Faiss.NET

C# bindings for Faiss.

• Faiss v1.14.1
• .NET 9.0

Warning

This library is under active construction and currently not usable yet.

Implemented Indexes

• IndexFlatIP
• IndexFlatL2
• IndexHNSW

Also, a generic Index factory is implemented which can be used to instantiate all other Faiss supported indexes.

This library aims to be as "bare metal" as possible while being straightforward to work with.

Additionally, Index serialization/deserialization and GPU Indexes + sharding is supported.

Supported Platforms:

Platform	x64	arm64
Windows	✅	✅
Linux	✅	✅
MacOS	✅	✅

GPU Acceleration:

Platform	x64	arm64
CUDA (Linux)	✅	✅
ROCm (Linux)	✅	❌

Installation

Prerequisites

Windows

You need a C++ redistributable installed. E.g.

$ winget install --id Microsoft.VCRedist.2015+.x64 --silent

Linux

OpenBLAS, OpenMP and Fortran runtimes must be installed. E.g.

$ sudo apt-get install -y libopenblas0 libgomp1 libgfortran5

MacOS

An OpenMP runtime is required. E.g.

$ brew install libomp

NuGet

Get the base NuGet:

$ dotnet add package Faiss.NET

Then pick a native package:

Either cpu only or with additional gpu support.

CPU only

$ dotnet add package Faiss.NET.Native

CPU+GPU

$ dotnet add package Faiss.NET.Native.Gpu.Cuda # Linux only
$ dotnet add package Faiss.NET.Native.Windows
$ dotnet add package Faiss.NET.Native.MacOS

or

$ dotnet add package Faiss.NET.Native.Gpu.Rocm # Linux x64 only
$ dotnet add package Faiss.NET.Native.Windows
$ dotnet add package Faiss.NET.Native.MacOS

Usage

Basic

using Faiss.NET;

int dimensions = 4;
using var index = new FaissIndex<FaissIndexFlatL2>(new FaissIndexFlatL2(dimensions));

float[] vectors = { 1.0f, 2.0f, 3.0f, 4.0f };

index.Add(vectors);

using var searchResult = index.Search(vectors, 1);

float distance = searchResult.Distances[0]; // 0.0f
long label = searchResult.Labels[0]; // 0

Examples

Low-Level index usage

using Faiss.NET;

int dimensions = 2;
using var index = new FaissIndexFlatIP(dimensions);

float[] vectors = { 2.0f, 3.0f };
index.Add(1, vectors);

float[] distances = new float[1];
long[] labels = new long[1];

index.Search(1, vectors, 1, distances, labels);

long foundLabel = labels[0]; // 0
float foundDistance = distances[0]; // 13

GPU Acceleration

Supported CUDA GPUs:

Compute Capability	Architecture	GPUs
75	Turing	RTX 2080/2080 Ti/2070/2060, Tesla T4
80	Ampere	A100
86	Ampere	RTX 3080/3090/3070/3060, A40, A10, A16, A30
89	Ada Lovelace	RTX 4090/4080/4070/4060, L40, L40S, L4
90	Hopper	H100, H200
120	Blackwell	B100, B200, GB200

Supported ROCm GPUs:

GFX Architecture	GPUs
gfx90a	AMD Instinct MI210, MI250, MI250X
gfx942	AMD Instinct MI300A
gfx950	AMD Instinct MI355X
gfx1030	Radeon PRO V620, RX 6800/6800 XT/6900 XT
gfx1031	Radeon RX 6700/6700 XT/6750 XT
gfx1032	Radeon RX 6600/6600 XT/6600M
gfx1100	Radeon RX 7600/7600 XT
gfx1101	Radeon RX 7700 XT/7800 XT
gfx1102	Radeon RX 7900 GRE/XT/Xt
gfx1200	Radeon RX 9070
gfx1201	Radeon RX 9070 XT/9060