NumPy-shaped arrays for .NET

NumSharp

A .NET port of NumPy focused on API parity, unmanaged NDArray storage, view semantics, broadcasting, and runtime-generated SIMD kernels.

Start with NDArray View benchmarks API reference

Install and run

Start from familiar NumPy-style calls in ordinary C#.

dotnet add package NumSharp

using NumSharp;
var a = np.array(new[] { 1, 2, 3, 4, 5 });
var b = np.arange(5);
Console.WriteLine(a + b);  // [1 3 5 7 9]

NumPy 2.x Compatibility target

15 Core dtypes

V128-V512 Runtime SIMD width

Views Slicing shares memory

Documentation Map

Pick the part of the stack you are working on.

Arrays NDArray fundamentals

Shape, strides, storage, indexing, and the mental model for NumSharp arrays.

Semantics Broadcasting

How NumSharp stretches operands without materializing repeated data.

Kernels NDIter

The iterator layer that schedules strided, buffered, and fused inner loops.

Runtime compiler IL Generation

DynamicMethod kernels, SIMD paths, cache families, and generator ownership.

Parity NumPy compliance

Compatibility notes, behavioral scope, and areas still being tightened.

Performance Benchmarks

Current NumSharp-vs-NumPy results with drill-down reports by subsystem.

What NumSharp Optimizes For

Behavior first, with fast paths where the layout and dtype make them possible.

NumPy-shaped API

Creation, indexing, broadcasting, math, reductions, random sampling, and file I/O use familiar NumPy names and behavior.

Unmanaged storage

Arrays use raw storage behind shape and stride metadata, so views and kernels can operate close to the metal.

Runtime specialization

Elementwise, cast, reduction, scan, and selection paths emit dtype-specific IL and cache the compiled delegates.

Release-tracked performance

The benchmark dashboard publishes stable history snapshots, not one-off scratch output.

GitHub Repository NuGet Package