Blake3 managed (ext)' 524288 27,097.21 ns 941.849 ns 51.626 ns 0.58 0.00 0.2136 30,033 B 3699 B NA
Blake2Fast 524288 480,848.50 ns 48,825.934 ns 2,676.314 ns 10.26 0.06 - 5,823 B 88 B NA
SHA256 524288 190,418.43 ns 44,029.824 ns 2,413.423 ns 4.06 0.05 - 342 B - NA
'Blake3 default' 1000000 91,670.84 ns 31,592.943 ns 1,731.716 ns 1.00 0.02 - 8,485 B - NA
'Blake3 managed (ext)' 1000000 35,480.01 ns 64,008.083 ns 3,508.499 ns 0.39 0.03 0.2441 29,502 B 4253 B NA
Blake2Fast 1000000 913,685.16 ns 30,384.972 ns 1,665.503 ns 9.97 0.16 - 5,539 B 88 B NA
SHA256 1000000 358,664.79 ns 10,127.695 ns 555.133 ns 3.91 0.06 - 342 B - NA
'Blake3 default' 10000000 925,091.31 ns 71,865.863 ns 3,939.210 ns 1.00 0.01 - 8,441 B - NA
'Blake3 managed (ext)' 10000000 221,220.17 ns 6,563.852 ns 359.787 ns 0.24 0.00 0.2441 37,399 B 7923 B NA
Blake2Fast 10000000 9,159,290.62 ns 32,710.881 ns 1,792.994 ns 9.90 0.04 - 5,823 B 88 B NA
SHA256 10000000 3,576,552.21 ns 10,767.420 ns 590.199 ns 3.87 0.01 - 342 B - NA

Explicit parallel hashing (AMD Ryzen 9 9950X, x64)

BenchmarkDotNet v0.15.8, Windows 11 (10.0.26200.8655/25H2/2025Update/HudsonValley2)
AMD Ryzen 9 9950X 4.30GHz, 1 CPU, 32 logical and 16 physical cores
.NET SDK 10.0.302
  [Host]   : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v4
  ShortRun : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v4

Job=ShortRun  IterationCount=3  LaunchCount=1 WarmupCount=3
Method N Mean Error StdDev Ratio RatioSD Gen0 Allocated Alloc Ratio
'Blake3 parallel' 524288 50.56 us 14.678 us 0.805 us 1.00 0.02 0.2441 4197 B 1.00
'Blake3 native parallel' 524288 17.23 us 0.411 us 0.023 us 0.34 0.00 - - 0.00
'Blake3 parallel' 1000000 26.95 us 3.921 us 0.215 us 1.00 0.01 0.4272 7072 B 1.00
'Blake3 native parallel' 1000000 184.98 us 46.628 us 2.556 us 6.86 0.09 - - 0.00
'Blake3 parallel' 10000000 207.25 us 30.229 us 1.657 us 1.00 0.01 1.4648 25776 B 1.00
'Blake3 native parallel' 10000000 116.22 us 14.228 us 0.780 us 0.56 0.01 - - 0.00