Benchmarks

All numbers are base64 decoding throughput in GB/s (higher is better) measured with BenchmarkDotNet on the enron email corpus (benchmark/data/email). The benchmarks are fully reproducible.

To reproduce on your own machine:

cd benchmark
dotnet run -c Release
# or a single benchmark:
dotnet run -c Release --filter "SimdUnicodeBenchmarks.RealDataBenchmark.AVX2DecodingRealDataUTF8(FileName: \"data/email/\")"

For the UTF-16 benchmarks (off by default):

cd benchmark
dotnet run -c Release --anyCategories UTF16

vs. the accelerated .NET functions

Compared against System.Buffers.Text.Base64.DecodeFromUtf8, the runtime's fast SIMD base64 decoder. SimdBase64 is 1.7×–2.6× faster on realistic inputs of a few kilobytes.

processor and base freq.	SimdBase64 (GB/s)	.NET (GB/s)	speed-up
Apple M2 (ARM, 3.5 GHz)	10	3.8	2.6×
AWS Graviton 3 (ARM, 2.6 GHz)	5.1	2.0	2.6×
Intel Ice Lake (2.0 GHz)	7.6	3.4	2.2×
AMD EPYC 7R32 (Zen 2, 2.8 GHz)	6.9	3.0	2.3×

vs. Convert.FromBase64String

The .NET runtime does not accelerate Convert.FromBase64String. Replacing it with SimdBase64.Base64.FromBase64String multiplies the decoding speed:

processor and base freq.	SimdBase64 (GB/s)	.NET (GB/s)	speed-up
Apple M2 (ARM, 3.5 GHz)	4.0	1.1	3.6×
Intel Ice Lake (2.0 GHz)	2.5	0.65	3.8×

Hardware, runtime version and input all affect these numbers. Treat the tables as representative, and measure on your own target hardware for decisions that matter.